megadetector 5.0.23__py3-none-any.whl → 5.0.25__py3-none-any.whl

megadetector/utils/wi_utils.py

@@ -0,0 +1,2691 @@
+"""
+
+wi_utils.py
+
+Functions related to working with the WI insights platform, specifically for:
+    
+* Retrieving images based on .csv downloads
+* Pushing results to the ProcessCVResponse() API (requires an API key)
+* Working with WI taxonomy records and geofencing data
+
+"""
+
+#%% Imports and constants
+
+import os
+import requests
+import json
+
+import numpy as np
+import pandas as pd
+
+from copy import deepcopy
+from collections import defaultdict
+from multiprocessing.pool import Pool, ThreadPool
+from functools import partial
+from tqdm import tqdm
+
+from megadetector.utils.path_utils import insert_before_extension
+from megadetector.utils.ct_utils import split_list_into_n_chunks
+from megadetector.utils.ct_utils import round_floats_in_nested_dict
+from megadetector.utils.ct_utils import is_list_sorted
+from megadetector.utils.ct_utils import invert_dictionary
+from megadetector.utils.ct_utils import sort_list_of_dicts_by_key
+from megadetector.utils.ct_utils import sort_dictionary_by_value
+from megadetector.utils.ct_utils import sort_dictionary_by_key
+from megadetector.utils.path_utils import find_images
+from megadetector.postprocessing.validate_batch_results import \
+    validate_batch_results, ValidateBatchResultsOptions
+
+md_category_id_to_name = {'1':'animal','2':'person','3':'vehicle'}
+md_category_name_to_id = invert_dictionary(md_category_id_to_name)
+
+# Only used when pushing results directly to the platform via the API; any detections we want 
+# to show in the UI should have at least this confidence value.
+min_md_output_confidence = 0.25
+
+# Fields expected to be present in a valid WI result
+wi_result_fields = ['wi_taxon_id','class','order','family','genus','species','common_name']
+
+
+#%% Miscellaneous WI support functions
+
+def is_valid_prediction_string(s):
+    """
+    Determine whether [s] is a valid WI prediction string.  Prediction strings look like:
+    
+    '90d950db-2106-4bd9-a4c1-777604c3eada;mammalia;rodentia;;;;rodent'
+    
+    Args:
+        s (str): the string to be tested for validity
+        
+    Returns:
+        bool: True if this looks more or less like a WI prediction string
+    """
+
+    # Note to self... don't get tempted to remove spaces here; spaces are used
+    # to indicate subspecies.
+    return isinstance(s,str) and (len(s.split(';')) == 7) and (s == s.lower())
+
+
+def is_valid_taxonomy_string(s):
+    """
+    Determine whether [s] is a valid 5-token WI taxonomy string.  Taxonmy strings look like:
+    
+    'mammalia;rodentia;;;;rodent'
+    'mammalia;chordata;canidae;canis;lupus dingo'
+    
+    Args:
+        s (str): the string to be tested for validity
+        
+    Returns:
+        bool: True if this looks more or less like a WI taxonomy string
+    """
+    return isinstance(s,str) and (len(s.split(';')) == 5) and (s == s.lower())
+
+    
+def wi_result_to_prediction_string(r):
+    """
+    Convert the dict [r] - typically loaded from a row in a downloaded .csv file - to
+    a valid prediction string, e.g.:
+        
+    1f689929-883d-4dae-958c-3d57ab5b6c16;;;;;;animal
+    90d950db-2106-4bd9-a4c1-777604c3eada;mammalia;rodentia;;;;rodent
+    
+    Args:
+        r (dict): dict containing WI prediction information, with at least the fields
+            specified in wi_result_fields.
+    
+    Returns:
+        str: the result in [r], as a semicolon-delimited prediction string
+    """
+    
+    values = []
+    for field in wi_result_fields:
+        if isinstance(r[field],str):
+            values.append(r[field].lower())
+        else:
+            assert isinstance(r[field],float) and np.isnan(r[field])
+            values.append('')
+    s = ';'.join(values)
+    assert is_valid_prediction_string(s)
+    return s
+        
+
+def compare_values(v0,v1):
+    """
+    Utility function for comparing two values when we want to return True if both
+    values are NaN.
+    
+    Args:
+        v0 (object): the first value to compare
+        v1 (object): the second value to compare
+        
+    Returns:
+        bool: True if v0 == v1, or if both v0 and v1 are NaN
+    """
+    
+    if isinstance(v0,float) and isinstance(v1,float) and np.isnan(v0) and np.isnan(v1):
+        return True
+    return v0 == v1
+
+
+def record_is_unidentified(record):
+    """
+    A record is considered "unidentified" if the "identified by" field is either NaN or "computer vision"
+    
+    Args:
+        record (dict): dict representing a WI result loaded from a .csv file, with at least the 
+            field "identified_by"
+    
+    Returns:
+        bool: True if the "identified_by" field is either NaN or a string indicating that this 
+        record has not yet been human-reviewed.
+    """    
+    
+    identified_by = record['identified_by']
+    assert isinstance(identified_by,float) or isinstance(identified_by,str)
+    if isinstance(identified_by,float):
+        assert np.isnan(identified_by)
+        return True
+    else:
+        return identified_by == 'Computer vision'
+
+
+def record_lists_are_identical(records_0,records_1,verbose=False):
+    """
+    Takes two lists of records in the form returned by read_images_from_download_bundle and
+    determines whether they are the same.
+    
+    Args:
+        records_0 (list of dict): the first list of records to compare
+        records_1 (list of dict): the second list of records to compare
+        verbose (bool, optional): enable additional debug output
+        
+    Returns:
+        bool: True if the two lists are identical
+    """
+    
+    if len(records_0) != len(records_1):
+        return False    
+    
+    # i_record = 0; record_0 = records_0[i_record]
+    for i_record,record_0 in enumerate(records_0):
+        record_1 = records_1[i_record]
+        assert set(record_0.keys()) == set(record_1.keys())
+        for k in record_0.keys():
+            if not compare_values(record_0[k],record_1[k]):
+                if verbose:
+                    print('Image ID: {} ({})\nRecord 0/{}: {}\nRecord 1/{}: {}'.format(
+                        record_0['image_id'],record_1['image_id'],
+                        k,record_0[k],k,record_1[k]))
+                return False
+            
+    return True
+    
+    
+#%% Functions for managing WI downloads
+
+def read_sequences_from_download_bundle(download_folder):
+    """
+    Reads sequences.csv from [download_folder], returning a list of dicts.  This is a
+    thin wrapper around pd.read_csv, it's just here for future-proofing.
+    
+    Args:
+        download_folder (str): a folder containing exactly one file called sequences.csv, typically
+            representing a Wildlife Insights download bundle.
+    
+    Returns:
+        list of dict: a direct conversion of the .csv file to a list of dicts
+    """
+
+    print('Reading sequences from {}'.format(download_folder))
+    
+    sequence_list_files = os.listdir(download_folder)
+    sequence_list_files = \
+        [fn for fn in sequence_list_files if fn == 'sequences.csv']
+    assert len(sequence_list_files) == 1, \
+        'Could not find sequences.csv in {}'.format(download_folder)
+    
+    sequence_list_file = os.path.join(download_folder,sequence_list_files[0])
+
+    df = pd.read_csv(sequence_list_file)
+    sequence_records = df.to_dict('records')
+    return sequence_records
+    
+    
+def read_images_from_download_bundle(download_folder):
+    """
+    Reads all images.csv files from [download_folder], returns a dict mapping image IDs
+    to a list of dicts that describe each image.  It's a list of dicts rather than a single dict
+    because images may appear more than once.
+    
+    Args:
+        download_folder (str): a folder containing one or more images.csv files, typically
+            representing a Wildlife Insights download bundle.
+    
+    Returns:
+        dict: Maps image GUIDs to dicts with at least the following fields:
+            * project_id (int)
+            * deployment_id (str)
+            * image_id (str, should match the key)
+            * filename (str, the filename without path at the time of upload)
+            * location (str, starting with gs://)
+            
+        May also contain clasification fields: wi_taxon_id (str), species, etc.        
+    """
+    
+    print('Reading images from {}'.format(download_folder))
+    
+    ##%% Find lists of images
+    
+    image_list_files = os.listdir(download_folder)
+    image_list_files = \
+        [fn for fn in image_list_files if fn.startswith('images_') and fn.endswith('.csv')]
+    image_list_files = \
+        [os.path.join(download_folder,fn) for fn in image_list_files]
+    print('Found {} image list files'.format(len(image_list_files)))
+
+
+    ##%% Read lists of images by deployment
+
+    image_id_to_image_records = defaultdict(list)
+    
+    # image_list_file = image_list_files[0]
+    for image_list_file in image_list_files:
+    
+        print('Reading images from list file {}'.format(
+            os.path.basename(image_list_file)))
+        
+        df = pd.read_csv(image_list_file)
+        
+        # i_row = 0; row = df.iloc[i_row]
+        for i_row,row in tqdm(df.iterrows(),total=len(df)):
+            
+            row_dict = row.to_dict()
+            image_id = row_dict['image_id']            
+            image_id_to_image_records[image_id].append(row_dict)
+            
+        # ...for each image
+        
+    # ...for each list file    
+
+    deployment_ids = set()
+    for image_id in image_id_to_image_records:
+        image_records = image_id_to_image_records[image_id]
+        for image_record in image_records:
+            deployment_ids.add(image_record['deployment_id'])
+        
+    print('Found {} rows in {} deployments'.format(
+        len(image_id_to_image_records),
+        len(deployment_ids)))
+
+    return image_id_to_image_records
+
+
+def find_images_in_identify_tab(download_folder_with_identify,download_folder_excluding_identify):
+    """
+    Based on extracted download packages with and without the "exclude images in 'identify' tab 
+    checkbox" checked, figure out which images are in the identify tab.  Returns a list of dicts (one
+    per image).
+    
+    Args:
+        download_folder_with_identify (str): the folder containing the download bundle that
+            includes images from the "identify" tab
+        download_folder_excluding_identify (str): the folder containing the download bundle that
+            excludes images from the "identify" tab
+        
+    Returns:
+        list of dict: list of image records that are present in the identify tab
+    """
+    
+    ##%% Read data (~30 seconds)
+    
+    image_id_to_image_records_with_identify = \
+        read_images_from_download_bundle(download_folder_with_identify)
+    image_id_to_image_records_excluding_identify = \
+        read_images_from_download_bundle(download_folder_excluding_identify)
+    
+    
+    ##%% Find images that have not been identified
+    
+    all_image_ids_with_identify = set(image_id_to_image_records_with_identify.keys())
+    all_image_ids_excluding_identify = set(image_id_to_image_records_excluding_identify.keys())
+    
+    image_ids_in_identify_tab = all_image_ids_with_identify.difference(all_image_ids_excluding_identify)
+    
+    assert len(image_ids_in_identify_tab) == \
+        len(all_image_ids_with_identify) - len(all_image_ids_excluding_identify)
+        
+    print('Found {} images with identify, {} in identify tab, {} excluding'.format(
+        len(all_image_ids_with_identify), 
+        len(image_ids_in_identify_tab),
+        len(all_image_ids_excluding_identify)))
+    
+    image_records_in_identify_tab = []
+    deployment_ids_for_downloaded_images = set()    
+    
+    for image_id in image_ids_in_identify_tab:
+        image_records_this_image = image_id_to_image_records_with_identify[image_id]
+        assert len(image_records_this_image) > 0
+        image_records_in_identify_tab.extend(image_records_this_image)
+        for image_record in image_records_this_image:
+            deployment_ids_for_downloaded_images.add(image_record['deployment_id'])
+        
+    print('Found {} records for {} unique images in {} deployments'.format(
+        len(image_records_in_identify_tab),
+        len(image_ids_in_identify_tab),
+        len(deployment_ids_for_downloaded_images)))
+
+    return image_records_in_identify_tab
+
+# ...def find_images_in_identify_tab(...)
+
+
+def write_download_commands(image_records_to_download,
+                            download_dir_base,
+                            force_download=False,
+                            n_download_workers=25,
+                            download_command_file_base=None):
+    """
+    Given a list of dicts with at least the field 'location' (a gs:// URL), prepare a set of "gcloud 
+    storage" commands to download images, and write those to a series of .sh scripts, along with one
+    .sh script that runs all the others and blocks.
+    
+    gcloud commands will use relative paths.
+    
+    image_records_to_download can also be a dict mapping IDs to lists of records.
+    
+    Args:
+        image_records_to_download (list of dict): list of dicts with at least the field 'location'
+        download_dir_base (str): local destination folder
+        force_download (bool, optional): include gs commands even if the target file exists
+        n_download_workers (int, optional): number of scripts to write (that's our hacky way
+            of controlling parallelization)
+        download_command_file (str, optional): path of the .sh script we should write, defaults
+            to "download_wi_images.sh" in the destination folder
+    """
+
+    if isinstance(image_records_to_download,dict):
+        
+        all_image_records = []
+        for k in image_records_to_download:
+            records_this_image = image_records_to_download[k]
+            all_image_records.extend(records_this_image)
+        return write_download_commands(all_image_records, 
+                                       download_dir_base=download_dir_base,
+                                       force_download=force_download,
+                                       n_download_workers=n_download_workers,
+                                       download_command_file_base=download_command_file_base)
+    
+    ##%% Make list of gcloud storage commands
+    
+    if download_command_file_base is None:
+        download_command_file_base = os.path.join(download_dir_base,'download_wi_images.sh')
+    
+    commands = []
+    skipped_urls = []
+    downloaded_urls = set()
+    
+    # image_record = image_records_to_download[0]
+    for image_record in tqdm(image_records_to_download):
+        
+        url = image_record['location']
+        if url in downloaded_urls:
+            continue
+        
+        assert url.startswith('gs://')
+        
+        relative_path = url.replace('gs://','')
+        abs_path = os.path.join(download_dir_base,relative_path)
+        
+        # Skip files that already exist
+        if (not force_download) and (os.path.isfile(abs_path)):
+            skipped_urls.append(url)
+            continue
+        
+        # command = 'gsutil cp "{}" "./{}"'.format(url,relative_path)
+        command = 'gcloud storage cp --no-clobber "{}" "./{}"'.format(url,relative_path)
+        commands.append(command)
+    
+    print('Generated {} commands for {} image records'.format(
+        len(commands),len(image_records_to_download)))
+    
+    print('Skipped {} URLs'.format(len(skipped_urls)))
+    
+    
+    ##%% Write those commands out to n .sh files
+    
+    commands_by_script = split_list_into_n_chunks(commands,n_download_workers)
+    
+    local_download_commands = []
+    
+    output_dir = os.path.dirname(download_command_file_base)
+    os.makedirs(output_dir,exist_ok=True)
+    
+    # Write out the download script for each chunk
+    # i_script = 0
+    for i_script in range(0,n_download_workers):
+        download_command_file = insert_before_extension(download_command_file_base,str(i_script).zfill(2))
+        local_download_commands.append(os.path.basename(download_command_file))
+        with open(download_command_file,'w',newline='\n') as f:
+            for command in commands_by_script[i_script]:
+                f.write(command + '\n')
+    
+    # Write out the main download script
+    with open(download_command_file_base,'w',newline='\n') as f:
+        for local_download_command in local_download_commands:
+            f.write('./' + local_download_command + ' &\n')
+        f.write('wait\n')
+        f.write('echo done\n')
+
+# ...def write_download_commands(...)
+
+
+#%% Functions and constants related to pushing results to the DB
+
+# Sample payload for validation
+sample_update_payload = {
+    
+    "predictions": [
+        {
+          "project_id": "1234",
+          "ignore_data_file_checks": True,
+          "prediction": "f1856211-cfb7-4a5b-9158-c0f72fd09ee6;;;;;;blank",
+          "prediction_score": 0.81218224763870239,
+            "classifications": {
+                "classes": [
+                    "f1856211-cfb7-4a5b-9158-c0f72fd09ee6;;;;;;blank",
+                    "b1352069-a39c-4a84-a949-60044271c0c1;aves;;;;;bird",
+                    "90d950db-2106-4bd9-a4c1-777604c3eada;mammalia;rodentia;;;;rodent",
+                    "f2d233e3-80e3-433d-9687-e29ecc7a467a;mammalia;;;;;mammal",
+                    "ac068717-6079-4aec-a5ab-99e8d14da40b;mammalia;rodentia;sciuridae;dremomys;rufigenis;red-cheeked squirrel"
+                ],
+                "scores": [
+                    0.81218224763870239,
+                    0.1096673980355263,
+                    0.02707692421972752,
+                    0.00771023565903306,
+                    0.0049269795417785636
+                ]
+            },
+            "detections": [
+                {
+                    "category": "1",
+                    "label": "animal",
+                    "conf": 0.181,
+                    "bbox": [
+                        0.02421,
+                        0.35823999999999989,
+                        0.051560000000000009,
+                        0.070826666666666746
+                    ]
+                }
+            ],
+            "model_version": "3.1.2",
+            "prediction_source": "manual_update",
+            "data_file_id": "2ea1d2b2-7f84-43f9-af1f-8be0e69c7015"
+        }
+    ]
+}
+
+blank_prediction_string = 'f1856211-cfb7-4a5b-9158-c0f72fd09ee6;;;;;;blank'
+no_cv_result_prediction_string = 'f2efdae9-efb8-48fb-8a91-eccf79ab4ffb;no cv result;no cv result;no cv result;no cv result;no cv result;no cv result'
+animal_prediction_string = '1f689929-883d-4dae-958c-3d57ab5b6c16;;;;;;animal'
+human_prediction_string = '990ae9dd-7a59-4344-afcb-1b7b21368000;mammalia;primates;hominidae;homo;sapiens;human'
+vehicle_prediction_string = 'e2895ed5-780b-48f6-8a11-9e27cb594511;;;;;;vehicle'
+
+non_taxonomic_prediction_strings = [blank_prediction_string,
+                                    no_cv_result_prediction_string,
+                                    animal_prediction_string,
+                                    vehicle_prediction_string]
+
+process_cv_response_url = 'https://placeholder'
+
+
+def prepare_data_update_auth_headers(auth_token_file):
+    """
+    Read the authorization token from a text file and prepare http headers.
+    
+    Args:        
+        auth_token_file (str): a single-line text file containing a write-enabled
+        API token.
+        
+    Returns:
+        dict: http headers, with fields 'Authorization' and 'Content-Type'
+    """
+
+    with open(auth_token_file,'r') as f:
+        auth_token = f.read()
+            
+    headers = {
+        'Authorization': 'Bearer ' + auth_token,
+        'Content-Type': 'application/json'
+    }
+    
+    return headers
+
+
+def push_results_for_images(payload,
+                            headers,
+                            url=process_cv_response_url,
+                            verbose=False):
+    """
+    Push results for one or more images represented in [payload] to the 
+    process_cv_response API, to write to the WI DB.
+    
+    Args:
+        payload (dict): payload to upload to the API
+        headers (dict): authorization headers, see prepare_data_update_auth_headers
+        url (str, optional): API URL
+        verbose (bool, optional): enable additional debug output            
+        
+    Return:
+        int: response status code
+    """
+    
+    if verbose:
+        print('Sending header {} to URL {}'.format(
+            headers,url))
+        
+    response = requests.post(url, headers=headers, json=payload)
+
+    # Check the response status code
+    if response.status_code in (200,201):  
+        if verbose:
+            print('Successfully pushed results for {} images'.format(len(payload['predictions'])))
+            print(response.headers)
+            print(str(response))
+    else:
+        print(f'Error: {response.status_code} {response.text}')
+        
+    return response.status_code
+
+
+def parallel_push_results_for_images(payloads,
+                                     headers,
+                                     url=process_cv_response_url,
+                                     verbose=False,
+                                     pool_type='thread',
+                                     n_workers=10):
+    """
+    Push results for the list of payloads in [payloads] to the process_cv_response API, 
+    parallelized over multiple workers.
+    
+    Args:
+        payloads (list of dict): payloads to upload to the API
+        headers (dict): authorization headers, see prepare_data_update_auth_headers
+        url (str, optional): API URL
+        verbose (bool, optional): enable additional debug output
+        pool_type (str, optional): 'thread' or 'process'
+        n_workers (int, optional): number of parallel workers
+        
+    Returns:
+        list of int: list of http response codes, one per payload
+    """
+    
+    if n_workers == 1:
+        
+        results = []
+        for payload in payloads:
+            results.append(push_results_for_images(payload,
+                                                   headers=headers,
+                                                   url=url,
+                                                   verbose=verbose))
+        return results
+    
+    else:
+        
+        assert pool_type in ('thread','process')
+        
+        if pool_type == 'thread':
+            pool_string = 'thread'
+            pool = ThreadPool(n_workers)
+        else:
+            pool_string = 'process'
+            pool = Pool(n_workers)
+    
+        print('Created a {} pool of {} workers'.format(
+            pool_string,n_workers))
+        
+        results = list(tqdm(pool.imap(
+            partial(push_results_for_images,headers=headers,url=url,verbose=verbose),payloads), 
+            total=len(payloads)))
+        
+        assert len(results) == len(payloads)
+        return results
+
+
+def generate_payload_with_replacement_detections(wi_result,
+                                                 detections,
+                                                 prediction_score=0.9,
+                                                 model_version='3.1.2',
+                                                 prediction_source='manual_update'):
+    """
+    Generate a payload for a single image that keeps the classifications from 
+    [wi_result], but replaces the detections with the MD-formatted list [detections].
+    
+    Args:
+        wi_result (dict): dict representing a WI prediction result, with at least the 
+            fields in the constant wi_result_fields
+        detections (list): list of WI-formatted detection dicts (with fields ['conf'] and ['category'])
+        prediction_score (float, optional): confidence value to use for the combined prediction
+        model_version (str, optional): model version string to include in the payload
+        prediction_source (str, optional): prediction source string to include in the payload
+        
+    Returns:
+        dict: dictionary suitable for uploading via push_results_for_images
+    """
+    
+    payload_detections = []
+    
+    # detection = detections[0]    
+    for detection in detections:
+        detection_out = detection.copy()
+        detection_out['label'] = md_category_id_to_name[detection['category']]
+        if detection_out['conf'] < min_md_output_confidence:
+            detection_out['conf'] = min_md_output_confidence
+        payload_detections.append(detection_out)
+    
+    prediction_string = wi_result_to_prediction_string(wi_result)
+    
+    prediction = {}
+    prediction['ignore_data_file_checks']  = True
+    prediction['prediction'] = prediction_string
+    prediction['prediction_score'] = prediction_score
+    
+    classifications = {}
+    classifications['classes'] = [prediction_string]
+    classifications['scores'] = [prediction_score]
+    
+    prediction['classifications'] = classifications
+    prediction['detections'] = payload_detections
+    prediction['model_version'] = model_version
+    prediction['prediction_source'] = prediction_source
+    prediction['data_file_id'] = wi_result['image_id']
+    prediction['project_id'] = str(wi_result['project_id'])
+    payload = {}
+    payload['predictions'] = [prediction]
+    
+    return payload
+
+
+def generate_blank_prediction_payload(data_file_id,
+                                      project_id,
+                                      blank_confidence=0.9,
+                                      model_version='3.1.2',
+                                      prediction_source='manual_update'):
+    """
+    Generate a payload that will set a single image to the blank classification, with
+    no detections.  Suitable for upload via push_results_for_images.
+    
+    Args:
+        data_file_id (str): unique identifier for this image used in the WI DB
+        project_id (int): WI project ID
+        blank_confidence (float, optional): confidence value to associate with this
+            prediction
+        model_version (str, optional): model version string to include in the payload
+        prediction_source (str, optional): prediction source string to include in the payload
+        
+    Returns:
+        dict: dictionary suitable for uploading via push_results_for_images
+    """
+    
+    prediction = {}
+    prediction['ignore_data_file_checks']  = True
+    prediction['prediction'] = blank_prediction_string
+    prediction['prediction_score'] = blank_confidence
+    prediction['classifications'] = {}
+    prediction['classifications']['classes'] = [blank_prediction_string]
+    prediction['classifications']['scores'] = [blank_confidence]
+    prediction['detections'] = []
+    prediction['model_version'] = model_version
+    prediction['prediction_source'] = prediction_source
+    prediction['data_file_id'] = data_file_id
+    prediction['project_id'] = project_id
+    payload = {}
+    payload['predictions'] = [prediction]
+    
+    return payload
+
+
+def generate_no_cv_result_payload(data_file_id,
+                                  project_id,
+                                  no_cv_confidence=0.9,
+                                  model_version='3.1.2',
+                                  prediction_source='manual_update'):
+    """
+    Generate a payload that will set a single image to the blank classification, with
+    no detections.  Suitable for uploading via push_results_for_images.
+    
+    Args:
+        data_file_id (str): unique identifier for this image used in the WI DB
+        project_id (int): WI project ID
+        no_cv_confidence (float, optional): confidence value to associate with this
+            prediction        
+        model_version (str, optional): model version string to include in the payload
+        prediction_source (str, optional): prediction source string to include in the payload
+    
+    Returns:
+        dict: dictionary suitable for uploading via push_results_for_images
+    """
+    
+    prediction = {}
+    prediction['ignore_data_file_checks']  = True
+    prediction['prediction'] = no_cv_result_prediction_string
+    prediction['prediction_score'] = no_cv_confidence
+    prediction['classifications'] = {}
+    prediction['classifications']['classes'] = [no_cv_result_prediction_string]
+    prediction['classifications']['scores'] = [no_cv_confidence]
+    prediction['detections'] = []
+    prediction['model_version'] = model_version
+    prediction['prediction_source'] = prediction_source
+    prediction['data_file_id'] = data_file_id
+    prediction['project_id'] = project_id
+    payload = {}
+    payload['predictions'] = [prediction]
+    
+    return payload
+
+
+def generate_payload_for_prediction_string(data_file_id,
+                                           project_id,
+                                           prediction_string,
+                                           prediction_confidence=0.8,
+                                           detections=None,
+                                           model_version='3.1.2',
+                                           prediction_source='manual_update'):
+    """
+    Generate a payload that will set a single image to a particular prediction, optionally
+    including detections.  Suitable for uploading via push_results_for_images.
+    
+    Args:
+        data_file_id (str): unique identifier for this image used in the WI DB
+        project_id (int): WI project ID
+        prediction_string (str): WI-formatted prediction string to include in the payload
+        prediction_confidence (float, optional): confidence value to associate with this
+            prediction        
+        detections (list, optional): list of MD-formatted detection dicts, with fields
+            ['category'] and 'conf'
+        model_version (str, optional): model version string to include in the payload
+        prediction_source (str, optional): prediction source string to include in the payload
+        
+    
+    Returns:
+        dict: dictionary suitable for uploading via push_results_for_images
+    """
+    
+    assert is_valid_prediction_string(prediction_string), \
+        'Invalid prediction string: {}'.format(prediction_string)
+        
+    payload_detections = []
+    
+    if detections is not None:
+        # detection = detections[0]    
+        for detection in detections:
+            detection_out = detection.copy()
+            detection_out['label'] = md_category_id_to_name[detection['category']]
+            if detection_out['conf'] < min_md_output_confidence:
+                detection_out['conf'] = min_md_output_confidence
+            payload_detections.append(detection_out)
+        
+    prediction = {}
+    prediction['ignore_data_file_checks']  = True
+    prediction['prediction'] = prediction_string
+    prediction['prediction_score'] = prediction_confidence
+    prediction['classifications'] = {}
+    prediction['classifications']['classes'] = [prediction_string]
+    prediction['classifications']['scores'] = [prediction_confidence]
+    prediction['detections'] = payload_detections
+    prediction['model_version'] = model_version
+    prediction['prediction_source'] = prediction_source
+    prediction['data_file_id'] = data_file_id
+    prediction['project_id'] = project_id
+    
+    payload = {}
+    payload['predictions'] = [prediction]
+    
+    return payload
+
+
+def validate_payload(payload):
+    """
+    Verifies that the dict [payload] is compatible with the ProcessCVResponse() API.  Throws an
+    error if [payload] is invalid.
+    
+    Args:
+        payload (dict): payload in the format expected by push_results_for_images.
+    
+    Returns:
+        bool: successful validation; this is just future-proofing, currently never returns False
+    """
+    
+    assert isinstance(payload,dict)
+    assert len(payload.keys()) == 1 and 'predictions' in payload
+    
+    # prediction = payload['predictions'][0]
+    for prediction in payload['predictions']:
+        
+        assert 'project_id' in prediction
+        if not isinstance(prediction['project_id'],int):
+            _ = int(prediction['project_id'])
+        assert 'ignore_data_file_checks' in prediction and \
+            isinstance(prediction['ignore_data_file_checks'],bool)
+        assert 'prediction' in prediction and \
+            isinstance(prediction['prediction'],str) and \
+            len(prediction['prediction'].split(';')) == 7
+        assert 'prediction_score' in prediction and \
+            isinstance(prediction['prediction_score'],float)
+        assert 'model_version' in prediction and \
+            isinstance(prediction['model_version'],str)
+        assert 'data_file_id' in prediction and \
+            isinstance(prediction['data_file_id'],str) and \
+            len(prediction['data_file_id']) == 36
+        assert 'classifications' in prediction and \
+            isinstance(prediction['classifications'],dict)
+        classifications = prediction['classifications']
+        assert 'classes' in classifications and isinstance(classifications['classes'],list)
+        assert 'scores' in classifications and isinstance(classifications['scores'],list)
+        assert len(classifications['classes']) == len(classifications['scores'])
+        for c in classifications['classes']:
+            assert is_valid_prediction_string(c)            
+        for score in classifications['scores']:
+            assert isinstance(score,float) and score >= 0 and score <= 1.0
+        assert 'detections' in prediction and isinstance(prediction['detections'],list)
+        
+        for detection in prediction['detections']:
+            
+            assert isinstance(detection,dict)
+            assert 'category' in detection and detection['category'] in ('1','2','3')
+            assert 'label' in detection and detection['label'] in ('animal','person','vehicle')
+            assert 'conf' in detection and \
+                isinstance(detection['conf'],float) and \
+                detection['conf'] >= 0 and detection['conf'] <= 1.0
+            assert 'bbox' in detection and \
+                isinstance(detection['bbox'],list) and \
+                len(detection['bbox']) == 4
+                
+         # ...for each detection
+         
+    # ...for each prediction
+    
+    return True
+
+# ...def validate_payload(...)
+
+
+#%% Validate constants
+
+# This is executed at the time this module gets imported.
+
+blank_payload = generate_blank_prediction_payload('70ede9c6-d056-4dd1-9a0b-3098d8113e0e','1234')
+validate_payload(sample_update_payload)
+validate_payload(blank_payload)
+
+
+#%% Functions and constants related to working with batch predictions
+
+def get_kingdom(prediction_string):
+    """
+    Return the kingdom field from a WI prediction string
+    
+    Args:
+        prediction_string (str): a string in the semicolon-delimited prediction string format
+        
+    Returns:
+        str: the kingdom field from the input string
+    """
+    tokens = prediction_string.split(';')
+    assert is_valid_prediction_string(prediction_string)
+    return tokens[1]
+
+
+def is_human_classification(prediction_string):
+    """
+    Determines whether the input string represents a human classification, which includes a variety
+    of common names (hiker, person, etc.)
+    
+    Args:
+        prediction_string (str): a string in the semicolon-delimited prediction string format
+        
+    Returns:
+        bool: whether this string corresponds to a human category
+    """
+    return prediction_string == human_prediction_string or 'homo;sapiens' in prediction_string
+    
+
+def is_animal_classification(prediction_string):
+    """
+    Determines whether the input string represents an animal classification, which excludes, e.g.,
+    humans, blanks, vehicles, unknowns
+    
+    Args:
+        prediction_string (str): a string in the semicolon-delimited prediction string format
+        
+    Returns:
+        bool: whether this string corresponds to an animal category
+    """
+    
+    if prediction_string == animal_prediction_string:
+        return True
+    if prediction_string == human_prediction_string or 'homo;sapiens' in prediction_string:
+        return False
+    if prediction_string == blank_prediction_string:
+        return False
+    if prediction_string == no_cv_result_prediction_string:
+        return False
+    if len(get_kingdom(prediction_string)) == 0:
+        return False
+    return True
+
+
+def generate_md_results_from_predictions_json(predictions_json_file,
+                                              md_results_file,
+                                              base_folder=None,
+                                              max_decimals=5):
+    """
+    Generate an MD-formatted .json file from a predictions.json file.  Typically,
+    MD results files use relative paths, and predictions.json files use absolute paths, so 
+    this function optionally removes the leading string [base_folder] from all file names.
+    
+    Currently just applies the top classification category to every detection.  If the top classification 
+    is "blank", writes an empty detection list.
+    
+    speciesnet_to_md.py is a command-line driver for this function.
+    
+    Args:
+        predictions_json_file (str): path to a predictions.json file, or a dict
+        md_results_file (str): path to which we should write an MD-formatted .json file
+        base_folder (str, optional): leading string to remove from each path in the 
+            predictions.json file
+        max_decimals (int, optional): number of decimal places to which we should round
+            all values
+    """
+        
+    # Read predictions file    
+    if isinstance(predictions_json_file,str):
+        with open(predictions_json_file,'r') as f:
+            predictions = json.load(f)
+    else:
+        assert isinstance(predictions_json_file,dict)
+        predictions = predictions_json_file
+    
+    # Round floating-point values (confidence scores, coordinates) to a
+    # reasonable number of decimal places
+    if max_decimals is not None and max_decimals > 0:
+        round_floats_in_nested_dict(predictions)
+        
+    predictions = predictions['predictions']
+    assert isinstance(predictions,list)
+    
+    # Convert backslashes to forward slashes in both filenames and the base folder string
+    for im in predictions:
+        im['filepath'] = im['filepath'].replace('\\','/')
+    if base_folder is not None:
+        base_folder = base_folder.replace('\\','/')
+    
+    detection_category_id_to_name = {}
+    classification_category_name_to_id = {}
+    
+    # Keep track of detections that don't have an assigned detection category; these 
+    # are fake detections we create for non-blank images with non-empty detection lists.
+    # We need to go back later and give them a legitimate detection category ID.
+    all_unknown_detections = []
+    
+    # Create the output images list
+    images_out = []
+    
+    base_folder_replacements = 0
+    
+    # im_in = predictions[0]
+    for im_in in predictions:
+        
+        # blank_prediction_string
+        im_out = {}
+        
+        fn = im_in['filepath']
+        if base_folder is not None:
+            if fn.startswith(base_folder):
+                base_folder_replacements += 1
+                fn = fn.replace(base_folder,'',1)
+        
+        im_out['file'] = fn
+        
+        if 'failures' in im_in:
+            
+            im_out['failure'] = str(im_in['failures'])
+            im_out['detections'] = None
+            
+        else:
+            
+            im_out['detections'] = []
+            
+            if 'detections' in im_in:
+                
+                if len(im_in['detections']) == 0:
+                    im_out['detections'] = []
+                else:
+                    # det_in = im_in['detections'][0]
+                    for det_in in im_in['detections']:
+                        det_out = {}
+                        if det_in['category'] in detection_category_id_to_name:
+                            assert detection_category_id_to_name[det_in['category']] == det_in['label']
+                        else:
+                            detection_category_id_to_name[det_in['category']] = det_in['label']
+                        det_out = {}
+                        for s in ['category','conf','bbox']:
+                            det_out[s] = det_in[s]
+                        im_out['detections'].append(det_out)
+                        
+            # ...if detections are present
+            
+            class_to_assign = None
+            class_confidence = None
+            
+            if 'classifications' in im_in:
+                
+                classifications = im_in['classifications']
+                assert len(classifications['scores']) == len(classifications['classes'])
+                assert is_list_sorted(classifications['scores'],reverse=True)
+                class_to_assign = classifications['classes'][0]
+                class_confidence = classifications['scores'][0]
+                
+            if 'prediction' in im_in:
+                
+                class_to_assign = im_in['prediction']
+                class_confidence = im_in['prediction_score']
+            
+            if class_to_assign is not None:
+                
+                if class_to_assign == blank_prediction_string:
+                
+                    # This is a scenario that's not captured well by the MD format: a blank prediction
+                    # with detections present.  But, for now, don't do anything special here, just making
+                    # a note of this.
+                    if len(im_out['detections']) > 0:
+                        pass                    
+                
+                else:
+                    
+                    assert not class_to_assign.endswith('blank')
+                    
+                    # This is a scenario that's not captured well by the MD format: no detections present,
+                    # but a non-blank prediction.  For now, create a fake detection to handle this prediction.
+                    if len(im_out['detections']) == 0:
+                    
+                        print('Warning: creating fake detection for non-blank whole-image classification')
+                        det_out = {}
+                        all_unknown_detections.append(det_out)
+                        
+                        # We will change this to a string-int later
+                        det_out['category'] = 'unknown'
+                        det_out['conf'] = class_confidence
+                        det_out['bbox'] = [0,0,1,1]
+                        im_out['detections'].append(det_out)
+                
+                # ...if this is/isn't a blank classification
+                
+                # Attach that classification to each detection
+                
+                # Create a new category ID if necessary
+                if class_to_assign in classification_category_name_to_id:
+                    classification_category_id = classification_category_name_to_id[class_to_assign]
+                else:
+                    classification_category_id = str(len(classification_category_name_to_id))
+                    classification_category_name_to_id[class_to_assign] = classification_category_id
+                
+                for det in im_out['detections']:
+                    det['classifications'] = []
+                    det['classifications'].append([classification_category_id,class_confidence])
+                
+            # ...if we have some type of classification for this image
+        
+        # ...if this is/isn't a failure
+        
+        images_out.append(im_out)
+        
+    # ...for each image
+    
+    if base_folder is not None:
+        if base_folder_replacements == 0:
+            print('Warning: you supplied {} as the base folder, but I made zero replacements'.format(
+                base_folder))
+            
+    # Fix the 'unknown' category
+    
+    if len(all_unknown_detections) > 0:
+        
+        max_detection_category_id = max([int(x) for x in detection_category_id_to_name.keys()])
+        unknown_category_id = str(max_detection_category_id + 1)
+        detection_category_id_to_name[unknown_category_id] = 'unknown'
+        
+        for det in all_unknown_detections:
+            assert det['category'] == 'unknown'
+            det['category'] = unknown_category_id
+                    
+    
+    # Sort by filename
+    
+    images_out = sort_list_of_dicts_by_key(images_out,'file')
+    
+    # Prepare friendly classification names
+    
+    classification_category_descriptions = \
+        invert_dictionary(classification_category_name_to_id)
+    classification_categories_out = {}
+    for category_id in classification_category_descriptions.keys():
+        category_name = classification_category_descriptions[category_id].split(';')[-1]
+        classification_categories_out[category_id] = category_name
+    
+    # Prepare the output dict
+    
+    detection_categories_out = detection_category_id_to_name
+    info = {}
+    info['format_version'] = 1.4
+    info['detector'] = 'converted_from_predictions_json'
+    
+    output_dict = {}
+    output_dict['info'] = info
+    output_dict['detection_categories'] = detection_categories_out
+    output_dict['classification_categories'] = classification_categories_out
+    output_dict['classification_category_descriptions'] = classification_category_descriptions
+    output_dict['images'] = images_out
+    
+    with open(md_results_file,'w') as f:
+        json.dump(output_dict,f,indent=1)
+    
+    validation_options = ValidateBatchResultsOptions()
+    validation_options.raise_errors = True    
+    _ = validate_batch_results(md_results_file, options=validation_options)
+    
+# ...def generate_md_results_from_predictions_json(...)
+        
+
+def generate_predictions_json_from_md_results(md_results_file,
+                                              predictions_json_file,
+                                              base_folder=None):
+    """
+    Generate a predictions.json file from the MD-formatted .json file [md_results_file].  Typically,
+    MD results files use relative paths, and predictions.json files use absolute paths, so 
+    this function optionally prepends [base_folder].  Does not handle classification results in
+    MD format, since this is intended to prepare data for passing through the WI classifier.
+    
+    md_to_wi.py is a command-line driver for this function.
+    
+    Args:
+        md_results_file (str): path to an MD-formatted .json file
+        predictions_json_file (str): path to which we should write a predictions.json file
+        base_folder (str, optional): folder name to prepend to each path in md_results_file,
+            to convert relative paths to absolute paths.    
+    """
+        
+    # Validate the input file
+    validation_options = ValidateBatchResultsOptions()
+    validation_options.raise_errors = True    
+    validation_options.return_data = True
+    md_results = validate_batch_results(md_results_file, options=validation_options)
+    category_id_to_name = md_results['detection_categories']
+    
+    output_dict = {}
+    output_dict['predictions'] = []
+        
+    # im = md_results['images'][0]
+    for im in md_results['images']:
+        
+        prediction = {}
+        fn = im['file']
+        if base_folder is not None:
+            fn = os.path.join(base_folder,fn)
+        fn = fn.replace('\\','/')
+        prediction['filepath'] = fn
+        if 'failure' in im and im['failure'] is not None:
+            prediction['failures'] = ['DETECTOR']
+        else:
+            assert 'detections' in im and im['detections'] is not None
+            detections = []
+            for det in im['detections']:
+                output_det = deepcopy(det)
+                output_det['label'] = category_id_to_name[det['category']]
+                detections.append(output_det)
+            
+            # detections *must* be sorted in descending order by confidence
+            detections = sort_list_of_dicts_by_key(detections,'conf', reverse=True)
+            prediction['detections'] = detections
+            
+        assert len(prediction.keys()) >= 2
+        output_dict['predictions'].append(prediction)
+            
+    # ...for each image
+        
+    os.makedirs(os.path.dirname(predictions_json_file),exist_ok=True)
+    with open(predictions_json_file,'w') as f:
+        json.dump(output_dict,f,indent=1)
+        
+# ...def generate_predictions_json_from_md_results(...)
+
+default_tokens_to_ignore = ['$RECYCLE.BIN']
+
+def generate_instances_json_from_folder(folder,
+                                        country=None,
+                                        admin1_region=None,
+                                        lat=None,
+                                        lon=None,
+                                        output_file=None,
+                                        filename_replacements=None,
+                                        tokens_to_ignore=default_tokens_to_ignore):
+    """
+    Generate an instances.json record that contains all images in [folder], optionally
+    including location information, in a format suitable for run_model.py.  Optionally writes
+    the results to [output_file].
+    
+    Args:
+        folder (str): the folder to recursively search for images
+        country (str, optional): a three-letter country code
+        lat (float, optional): latitude to associate with all images
+        lon (float, optional): longitude to associate with all images
+        output_file (str, optional): .json file to which we should write instance records
+        filename_replacements (dict, optional): str --> str dict indicating filename substrings
+            that should be replaced with other strings.  Replacement occurs *after* converting
+            backslashes to forward slashes.
+        tokens_to_ignore (list, optional): ignore any images with these tokens in their
+            names, typically used to avoid $RECYCLE.BIN.  Can be None.
+        
+    Returns:
+        dict: dict with at least the field "instances"
+    """
+    
+    assert os.path.isdir(folder)
+        
+    image_files_abs = find_images(folder,recursive=True,return_relative_paths=False)
+    
+    if tokens_to_ignore is not None:
+        n_images_before_ignore_tokens = len(image_files_abs)
+        for token in tokens_to_ignore:
+            image_files_abs = [fn for fn in image_files_abs if token not in fn]
+        print('After ignoring {} tokens, kept {} of {} images'.format(
+            len(tokens_to_ignore),len(image_files_abs),n_images_before_ignore_tokens))
+    
+    instances = []
+    
+    # image_fn_abs = image_files_abs[0]
+    for image_fn_abs in image_files_abs:
+        instance = {}
+        instance['filepath'] = image_fn_abs.replace('\\','/')
+        if filename_replacements is not None:
+            for s in filename_replacements:
+                instance['filepath'] = instance['filepath'].replace(s,filename_replacements[s])
+        if country is not None:
+            instance['country'] = country
+        if admin1_region is not None:
+            instance['admin1_region'] = admin1_region
+        if lat is not None:
+            assert lon is not None, 'Latitude provided without longitude'
+            instance['latitude'] = lat
+        if lon is not None:
+            assert lat is not None, 'Longitude provided without latitude'
+            instance['longitude'] = lon
+        instances.append(instance)
+    
+    to_return = {'instances':instances}
+    
+    if output_file is not None:
+        os.makedirs(os.path.dirname(output_file),exist_ok=True)
+        with open(output_file,'w') as f:
+            json.dump(to_return,f,indent=1)
+    
+    return to_return
+        
+# ...def generate_instances_json_from_folder(...)
+
+
+def split_instances_into_n_batches(instances_json,n_batches,output_files=None):
+    """
+    Given an instances.json file, split it into batches of equal size.
+    
+    Args:
+        instances_json (str): input .json file in
+        n_batches (int): number of new files to generate
+        output_files (list, optional): output .json files for each
+            batch.  If supplied, should have length [n_batches].  If not 
+            supplied, filenames will be generated based on [instances_json].
+            
+    Returns:
+        list: list of output files that were written; identical to [output_files]
+        if it was supplied as input.
+    """
+    
+    with open(instances_json,'r') as f:
+        instances = json.load(f)
+    assert isinstance(instances,dict) and 'instances' in instances
+    instances = instances['instances']
+    
+    if output_files is not None:
+        assert len(output_files) == n_batches, \
+            'Expected {} output files, received {}'.format(
+                n_batches,len(output_files))
+    else:
+        output_files = []
+        for i_batch in range(0,n_batches):
+            batch_string = 'batch_{}'.format(str(i_batch).zfill(3))
+            output_files.append(insert_before_extension(instances_json,batch_string))
+                    
+    batches = split_list_into_n_chunks(instances, n_batches)
+    
+    for i_batch,batch in enumerate(batches):
+        batch_dict = {'instances':batch}
+        with open(output_files[i_batch],'w') as f:
+            json.dump(batch_dict,f,indent=1)
+    
+    print('Wrote {} batches to file'.format(n_batches))
+    
+    return output_files
+    
+    
+def merge_prediction_json_files(input_prediction_files,output_prediction_file):
+    """
+    Merge all predictions.json files in [files] into a single .json file.
+    
+    Args:
+        files (list): list of predictions.json files to merge
+        output_file (str): output .json file        
+    """
+    
+    predictions = []
+    image_filenames_processed = set()
+    
+    # input_json_fn = input_prediction_files[0]
+    for input_json_fn in tqdm(input_prediction_files):
+        
+        assert os.path.isfile(input_json_fn), \
+            'Could not find prediction file {}'.format(input_json_fn)            
+        with open(input_json_fn,'r') as f:
+            results_this_file = json.load(f)
+        assert isinstance(results_this_file,dict)
+        predictions_this_file = results_this_file['predictions']
+        for prediction in predictions_this_file:
+            image_fn = prediction['filepath']
+            assert image_fn not in image_filenames_processed
+        predictions.extend(predictions_this_file)
+        
+    output_dict = {'predictions':predictions}
+    
+    os.makedirs(os.path.dirname(output_prediction_file),exist_ok=True)
+    with open(output_prediction_file,'w') as f:
+        json.dump(output_dict,f,indent=1)
+
+# ...def merge_prediction_json_files(...)    
+
+
+def validate_predictions_file(fn,instances=None,verbose=True):
+    """
+    Validate the predictions.json file [fn].
+    
+    Args:
+        fn (str): a .json file in predictions.json (SpeciesNet) format
+        instances (str or list, optional): a folder, instances.json file,
+            or dict loaded from an instances.json file.  If supplied, this
+            function will verify that [fn] contains the same number of 
+            images as [instances].
+        verbose (bool, optional): enable additional debug output
+        
+    Returns:
+        dict: the contents of [fn]        
+    """
+    
+    with open(fn,'r') as f:
+        d = json.load(f)
+    predictions = d['predictions']
+        
+    failures = []
+
+    for im in predictions:
+        if 'failures' in im:
+            failures.append(im)
+
+    if verbose:
+        print('Read detector results for {} images, with {} failure(s)'.format(
+            len(d['predictions']),len(failures)))
+        
+    if instances is not None:
+        
+        if isinstance(instances,str):
+            if os.path.isdir(instances):
+                instances = generate_instances_json_from_folder(folder=instances)
+            elif os.path.isfile(instances):
+                with open(instances,'r') as f:
+                    instances = json.load(f)
+            else:
+                raise ValueError('Could not find instances file/folder {}'.format(
+                    instances))
+        assert isinstance(instances,dict)
+        assert 'instances' in instances
+        instances = instances['instances']
+        if verbose:
+            print('Expected results for {} files'.format(len(instances)))
+        assert len(instances) == len(predictions), \
+            '{} instances expected, {} found'.format(
+                len(instances),len(predictions))
+        
+        expected_files = set([instance['filepath'] for instance in instances])
+        found_files = set([prediction['filepath'] for prediction in predictions])
+        assert expected_files == found_files
+        
+    # ...if a list of instances was supplied
+
+    return d
+
+# ...def validate_predictions_file(...)        
+
+
+def find_geofence_adjustments(ensemble_json_file,use_latin_names=False):
+    """
+    Count the number of instances of each unique change made by the geofence.
+    
+    Args:
+        ensemble_json_file (str): SpeciesNet-formatted .json file produced
+            by the full ensemble.
+        use_latin_names (bool, optional): return a mapping using binomial names
+            rather than common names.
+    
+    Returns:
+        dict: maps strings that look like "puma,felidae family" to integers, 
+            where that entry would indicate the number of times that "puma" was 
+            predicted, but mapped to family level by the geofence.  Sorted in 
+            descending order by count.
+    """
+    
+    ensemble_results = validate_predictions_file(ensemble_json_file)
+    
+    assert isinstance(ensemble_results,dict)
+    predictions = ensemble_results['predictions']
+    
+    # Maps comma-separated pairs of common names (or binomial names) to
+    # the number of times that transition (first --> second) happened
+    rollup_pair_to_count = defaultdict(int)
+    
+    # prediction = predictions[0]
+    for prediction in tqdm(predictions):
+        
+        if 'failures' in prediction and \
+            prediction['failures'] is not None and \
+            len(prediction['failures']) > 0:
+                continue
+            
+        assert 'prediction_source' in prediction, \
+            'Prediction present without [prediction_source] field, are you sure this ' + \
+            'is an ensemble output file?'
+        
+        if 'geofence' in prediction['prediction_source']:
+            
+            classification_taxonomy_string = \
+                prediction['classifications']['classes'][0]
+            prediction_taxonomy_string = prediction['prediction']
+            assert is_valid_prediction_string(classification_taxonomy_string)
+            assert is_valid_prediction_string(prediction_taxonomy_string)
+
+            # Typical examples:
+            # '86f5b978-4f30-40cc-bd08-be9e3fba27a0;mammalia;rodentia;sciuridae;sciurus;carolinensis;eastern gray squirrel'
+            # 'e4d1e892-0e4b-475a-a8ac-b5c3502e0d55;mammalia;rodentia;sciuridae;;;sciuridae family'
+            classification_common_name = classification_taxonomy_string.split(';')[-1]
+            prediction_common_name = prediction_taxonomy_string.split(';')[-1]
+            classification_binomial_name = classification_taxonomy_string.split(';')[-2]
+            prediction_binomial_name = prediction_taxonomy_string.split(';')[-2]
+            
+            input_name = classification_binomial_name if use_latin_names else \
+                classification_common_name
+            output_name = prediction_binomial_name if use_latin_names else \
+                prediction_common_name
+                
+            rollup_pair = input_name.strip() + ',' + output_name.strip()
+            rollup_pair_to_count[rollup_pair] += 1
+            
+        # ...if we made a geofencing change
+        
+    # ...for each prediction
+           
+    rollup_pair_to_count = sort_dictionary_by_value(rollup_pair_to_count,reverse=True)
+    
+    return rollup_pair_to_count
+
+# ...def find_geofence_adjustments(...)    
+
+
+#%% Module-level globals related to taxonomy mapping and geofencing    
+
+# This maps a taxonomy string (e.g. mammalia;cetartiodactyla;cervidae;odocoileus;virginianus) to 
+# a dict with keys taxon_id, common_name, kingdom, phylum, class, order, family, genus, species
+taxonomy_string_to_taxonomy_info = None
+
+# Maps a binomial name (possibly three tokens, if it's a subspecies) to the same dict
+# described above.
+binomial_name_to_taxonomy_info = None
+
+# Maps a common name to the same dict described above
+common_name_to_taxonomy_info = None
+
+# Dict mapping 5-token semicolon-delimited taxonomy strings to geofencing rules
+taxonomy_string_to_geofencing_rules = None
+
+# Maps lower-case country names to upper-case country codes
+country_to_country_code = None
+
+# Maps upper-case country codes to lower-case country names
+country_code_to_country = None
+
+
+#%% Functions related to geofencing and taxonomy mapping
+
+def taxonomy_info_to_taxonomy_string(taxonomy_info):
+    """
+    Convert a taxonomy record in dict format to a semicolon-delimited string
+    
+    Args:
+        taxonomy_info (dict): dict in the format stored in, e.g., taxonomy_string_to_taxonomy_info
+    
+    Returns:
+        str: string in the format used as keys in, e.g., taxonomy_string_to_taxonomy_info
+    """
+    return taxonomy_info['class'] + ';' + \
+        taxonomy_info['order'] + ';'  + \
+        taxonomy_info['family'] + ';'  + \
+        taxonomy_info['genus'] + ';'  + \
+        taxonomy_info['species']
+        
+
+def initialize_taxonomy_info(taxonomy_file,force_init=False,encoding='cp1252'):
+    """
+    Load WI taxonomy information from a .json file.  Stores information in the global
+    dicts [taxonomy_string_to_taxonomy_info], [binomial_name_to_taxonomy_info], and
+    [common_name_to_taxonomy_info].
+    
+    Args:
+        taxonomy_file (str): .json file containing mappings from the short taxonomy strings 
+            to the longer strings with GUID and common name, see example below.            
+        force_init (bool, optional): if the output dicts already exist, should we
+            re-initialize anyway?
+        encoding (str, optional): character encoding to use when opening the .json file
+    """
+    
+    if encoding is None:
+        encoding = 'cp1252'
+        
+    global taxonomy_string_to_taxonomy_info
+    global binomial_name_to_taxonomy_info
+    global common_name_to_taxonomy_info
+    
+    if (taxonomy_string_to_taxonomy_info is not None) and (not force_init):
+        return
+    
+    """
+    Taxonomy keys are taxonomy strings, e.g.:
+    
+    'mammalia;cetartiodactyla;cervidae;odocoileus;virginianus'
+    
+    Taxonomy values are extended strings w/Taxon IDs and common names, e.g.:
+        
+    '5c7ce479-8a45-40b3-ae21-7c97dfae22f5;mammalia;cetartiodactyla;cervidae;odocoileus;virginianus;white-tailed deer'
+    """
+    
+    with open(taxonomy_file,encoding=encoding,errors='ignore') as f:
+        taxonomy_table = json.load(f,strict=False)    
+    
+    # Right now I'm punting on some unusual-character issues, but here is some scrap that
+    # might help address this in the future
+    if False:
+        import codecs
+        with codecs.open(taxonomy_file,'r',encoding=encoding,errors='ignore') as f:
+            s = f.read()        
+        import unicodedata
+        s = unicodedata.normalize('NFKD', s).encode('ascii', 'ignore')    
+        taxonomy_table = json.loads(s,strict=False)
+        
+    taxonomy_string_to_taxonomy_info = {}
+    binomial_name_to_taxonomy_info = {}
+    common_name_to_taxonomy_info = {}
+    
+    # taxonomy_string = next(iter(taxonomy_table.keys()))
+    for taxonomy_string in taxonomy_table.keys():
+        
+        taxonomy_string = taxonomy_string.lower()
+        
+        taxon_info = {}
+        extended_string = taxonomy_table[taxonomy_string]
+        tokens = extended_string.split(';')
+        assert len(tokens) == 7
+        taxon_info['taxon_id'] = tokens[0]
+        assert len(taxon_info['taxon_id']) == 36
+        taxon_info['kingdom'] = 'animal'
+        taxon_info['phylum'] = 'chordata'
+        taxon_info['class'] = tokens[1]
+        taxon_info['order'] = tokens[2]
+        taxon_info['family'] = tokens[3]
+        taxon_info['genus'] = tokens[4]
+        taxon_info['species'] = tokens[5]
+        taxon_info['common_name'] = tokens[6]
+        
+        if taxon_info['common_name'] != '':
+            common_name_to_taxonomy_info[taxon_info['common_name']] = taxon_info
+            
+        taxonomy_string_to_taxonomy_info[taxonomy_string] = taxon_info
+        if tokens[4] == '' or tokens[5] == '':
+            # print('Warning: no binomial name for {}'.format(taxonomy_string))
+            pass
+        else:
+            # strip(), but don't remove spaces from the species name; 
+            # subspecies are separated with a space, e.g. canis;lupus dingo
+            binomial_name = tokens[4].strip() + ' ' + tokens[5].strip()
+            binomial_name_to_taxonomy_info[binomial_name] = taxon_info
+    
+    print('Created {} records in taxonomy_string_to_taxonomy_info'.format(
+        len(taxonomy_string_to_taxonomy_info)))
+        
+# ...def initialize_taxonomy_info(...)
+
+
+def _parse_code_list(codes):
+    """
+    Turn a list of country or state codes in string, delimited string, or list format
+    into a list.  Also does basic validity checking.
+    """
+    
+    if not isinstance(codes,list):
+        
+        assert isinstance(codes,str)
+        
+        codes = codes.strip()
+        
+        # This is just a single codes
+        if ',' not in codes:
+            codes = [codes]
+        else:
+            codes = codes.split(',')
+        codes = [c.strip() for c in codes]
+        
+    assert isinstance(codes,list)
+    
+    codes = [c.upper().strip() for c in codes]
+    
+    for c in codes:
+        assert len(c) in (2,3)
+        
+    return codes
+        
+        
+def _generate_csv_rows_to_block_all_countries_except(
+        species_string,
+        block_except_list):
+    """
+    Generate rows in the format expected by geofence_fixes.csv, representing a list of
+    allow and block rules to block all countries currently allowed for this species
+    except [allow_countries], and add allow rules these countries.
+    """
+    
+    assert is_valid_taxonomy_string(species_string), \
+        '{} is not a valid taxonomy string'.format(species_string)
+        
+    global taxonomy_string_to_taxonomy_info
+    global binomial_name_to_taxonomy_info
+    global common_name_to_taxonomy_info
+    
+    assert taxonomy_string_to_geofencing_rules is not None, \
+        'Initialize geofencing prior to species lookup'
+    assert taxonomy_string_to_taxonomy_info is not None, \
+        'Initialize taxonomy lookup prior to species lookup'
+        
+    geofencing_rules_this_species = \
+        taxonomy_string_to_geofencing_rules[species_string]
+        
+    allowed_countries = []
+    if 'allow' in geofencing_rules_this_species:
+        allowed_countries.extend(geofencing_rules_this_species['allow'])
+        
+    blocked_countries = []
+    if 'block' in geofencing_rules_this_species:
+        blocked_countries.extend(geofencing_rules_this_species['block'])
+        
+    block_except_list = _parse_code_list(block_except_list)
+    
+    countries_to_block = []
+    countries_to_allow = []
+    
+    # country = allowed_countries[0]
+    for country in allowed_countries:
+        if country not in block_except_list and country not in blocked_countries:
+            countries_to_block.append(country)
+    
+    for country in block_except_list:
+        if country in blocked_countries:
+            raise ValueError("I can't allow a country that has already been blocked")
+        if country not in allowed_countries:
+            countries_to_allow.append(country)
+            
+    rows = generate_csv_rows_for_species(species_string,
+                                         allow_countries=countries_to_allow,
+                                         block_countries=countries_to_block)
+    
+    return rows
+   
+# ...def _generate_csv_rows_to_block_all_countries_except(...)
+    
+    
+def generate_csv_rows_for_species(species_string,
+                                  allow_countries=None,
+                                  block_countries=None,
+                                  allow_states=None,
+                                  block_states=None,
+                                  blockexcept_countries=None):
+    """
+    Generate rows in the format expected by geofence_fixes.csv, representing a list of
+    allow and/or block rules for the specified species and countries/states.  Does not check 
+    that the rules make sense; e.g. nothing will stop you in this function from both allowing
+    and blocking a country.
+    
+    Args:
+        species_string (str): string in semicolon-delimited WI taxonomy format
+        allow_countries (optional, list or str): three-letter country codes, list of
+            country codes, or comma-separated list of country codes to allow
+        block_countries (optional, list or str): three-letter country codes, list of
+            country codes, or comma-separated list of country codes to block
+        allow_states (optional, list or str): two-letter state codes, list of
+            state codes, or comma-separated list of state codes to allow
+        block_states (optional, list or str): two-letter state code, list of
+            state codes, or comma-separated list of state codes to block        
+            
+    Returns:
+        list of str: lines ready to be pasted into geofence_fixes.csv
+    """
+    
+    assert is_valid_taxonomy_string(species_string), \
+        '{} is not a valid taxonomy string'.format(species_string)
+    
+    lines = []
+    
+    if allow_countries is not None:
+        allow_countries = _parse_code_list(allow_countries)
+        for country in allow_countries:
+            lines.append(species_string + ',allow,' + country + ',')
+    
+    if block_countries is not None:
+        block_countries = _parse_code_list(block_countries)
+        for country in block_countries:
+            lines.append(species_string + ',block,' + country + ',')
+            
+    if allow_states is not None:
+        allow_states = _parse_code_list(allow_states)
+        for state in allow_states:
+            lines.append(species_string + ',allow,USA,' + state)
+    
+    if block_states is not None:
+        block_states = _parse_code_list(block_states)
+        for state in block_states:
+            lines.append(species_string + ',block,USA,' + state)
+         
+    return lines
+
+# ...def generate_csv_rows_for_species(...)
+
+
+def initialize_geofencing(geofencing_file,country_code_file,force_init=False):
+    """
+    Load geofencing information from a .json file, and country code mappings from
+    a .csv file.  Stores results in the global tables [taxonomy_string_to_geofencing_rules],
+    [country_to_country_code], and [country_code_to_country].
+    
+    Args:
+        geofencing_file (str): .json file with geofencing rules
+        country_code_file (str): .csv file with country code mappings, in columns 
+            called "name" and "alpha-3", e.g. from
+            https://github.com/lukes/ISO-3166-Countries-with-Regional-Codes/blob/master/all/all.csv
+        force_init (bool, optional): if the output dicts already exist, should we
+            re-initialize anyway?
+    """
+    
+    global taxonomy_string_to_geofencing_rules
+    global country_to_country_code
+    global country_code_to_country
+    
+    if (country_to_country_code is not None) and \
+        (country_code_to_country is not None) and \
+        (taxonomy_string_to_geofencing_rules is not None) and \
+        (not force_init):
+        return
+    
+    # Read country code information
+    country_code_df = pd.read_csv(country_code_file)
+    country_to_country_code = {}
+    country_code_to_country = {}
+    for i_row,row in country_code_df.iterrows():
+        country_to_country_code[row['name'].lower()] = row['alpha-3'].upper()
+        country_code_to_country[row['alpha-3'].upper()] = row['name'].lower()
+                    
+    # Read geofencing information
+    with open(geofencing_file,'r',encoding='utf-8') as f:
+        taxonomy_string_to_geofencing_rules = json.load(f)
+    
+    """
+    Geofencing keys are taxonomy strings, e.g.:
+      
+    'mammalia;cetartiodactyla;cervidae;odocoileus;virginianus'
+    
+    Geofencing values are tables mapping allow/block to country codes, optionally including region/state codes, e.g.:
+        
+    {'allow': {
+      'ALA': [],
+      'ARG': [],
+      ...
+      'SUR': [],
+      'TTO': [],
+      'USA': ['AL',
+       'AR',
+       'AZ',
+       ...
+    }
+    """        
+        
+    # Validate
+    
+    # species_string = next(iter(taxonomy_string_to_geofencing_rules.keys()))
+    for species_string in taxonomy_string_to_geofencing_rules.keys():
+        
+        species_rules = taxonomy_string_to_geofencing_rules[species_string]
+        
+        # Every country should *either* have allow rules or block rules, no countries 
+        # currently have both        
+        assert len(species_rules.keys()) == 1
+        rule_type = list(species_rules.keys())[0]
+        assert rule_type in ('allow','block')
+
+        all_country_rules_this_species = species_rules[rule_type]
+        for country_code in all_country_rules_this_species.keys():
+            
+            assert country_code in country_code_to_country
+        
+            region_rules = all_country_rules_this_species[country_code]
+            
+            # Right now we only have regional rules for the USA; these may be part of 
+            # allow or block rules.
+            if len(region_rules) > 0:
+                assert country_code == 'USA'
+    
+    # ...for each species
+    
+# ...def initialize_geofencing(...)
+    
+
+def _species_string_to_canonical_species_string(species):
+    """
+    Convert a string that may be a 5-token species string, a binomial name,
+    or a common name into a 5-token species string.
+    """
+    
+    global taxonomy_string_to_taxonomy_info
+    global binomial_name_to_taxonomy_info
+    global common_name_to_taxonomy_info
+    
+    assert taxonomy_string_to_geofencing_rules is not None, \
+        'Initialize geofencing prior to species lookup'
+    assert taxonomy_string_to_taxonomy_info is not None, \
+        'Initialize taxonomy lookup prior to species lookup'
+            
+    species = species.lower()
+    
+    # Turn "species" into a taxonomy string
+
+    # If this is already a taxonomy string...    
+    if len(species.split(';')) == 5:
+        pass
+    # If this is a binomial name (which may include a subspecies)...
+    elif (len(species.split(' ')) in (2,3)) and (species in binomial_name_to_taxonomy_info):
+        taxonomy_info = binomial_name_to_taxonomy_info[species]
+        taxonomy_string = taxonomy_info_to_taxonomy_string(taxonomy_info)        
+    # If this is a common name...
+    elif species in common_name_to_taxonomy_info:
+        taxonomy_info = common_name_to_taxonomy_info[species]
+        taxonomy_string = taxonomy_info_to_taxonomy_string(taxonomy_info)        
+    else:
+        raise ValueError('Could not find taxonomic information for {}'.format(species))
+        
+    return taxonomy_string
+            
+
+def species_allowed_in_country(species,country,state=None,return_status=False):
+    """
+    Determines whether [species] is allowed in [country], according to 
+    already-initialized geofencing rules.
+    
+    Args:
+        species (str): can be a common name, a binomial name, or a species string
+        country (str): country name or three-letter code
+        state (str, optional): two-letter US state code
+        return_status (bool, optional): by default, this function returns a bool;
+            if you want to know *why* [species] is allowed/not allowed, settings
+            return_status to True will return additional information.        
+    
+    Returns:
+        bool or str: typically returns True if [species] is allowed in [country], else
+        False.  Returns a more detailed string if return_status is set.
+    """
+
+    global taxonomy_string_to_taxonomy_info
+    global binomial_name_to_taxonomy_info
+    global common_name_to_taxonomy_info
+    
+    assert taxonomy_string_to_geofencing_rules is not None, \
+        'Initialize geofencing prior to species lookup'
+    assert taxonomy_string_to_taxonomy_info is not None, \
+        'Initialize taxonomy lookup prior to species lookup'
+        
+    taxonomy_string = _species_string_to_canonical_species_string(species)
+    
+    # Normalize [state]
+    
+    if state is not None:
+        state = state.upper()
+        assert len(state) == 2
+        
+    # Turn "country" into a country code
+    
+    if len(country) == 3:
+        assert country.upper() in country_code_to_country        
+        country = country.upper()
+    else:
+        assert country.lower() in country_to_country_code
+        country = country_to_country_code[country.lower()]
+        
+    country_code = country.upper()
+        
+    # Species with no rules are allowed everywhere
+    if taxonomy_string not in taxonomy_string_to_geofencing_rules:
+        status = 'allow_by_default'
+        if return_status:
+            return status
+        else:
+            return True
+    
+    geofencing_rules_this_species = taxonomy_string_to_geofencing_rules[taxonomy_string]
+    allowed_countries = []
+    blocked_countries = []
+    
+    assert len(geofencing_rules_this_species.keys()) == 1
+    rule_type = list(geofencing_rules_this_species.keys())[0]
+    assert rule_type in ('allow','block')
+    
+    if rule_type == 'allow':    
+        allowed_countries = list(geofencing_rules_this_species['allow'])
+    else:
+        assert rule_type == 'block'
+        blocked_countries = list(geofencing_rules_this_species['block'])
+    
+    status = None
+    
+    # The convention is that block rules win over allow rules
+    if country_code in blocked_countries:
+        status = 'blocked'
+    elif country_code in allowed_countries:
+        status = 'allowed'
+    else:
+        # The convention is that if allow rules exist, any country not on that list
+        # is blocked.
+        assert len(allowed_countries) > 0
+        return 'not_on_country_allow_list'
+    
+    # Now let's see whether we have to deal with any regional rules
+    if state is None:
+        
+        # If state rules are provided, we need to have a state
+        if country_code == 'USA':
+            state_list = geofencing_rules_this_species[rule_type][country_code]
+            if len(state_list) > 0:
+                raise ValueError('Cannot determine status for a species with state-level rules with no state information')
+                
+    else:
+        
+        # Right now state-level rules only exist for the US
+        assert country_code == 'USA'
+        state_list = geofencing_rules_this_species[rule_type][country_code]
+        
+        if state in state_list:
+            # If the state is on the list, do what the list says
+            if rule_type == 'allow':
+                status = 'allow_on_state_allow_list'
+            else:
+                status = 'block_on_state_block_list'
+        else:
+            # If the state is not on the list, do the opposite of what the list says
+            if rule_type == 'allow':
+                status = 'block_not_on_state_allow_list'
+            else:
+                status = 'allow_not_on_state_block_list'
+        
+    if return_status:
+        return status
+    else:
+        if status.startswith('allow'):
+            return True
+        else:
+            assert status.startswith('block')
+            return False
+
+# ...def species_allowed_in_country(...)
+
+
+def restrict_to_taxa_list(taxa_list,
+                          speciesnet_taxonomy_file,
+                          input_file,
+                          output_file,
+                          allow_walk_down=False):
+    """
+    Given a prediction file in MD .json format, likely without having had
+    a geofence applied, apply a custom taxa list.
+    
+    Args:
+        taxa_list (str or list): list of latin names, or a text file containing
+            a list of latin names.  Optionally may contain a second (comma-delimited)
+            column containing common names, used only for debugging.  Latin names
+            must exist in the SpeciesNet taxonomy.
+        taxonomy_file (str): taxonomy filename, in the same format used for model
+            release (with 7-token taxonomy entries)            
+        output_file (str): .json file to write, in MD format      
+        allow_walk_down (bool, optional): should we walk down the taxonomy tree
+            when making mappings if a parent has only a single allowable child?
+            For example, if only a single felid species is allowed, should other
+            felid predictions be mapped to that species, as opposed to being mapped
+            to the family?
+    """
+
+    ##%% Read target taxa list
+    
+    if isinstance(taxa_list,str):
+        assert os.path.isfile(taxa_list), \
+            'Could not find taxa list file {}'.format(taxa_list)
+        with open(taxa_list,'r') as f:
+            taxa_list = f.readlines()
+    
+    taxa_list = [s.strip().lower() for s in taxa_list]
+    taxa_list = [s for s in taxa_list if len(s) > 0]
+    
+    target_latin_to_common = {}
+    for s in taxa_list:
+        if s.strip().startswith('#'):
+            continue
+        tokens = s.split(',')
+        assert len(tokens) <= 2
+        binomial_name = tokens[0]
+        assert len(binomial_name.split(' ')) in (1,2,3), \
+            'Illegal binomial name in species list: {}'.format(binomial_name)
+        if len(tokens) > 0:
+            common_name = tokens[1].strip().lower()
+        else:
+            common_name = None
+        assert binomial_name not in target_latin_to_common
+        target_latin_to_common[binomial_name] = common_name
+    
+
+    ##%% Read taxonomy file
+    
+    with open(speciesnet_taxonomy_file,'r') as f:
+        speciesnet_taxonomy_list = f.readlines()
+    speciesnet_taxonomy_list = [s.strip() for s in \
+                                speciesnet_taxonomy_list if len(s.strip()) > 0]
+    
+    # Maps the latin name of every taxon to the corresponding full taxon string
+    #
+    # For species, the key is a binomial name
+    speciesnet_latin_name_to_taxon_string = {}
+    speciesnet_common_name_to_taxon_string = {}
+    
+    def _insert_taxonomy_string(s):
+        
+        tokens = s.split(';')
+        assert len(tokens) == 7
+        
+        guid = tokens[0] # noqa
+        class_name = tokens[1]
+        order = tokens[2]
+        family = tokens[3]
+        genus = tokens[4]
+        species = tokens[5]        
+        common_name = tokens[6]
+        
+        if len(class_name) == 0:
+            assert common_name in ('animal','vehicle','blank')
+            return
+        
+        if len(species) > 0:
+            assert all([len(s) > 0 for s in [genus,family,order]])
+            binomial_name = genus + ' ' + species
+            if binomial_name not in speciesnet_latin_name_to_taxon_string:
+                speciesnet_latin_name_to_taxon_string[binomial_name] = s
+        elif len(genus) > 0:
+            assert all([len(s) > 0 for s in [family,order]])
+            if genus not in speciesnet_latin_name_to_taxon_string:
+                speciesnet_latin_name_to_taxon_string[genus] = s
+        elif len(family) > 0:
+            assert len(order) > 0
+            if family not in speciesnet_latin_name_to_taxon_string:
+                speciesnet_latin_name_to_taxon_string[family] = s
+        elif len(order) > 0:
+            if order not in speciesnet_latin_name_to_taxon_string:
+                speciesnet_latin_name_to_taxon_string[order] = s
+        else:
+            if class_name not in speciesnet_latin_name_to_taxon_string:
+                speciesnet_latin_name_to_taxon_string[class_name] = s
+        
+        if len(common_name) > 0:
+            if common_name not in speciesnet_common_name_to_taxon_string:
+                speciesnet_common_name_to_taxon_string[common_name] = s
+    
+    for s in speciesnet_taxonomy_list:
+        
+        _insert_taxonomy_string(s)
+    
+    
+    ##%% Make sure all parent taxa are represented in the taxonomy
+    
+    # In theory any taxon that appears as the parent of another taxon should
+    # also be in the taxonomy, but this isn't always true, so we fix it here.
+    
+    new_taxon_string_to_missing_tokens = defaultdict(list)
+    
+    # latin_name = next(iter(speciesnet_latin_name_to_taxon_string.keys()))
+    for latin_name in speciesnet_latin_name_to_taxon_string.keys():
+        
+        if 'no cv result' in latin_name:
+            continue
+        
+        taxon_string = speciesnet_latin_name_to_taxon_string[latin_name]
+        tokens = taxon_string.split(';')
+        
+        # Don't process GUID, species, or common name
+        # i_token = 6
+        for i_token in range(1,len(tokens)-2):
+            
+            test_token = tokens[i_token]            
+            if len(test_token) == 0:
+                continue
+            
+            # Do we need to make up a taxon for this token?
+            if test_token not in speciesnet_latin_name_to_taxon_string:
+                
+                new_tokens = [''] * 7
+                new_tokens[0] = 'fake_guid'
+                for i_copy_token in range(1,i_token+1):
+                    new_tokens[i_copy_token] = tokens[i_copy_token]
+                new_tokens[-1] = test_token + ' species'
+                assert new_tokens[-2] == ''
+                new_taxon_string = ';'.join(new_tokens)
+                # assert new_taxon_string not in new_taxon_strings
+                new_taxon_string_to_missing_tokens[new_taxon_string].append(test_token)
+                
+        # ...for each token
+        
+    # ...for each taxon
+        
+    print('Found {} taxa that need to be inserted to make the taxonomy valid:\n'.format(
+        len(new_taxon_string_to_missing_tokens)))
+    
+    new_taxon_string_to_missing_tokens = \
+        sort_dictionary_by_key(new_taxon_string_to_missing_tokens)
+    for taxon_string in new_taxon_string_to_missing_tokens:
+        missing_taxa = ','.join(new_taxon_string_to_missing_tokens[taxon_string])
+        print('{} ({})'.format(taxon_string,missing_taxa))
+        
+    for new_taxon_string in new_taxon_string_to_missing_tokens:
+        _insert_taxonomy_string(new_taxon_string)
+        
+        
+    ##%% Make sure all species on the allow-list are in the taxonomy
+    
+    n_failed_mappings = 0
+    
+    for target_taxon_latin_name in target_latin_to_common.keys():
+        if target_taxon_latin_name not in speciesnet_latin_name_to_taxon_string:
+            common_name = target_latin_to_common[target_taxon_latin_name]
+            s = '{} ({}) not in speciesnet taxonomy'.format(
+                target_taxon_latin_name,common_name)
+            if common_name in speciesnet_common_name_to_taxon_string:
+                s += ' (common name maps to {})'.format(
+                    speciesnet_common_name_to_taxon_string[common_name])
+            print(s)
+            n_failed_mappings += 1
+    
+    if n_failed_mappings > 0:
+        raise ValueError('Cannot continue with geofence generation')
+    
+    
+    ##%% For the allow-list, map each parent taxon to a set of allowable child taxa
+    
+    # Maps parent names to all allowed child names, or None if this is the 
+    # lowest-level allowable taxon on this path
+    allowed_parent_taxon_to_child_taxa = defaultdict(set)
+    
+    # latin_name = next(iter(target_latin_to_common.keys()))
+    for latin_name in target_latin_to_common:
+        
+        taxon_string = speciesnet_latin_name_to_taxon_string[latin_name]
+        tokens = taxon_string.split(';')
+        assert len(tokens) == 7
+        
+        # Remove GUID and common mame
+        #
+        # This is now always class/order/family/genus/species
+        tokens = tokens[1:-1]
+        
+        child_taxon = None
+        
+        # If this is a species
+        if len(tokens[-1]) > 0:
+            binomial_name = tokens[-2] + ' ' + tokens[-1]
+            assert binomial_name == latin_name
+            allowed_parent_taxon_to_child_taxa[binomial_name].add(None)
+            child_taxon = binomial_name
+            
+        # The first candidate parent is the genus        
+        parent_token_index = len(tokens) - 2
+
+        while(parent_token_index >= 0):
+            
+            parent_taxon = tokens[parent_token_index]
+            allowed_parent_taxon_to_child_taxa[parent_taxon].add(child_taxon)
+            child_taxon = parent_taxon
+            parent_token_index -= 1                
+    
+    # ...for each allowed latin name
+    
+    allowed_parent_taxon_to_child_taxa = \
+        sort_dictionary_by_key(allowed_parent_taxon_to_child_taxa)
+    
+    
+    ##%% Map all predictions that exist in this dataset...
+    
+    # ...to the prediction we should generate.
+    
+    with open(input_file,'r') as f:
+        input_data = json.load(f)
+    
+    input_category_id_to_common_name = input_data['classification_categories'] #noqa
+    input_category_id_to_taxonomy_string = \
+        input_data['classification_category_descriptions']
+    
+    input_category_id_to_output_taxon_string = {}
+    
+    # input_category_id = next(iter(input_category_id_to_taxonomy_string.keys()))
+    for input_category_id in input_category_id_to_taxonomy_string.keys():
+        
+        input_taxon_string = input_category_id_to_taxonomy_string[input_category_id]
+        input_taxon_tokens = input_taxon_string.split(';')
+        assert len(input_taxon_tokens) == 7
+    
+        # Don't mess with blank/no-cv-result/animal/human
+        if (input_taxon_string in non_taxonomic_prediction_strings) or \
+           (input_taxon_string == human_prediction_string):
+            input_category_id_to_output_taxon_string[input_category_id] = \
+                input_taxon_string
+            continue
+    
+        # Remove GUID and common mame
+        #
+        # This is now always class/order/family/genus/species
+        input_taxon_tokens = input_taxon_tokens[1:-1]
+        
+        test_index = len(input_taxon_tokens) - 1
+        target_taxon = None
+        
+        # Start at the species level, and see whether each taxon is allowed
+        while((test_index >= 0) and (target_taxon is None)):
+            
+            # Species are represented as binomial names
+            if (test_index == (len(input_taxon_tokens) - 1)) and \
+                (len(input_taxon_tokens[-1]) > 0):
+                test_taxon_name = \
+                    input_taxon_tokens[-2] + ' ' + input_taxon_tokens[-1]
+            else:
+                test_taxon_name = input_taxon_tokens[test_index]
+            
+            # If we haven't yet found the level at which this taxon is non-empty,
+            # keep going up
+            if len(test_taxon_name) == 0:            
+                test_index -= 1
+                continue
+            
+            assert test_taxon_name in speciesnet_latin_name_to_taxon_string
+            
+            # Is this taxon allowed according to the custom species list?
+            if test_taxon_name in allowed_parent_taxon_to_child_taxa:
+                
+                allowed_child_taxa = allowed_parent_taxon_to_child_taxa[test_taxon_name]
+                assert allowed_child_taxa is not None
+                
+                # If this is the lowest-level allowable token or there is not a 
+                # unique child, don't walk any further, even if walking down
+                # is enabled.
+                if (None in allowed_child_taxa):
+                    assert len(allowed_child_taxa) == 1
+                
+                if (None in allowed_child_taxa) or (len(allowed_child_taxa) > 1):
+                    target_taxon = test_taxon_name
+                elif not allow_walk_down:
+                    target_taxon = test_taxon_name
+                else:
+                    # If there's a unique child, walk back *down* the allowable
+                    # taxa until we run out of unique children
+                    while ((next(iter(allowed_child_taxa)) is not None) and \
+                          (len(allowed_child_taxa) == 1)):
+                        candidate_taxon = next(iter(allowed_child_taxa))
+                        assert candidate_taxon in allowed_parent_taxon_to_child_taxa
+                        assert candidate_taxon in speciesnet_latin_name_to_taxon_string
+                        allowed_child_taxa = \
+                            allowed_parent_taxon_to_child_taxa[candidate_taxon]
+                    target_taxon = candidate_taxon
+                    
+            # ...if this is an allowed taxon
+                    
+            test_index -= 1
+    
+        # ...for each token
+        
+        if target_taxon is None:
+            output_taxon_string = animal_prediction_string                        
+        else:
+            output_taxon_string = speciesnet_latin_name_to_taxon_string[target_taxon]
+        input_category_id_to_output_taxon_string[input_category_id] = output_taxon_string        
+    
+    # ...for each category
+        
+    
+    ##%% Build the new tables
+    
+    input_category_id_to_output_category_id = {}
+    output_taxon_string_to_category_id = {}
+    output_category_id_to_common_name = {}
+    
+    for input_category_id in input_category_id_to_output_taxon_string:
+        
+        original_common_name = \
+            input_category_id_to_common_name[input_category_id]
+        original_taxon_string = \
+            input_category_id_to_taxonomy_string[input_category_id]
+        output_taxon_string = \
+            input_category_id_to_output_taxon_string[input_category_id]
+    
+        output_common_name = output_taxon_string.split(';')[-1]
+        
+        # Do we need to create a new output category?
+        if output_taxon_string not in output_taxon_string_to_category_id:
+            output_category_id = str(len(output_taxon_string_to_category_id))
+            output_taxon_string_to_category_id[output_taxon_string] = \
+                output_category_id
+            output_category_id_to_common_name[output_category_id] = \
+                output_common_name            
+        else:
+            output_category_id = \
+                output_taxon_string_to_category_id[output_taxon_string]
+        
+        input_category_id_to_output_category_id[input_category_id] = \
+            output_category_id
+            
+        if False:
+            print('Mapping {} ({}) to:\n{} ({})\n'.format(
+                original_common_name,original_taxon_string,
+                output_common_name,output_taxon_string))
+        if False:    
+            print('Mapping {} to {}'.format(
+                original_common_name,output_common_name,))
+    
+    # ...for each category
+    
+    
+    ##%% Remap all category labels
+    
+    assert len(set(output_taxon_string_to_category_id.keys())) == \
+           len(set(output_taxon_string_to_category_id.values()))
+           
+    output_category_id_to_taxon_string = \
+        invert_dictionary(output_taxon_string_to_category_id)
+        
+    with open(input_file,'r') as f:
+        output_data = json.load(f)
+        
+    for im in tqdm(output_data['images']):
+        if 'detections' in im and im['detections'] is not None:
+            for det in im['detections']:
+                if 'classifications' in det:
+                    for classification in det['classifications']:
+                        classification[0] = \
+                            input_category_id_to_output_category_id[classification[0]]
+    
+    output_data['classification_categories'] = output_category_id_to_common_name
+    output_data['classification_category_descriptions'] = \
+        output_category_id_to_taxon_string
+        
+        
+    ##%% Write output
+    
+    with open(output_file,'w') as f:
+        json.dump(output_data,f,indent=1)
+    
+    
+#%% Interactive driver(s)
+
+if False:
+    
+    pass
+
+    #%% Shared cell to initialize geofencing and taxonomy information
+    
+    from megadetector.utils.wi_utils import species_allowed_in_country # noqa
+    from megadetector.utils.wi_utils import initialize_geofencing, initialize_taxonomy_info # noqa
+    from megadetector.utils.wi_utils import _species_string_to_canonical_species_string # noqa
+    from megadetector.utils.wi_utils import generate_csv_rows_for_species # noqa
+    from megadetector.utils.wi_utils import _generate_csv_rows_to_block_all_countries_except # noqa
+
+    from megadetector.utils.wi_utils import taxonomy_string_to_geofencing_rules # noqa
+    from megadetector.utils.wi_utils import taxonomy_string_to_taxonomy_info # noqa
+    
+    geofencing_file = r'c:\git\cameratrapai\data\geofence_base.json'
+    
+    country_code_file = r'g:\temp\country-codes.csv'
+    # encoding = 'cp1252'; taxonomy_file = r'g:\temp\taxonomy_mapping-' + encoding + '.json'    
+    encoding = None; taxonomy_file = r'g:\temp\taxonomy_mapping.json'
+    
+    initialize_geofencing(geofencing_file, country_code_file, force_init=True)
+    initialize_taxonomy_info(taxonomy_file, force_init=True, encoding=encoding)
+    
+    
+    #%% Test driver for geofence_fixes.csv function
+    
+    block_except_list = 'AUS, PNG, THA, IDN, MYS'
+    species = 'dingo'
+    species_string = _species_string_to_canonical_species_string(species)
+    rows = _generate_csv_rows_to_block_all_countries_except(species_string,block_except_list)
+        
+    import clipboard; clipboard.copy('\n'.join(rows))
+    
+    
+    #%%
+    
+    generate_csv_rows_for_species(species_string=species_string,
+                                  allow_countries=None,
+                                  block_countries=None,
+                                  allow_states=None,
+                                  block_states=None,
+                                  blockexcept_countries=None)
+    
+    
+    _generate_csv_rows_to_block_all_countries_except(species_string,'AUS')
+    
+    
+    #%% Test the effects of geofence changes
+
+    species = 'canis lupus dingo'
+    country = 'guatemala'
+    species_allowed_in_country(species,country,state=None,return_status=False)
+        
+    
+    #%% instances.json generation test
+    
+    from megadetector.utils.wi_utils import generate_instances_json_from_folder # noqa
+    
+    instances_file = r'g:\temp\water-hole\instances.json'
+    
+    _ = generate_instances_json_from_folder(folder=r'g:\temp\water-hole',
+                                            country='NAM',
+                                            lat=None,
+                                            lon=None,
+                                            output_file=instances_file,
+                                            filename_replacements={'g:/temp':'/mnt/g/temp'})
+
+    # from megadetector.utils.path_utils import open_file; open_file(instances_file)
+
+
+    #%% MD --> prediction conversion test
+    
+    from megadetector.utils.wi_utils import generate_predictions_json_from_md_results # noqa
+    md_results_file = r'G:\temp\md-test-images\mdv5a.relpaths.json'
+    predictions_json_file = r'\\wsl$\Ubuntu\home\dmorris\tmp\speciesnet-tests\mdv5a.abspaths.predictions-format.json'
+    generate_predictions_json_from_md_results(md_results_file,predictions_json_file,base_folder=
+                                              '/home/dmorris/tmp/md-test-images/')
+    
+    from megadetector.utils.wi_utils import generate_predictions_json_from_md_results # noqa
+    md_results_file = r"G:\temp\water-hole\md_results.json"
+    predictions_json_file = r"G:\temp\water-hole\md_results-prediction_format.json"
+    generate_predictions_json_from_md_results(md_results_file,predictions_json_file,base_folder=
+                                              '/mnt/g/temp/water-hole')    
+    
+    
+    #%% Geofencing tests
+    
+    species = 'didelphis marsupialis'
+    print(binomial_name_to_taxonomy_info[species])
+    country = 'Guatemala'
+    assert species_allowed_in_country(species, country)
+    
+    species = 'virginia opossum'
+    print(common_name_to_taxonomy_info[species])
+    country = 'USA'
+    assert species_allowed_in_country(species, country)
+    
+    
+    #%% Test several species
+    
+    if True:
+        
+        # Make sure some Guatemalan species are allowed in Guatemala
+        all_species = [
+            'didelphis marsupialis',
+            'didelphis virginiana',
+            'dasypus novemcinctus',
+            'urocyon cinereoargenteus',
+            'nasua narica',
+            'eira barbara',
+            'conepatus semistriatus',
+            'leopardus wiedii',
+            'leopardus pardalis',
+            'puma concolor',
+            'panthera onca',
+            'tapirus bairdii',
+            'pecari tajacu',
+            'tayassu pecari',
+            'mazama temama',
+            'mazama pandora',
+            'odocoileus virginianus',
+            'dasyprocta punctata',
+            'tinamus major',
+            'crax rubra',
+            'meleagris ocellata',
+            'gulo gulo' # Consistency check; this species should be blocked
+            ]
+        
+        country ='guatemala'
+        state = None
+    
+    if True:
+        
+        # Make sure some PNW species are allowed in the right states
+        all_species = \
+            ['Taxidea taxus',
+            'Martes americana',
+            'Ovis canadensis',
+            'Ursus americanus',
+            'Lynx rufus',
+            'Lynx canadensis',
+            'Puma concolor',
+            'Canis latrans',
+            'Cervus canadensis',
+            'Canis lupus',
+            'Ursus arctos',
+            'Marmota caligata',
+            'Alces alces',
+            'Oreamnos americanus',
+            'Odocoileus hemionus',
+            'Vulpes vulpes',
+            'Lepus americanus',
+            'Mephitis mephitis',
+            'Odocoileus virginianus',
+            'Marmota flaviventris',
+            'tapirus bairdii' # Consistency check; this species should be blocked
+            ]
+        
+        all_species = [s.lower() for s in all_species]
+        
+        country = 'USA'
+        state = 'WA'
+        # state = 'MT'
+    
+    if True:
+        
+        all_species = ['ammospermophilus harrisii']
+        country = 'USA'
+        state = 'CA'
+        
+    for species in all_species:
+        
+        taxonomy_info = binomial_name_to_taxonomy_info[species]
+        allowed = species_allowed_in_country(species, country, state=state, return_status=True)
+        state_string = ''
+        if state is not None:
+            state_string = ' ({})'.format(state)
+        print('{} ({}) for {}{}: {}'.format(taxonomy_info['common_name'],species,country,state_string,allowed))
+            
+        
+    #%% Test conversion from predictons.json to MD format
+    
+    import os # noqa
+    from megadetector.utils.wi_utils import generate_md_results_from_predictions_json # noqa
+    
+    # detector_source = 'speciesnet'
+    detector_source = 'md'
+    
+    if False:
+        image_folder = r'g:\temp\md-test-images'
+        base_folder = '/home/dmorris/tmp/md-test-images/'
+        if detector_source == 'speciesnet':    
+            predictions_json_file = r"\\wsl$\Ubuntu\home\dmorris\tmp\speciesnet-tests\ensemble-output.json"
+            md_results_file = r"\\wsl$\Ubuntu\home\dmorris\tmp\speciesnet-tests\ensemble-output-md-format.json"
+        else:
+            assert detector_source == 'md'
+            predictions_json_file = r"\\wsl$\Ubuntu\home\dmorris\tmp\speciesnet-tests\ensemble-output-from-md-results.json"
+            md_results_file = r"\\wsl$\Ubuntu\home\dmorris\tmp\speciesnet-tests\ensemble-output-md-format-from-md-results.json"        
+    else:
+        image_folder = r'g:\temp\water-hole'
+        base_folder = '/mnt/g/temp/water-hole/'
+        if detector_source == 'speciesnet':    
+            predictions_json_file = r'g:\temp\water-hole\ensemble-output.json'
+            md_results_file = r'g:\temp\water-hole\ensemble-output.md_format.json'
+        else:
+            assert detector_source == 'md'
+            predictions_json_file = r'g:\temp\water-hole\ensemble-output-md.json'
+            md_results_file = r'g:\temp\water-hole\ensemble-output-md.md_format.json'
+    
+    generate_md_results_from_predictions_json(predictions_json_file=predictions_json_file,
+                                              md_results_file=md_results_file,
+                                              base_folder=base_folder)
+    
+    # from megadetector.utils.path_utils import open_file; open_file(md_results_file)
+    
+    assert os.path.isdir(image_folder)
+    
+   
+    #%% Preview
+    
+    from megadetector.postprocessing.postprocess_batch_results import \
+        PostProcessingOptions, process_batch_results
+    from megadetector.utils import path_utils
+    
+    render_animals_only = False
+    
+    options = PostProcessingOptions()
+    options.image_base_dir = image_folder
+    options.include_almost_detections = True
+    options.num_images_to_sample = None
+    options.confidence_threshold = 0.2
+    options.almost_detection_confidence_threshold = options.confidence_threshold - 0.05
+    options.ground_truth_json_file = None
+    options.separate_detections_by_category = True
+    options.sample_seed = 0
+    options.max_figures_per_html_file = 5000
+    
+    options.parallelize_rendering = True
+    options.parallelize_rendering_n_cores = 10
+    options.parallelize_rendering_with_threads = True
+    options.sort_classification_results_by_count = True
+    
+    if render_animals_only:
+        # Omit some pages from the output, useful when animals are rare
+        options.rendering_bypass_sets = ['detections_person','detections_vehicle',
+                                          'detections_person_vehicle','non_detections']    
+    
+    output_base = r'g:\temp\preview' + '_' + detector_source
+    if render_animals_only:
+        output_base = output_base + '_render_animals_only'
+    os.makedirs(output_base, exist_ok=True)
+    
+    print('Writing preview to {}'.format(output_base))
+    
+    options.md_results_file = md_results_file
+    options.output_dir = output_base
+    ppresults = process_batch_results(options)
+    html_output_file = ppresults.output_html_file
+    
+    path_utils.open_file(html_output_file,attempt_to_open_in_wsl_host=True,browser_name='chrome')
+    # import clipboard; clipboard.copy(html_output_file)