megadetector 10.0.13__py3-none-any.whl

megadetector/taxonomy_mapping/prepare_lila_taxonomy_release.py

@@ -0,0 +1,165 @@
+"""
+
+prepare_lila_taxonomy_release.py
+
+Given the private intermediate taxonomy mapping (produced by map_new_lila_datasets.py),
+prepare the public (release) taxonomy mapping file.
+
+"""
+
+#%% Imports and constants
+
+import os
+import json
+import pandas as pd
+
+
+#%% Prevent execution during infrastructural imports
+
+if False:
+
+    #%% Filenames
+
+    lila_taxonomy_file = 'c:/git/agentmorrisprivate/lila-taxonomy/lila-taxonomy-mapping.csv'
+    release_taxonomy_file = os.path.expanduser('~/lila/lila-taxonomy-mapping_release.csv')
+    # import clipboard; clipboard.copy(release_taxonomy_file)
+
+    # Created by get_lila_annotation_counts.py... contains counts for each category
+    lila_dataset_to_categories_file = os.path.expanduser('~/lila/lila_categories_list/lila_dataset_to_categories.json')
+
+    assert os.path.isfile(lila_dataset_to_categories_file)
+    assert os.path.isfile(lila_taxonomy_file)
+
+
+    #%% Find out which categories are actually used
+
+    df = pd.read_csv(lila_taxonomy_file)
+
+    with open(lila_dataset_to_categories_file,'r') as f:
+        lila_dataset_to_categories = json.load(f)
+
+    used_category_mappings = []
+
+    # dataset_name = datasets_to_map[0]
+    for dataset_name in lila_dataset_to_categories.keys():
+
+        ds_categories = lila_dataset_to_categories[dataset_name]
+        for category in ds_categories:
+            category_name = category['name'].lower()
+            assert ':' not in category_name
+            mapping_name = dataset_name + ':' + category_name
+            used_category_mappings.append(mapping_name)
+
+    df['used'] = False
+
+    n_dropped = 0
+
+    # i_row = 0; row = df.iloc[i_row]; row
+    for i_row,row in df.iterrows():
+        ds_name = row['dataset_name']
+        query = row['query']
+        mapping_name = ds_name + ':' + query
+        if mapping_name in used_category_mappings:
+            df.loc[i_row,'used'] = True
+        else:
+            n_dropped += 1
+            print('Dropping unused mapping {}'.format(mapping_name))
+
+    print('Dropping {} of {} mappings'.format(n_dropped,len(df)))
+
+    df = df[df.used]
+    df = df.drop('used',axis=1)
+
+
+    #%% Generate the final output file
+
+    assert not os.path.isfile(release_taxonomy_file), \
+        'File {} exists, delete it manually before proceeding'.format(release_taxonomy_file)
+
+    levels_to_include = ['kingdom',
+                         'phylum',
+                         'subphylum',
+                         'superclass',
+                         'class',
+                         'subclass',
+                         'infraclass',
+                         'superorder',
+                         'order',
+                         'suborder',
+                         'infraorder',
+                         'superfamily',
+                         'family',
+                         'subfamily',
+                         'tribe',
+                         'genus',
+                         'subgenus',
+                         'species',
+                         'subspecies',
+                         'variety']
+
+    levels_to_exclude = ['stateofmatter',
+                         'zoosection',
+                         'parvorder',
+                         'complex',
+                         'epifamily']
+
+    for x in [levels_to_include,levels_to_exclude]:
+        assert len(x) == len(set(x))
+
+    for s in levels_to_exclude:
+        assert s not in levels_to_include
+
+    known_levels = levels_to_include + levels_to_exclude
+
+    levels_used = set()
+
+    # i_row = 0; row = df.iloc[i_row]; row
+    for i_row,row in df.iterrows():
+
+        if not isinstance(row['scientific_name'],str):
+            assert not isinstance(row['taxonomy_string'],str)
+            continue
+
+        # This is a list of length-4 tuples that each look like:
+        #
+        # (41789, 'species', 'taxidea taxus', ['american badger'])
+        taxonomic_match = eval(row['taxonomy_string'])
+
+        # match_at_level = taxonomic_match[0]
+        for match_at_level in taxonomic_match:
+            assert len(match_at_level) == 4
+            # E.g. "species"
+            levels_used.add(match_at_level[1])
+
+    levels_used = [s for s in levels_used if isinstance(s,str)]
+
+    for s in levels_used:
+        assert s in known_levels, 'Unrecognized level {}'.format(s)
+
+    for s in levels_to_include:
+        assert s in levels_used
+
+    for s in levels_to_include:
+        df[s] = ''
+
+    # i_row = 0; row = df.iloc[i_row]; row
+    for i_row,row in df.iterrows():
+
+        if not isinstance(row['scientific_name'],str):
+            assert not isinstance(row['taxonomy_string'],str)
+            continue
+
+        # E.g.: (43117, 'genus', 'lepus', ['hares and jackrabbits']
+        taxonomic_match = eval(row['taxonomy_string'])
+
+        for match_at_level in taxonomic_match:
+            level = match_at_level[1]
+            if level in levels_to_include:
+                df.loc[i_row,level] = match_at_level[2]
+
+    df = df.drop('source',axis=1)
+    df.to_csv(release_taxonomy_file,header=True,index=False)
+
+    print('Wrote final output to {}'.format(release_taxonomy_file))
+
+

megadetector/taxonomy_mapping/preview_lila_taxonomy.py

@@ -0,0 +1,543 @@
+"""
+
+preview_lila_taxonomy.py
+
+Does some consistency-checking on the LILA taxonomy file, and generates
+an HTML preview page that we can use to determine whether the mappings
+make sense.
+
+"""
+
+#%% Imports and constants
+
+from tqdm import tqdm
+
+import os
+import pandas as pd
+
+# lila_taxonomy_file = r"c:\git\agentmorrisprivate\lila-taxonomy\lila-taxonomy-mapping.csv"
+lila_taxonomy_file = os.path.expanduser('~/lila/lila_additions_2025.10.07.csv')
+
+preview_base = os.path.expanduser('~/lila/lila_taxonomy_preview')
+os.makedirs(preview_base,exist_ok=True)
+html_output_file = os.path.join(preview_base,'index.html')
+
+
+#%% Support functions
+
+def parse_taxonomy_string(taxonomy_string):
+
+    taxonomic_match = eval(taxonomy_string)
+    matched_entity = taxonomic_match[0]
+    assert len(matched_entity) == 4
+
+    level = matched_entity[1]
+
+    scientific_name = matched_entity[2]
+
+    common_names = matched_entity[3]
+    if len(common_names) == 1:
+        common_name = common_names[0]
+    else:
+        common_name = str(common_names)
+
+    return scientific_name,common_name,level,taxonomic_match
+
+def taxonomy_string_to_common_name(taxonomy_string):
+    _,cn,_,_ = parse_taxonomy_string(taxonomy_string)
+    return cn
+
+def taxonomy_string_to_scientific(taxonomy_string):
+    sn,_,_,_ = parse_taxonomy_string(taxonomy_string)
+    return sn
+
+def taxonomy_string_to_level(taxonomy_string):
+    _,_,level,_ = parse_taxonomy_string(taxonomy_string)
+    return level
+
+
+#%% Prepare taxonomy lookup
+
+from megadetector.taxonomy_mapping.species_lookup import \
+    initialize_taxonomy_lookup, get_preferred_taxonomic_match
+
+initialize_taxonomy_lookup()
+
+
+#%% Check for mappings that disagree with the taxonomy string
+
+# For example, cases where the "level" column says "species", but the taxonomy string says it's a genus.
+
+df = pd.read_csv(lila_taxonomy_file)
+
+n_taxonomy_changes = 0
+
+# Look for internal inconsistency
+for i_row,row in df.iterrows():
+
+    sn = row['scientific_name']
+    if not isinstance(sn,str):
+        continue
+
+    ts = row['taxonomy_string']
+    assert sn == taxonomy_string_to_scientific(ts)
+
+    assert row['taxonomy_level'] == taxonomy_string_to_level(ts)
+
+# Look for outdated mappings
+taxonomy_preference = 'inat'
+
+# i_row = 0; row = df.iloc[i_row]
+for i_row,row in tqdm(df.iterrows(),total=len(df)):
+
+    try:
+
+        sn = row['scientific_name']
+        if not isinstance(sn,str):
+            continue
+
+        m = get_preferred_taxonomic_match(sn,taxonomy_preference)
+        assert m.scientific_name == sn
+
+        ts = row['taxonomy_string']
+        assert m.taxonomy_string[0:50] == ts[0:50], 'Mismatch for {}:\n\n{}\n\n{}\n'.format(
+            row['dataset_name'],ts,m.taxonomy_string)
+
+        if ts != m.taxonomy_string:
+            n_taxonomy_changes += 1
+            df.loc[i_row,'taxonomy_string'] = m.taxonomy_string
+
+    except Exception as e:
+
+        print('Error at row {}: {}'.format(i_row,str(e)))
+        raise
+
+# ...for each row
+
+print('\nMade {} taxonomy changes'.format(n_taxonomy_changes))
+
+# Optionally re-write
+if False:
+    df.to_csv(lila_taxonomy_file,header=True,index=False)
+
+
+#%% List null mappings
+
+# These should all be things like "empty", "unidentified", "fire", "car", etc.
+
+# i_row = 0; row = df.iloc[i_row]
+for i_row,row in df.iterrows():
+    if (not isinstance(row['taxonomy_string'],str)) or (len(row['taxonomy_string']) == 0):
+        print('No mapping for {}:{}'.format(row['dataset_name'],row['query']))
+
+
+#%% List mappings with scientific names but no common names
+
+for i_row,row in df.iterrows():
+    cn = row['common_name']
+    sn = row['scientific_name']
+    ts = row['taxonomy_string']
+    if (isinstance(ts,str)) and (len(ts) >= 0):
+        if (not isinstance(cn,str)) or (len(cn) == 0):
+            print('No mapping for {}:{}:{}'.format(row['dataset_name'],row['query'],row['scientific_name']))
+
+
+#%% List mappings that map to different things in different data sets
+
+import numpy as np
+def isnan(x):
+    if not isinstance(x,float):
+        return False
+    return np.isnan(x)
+
+from collections import defaultdict
+query_to_rows = defaultdict(list)
+
+queries_with_multiple_mappings = set()
+
+n_suppressed = 0
+
+suppress_multiple_matches = [
+    ['porcupine','Snapshot Camdeboo','Idaho Camera Traps'],
+    ['porcupine','Snapshot Enonkishu','Idaho Camera Traps'],
+    ['porcupine','Snapshot Karoo','Idaho Camera Traps'],
+    ['porcupine','Snapshot Kgalagadi','Idaho Camera Traps'],
+    ['porcupine','Snapshot Kruger','Idaho Camera Traps'],
+    ['porcupine','Snapshot Mountain Zebra','Idaho Camera Traps'],
+    ['porcupine','Snapshot Serengeti','Idaho Camera Traps'],
+
+    ['porcupine','Snapshot Serengeti','Snapshot Mountain Zebra'],
+    ['porcupine','Snapshot Serengeti','Snapshot Kruger'],
+    ['porcupine','Snapshot Serengeti','Snapshot Kgalagadi'],
+    ['porcupine','Snapshot Serengeti','Snapshot Karoo'],
+    ['porcupine','Snapshot Serengeti','Snapshot Camdeboo'],
+
+    ['porcupine','Snapshot Enonkishu','Snapshot Camdeboo'],
+    ['porcupine','Snapshot Enonkishu','Snapshot Mountain Zebra'],
+    ['porcupine','Snapshot Enonkishu','Snapshot Kruger'],
+    ['porcupine','Snapshot Enonkishu','Snapshot Kgalagadi'],
+    ['porcupine','Snapshot Enonkishu','Snapshot Karoo'],
+
+    ['kudu','Snapshot Serengeti','Snapshot Mountain Zebra'],
+    ['kudu','Snapshot Serengeti','Snapshot Kruger'],
+    ['kudu','Snapshot Serengeti','Snapshot Kgalagadi'],
+    ['kudu','Snapshot Serengeti','Snapshot Karoo'],
+    ['kudu','Snapshot Serengeti','Snapshot Camdeboo'],
+
+    ['fox','Caltech Camera Traps','Channel Islands Camera Traps'],
+    ['fox','Idaho Camera Traps','Channel Islands Camera Traps'],
+    ['fox','Idaho Camera Traps','Caltech Camera Traps'],
+
+    ['pangolin','Snapshot Serengeti','SWG Camera Traps'],
+
+    ['deer', 'Wellington Camera Traps', 'Idaho Camera Traps'],
+    ['deer', 'Wellington Camera Traps', 'Caltech Camera Traps'],
+
+    ['unknown cervid', 'WCS Camera Traps', 'Idaho Camera Traps']
+
+]
+
+for i_row,row in df.iterrows():
+
+    query = row['query']
+    taxonomy_string = row['taxonomy_string']
+
+    for previous_i_row in query_to_rows[query]:
+
+        previous_row = df.iloc[previous_i_row]
+        assert previous_row['query'] == query
+        query_match = False
+        if isnan(row['taxonomy_string']):
+            query_match = isnan(previous_row['taxonomy_string'])
+        elif isnan(previous_row['taxonomy_string']):
+            query_match = isnan(row['taxonomy_string'])
+        else:
+            query_match = previous_row['taxonomy_string'][0:10] == taxonomy_string[0:10]
+
+        if not query_match:
+
+            suppress = False
+
+            # x = suppress_multiple_matches[-1]
+            for x in suppress_multiple_matches:
+                if x[0] == query and \
+                    ( \
+                    (x[1] == row['dataset_name'] and x[2] == previous_row['dataset_name']) \
+                    or \
+                    (x[2] == row['dataset_name'] and x[1] == previous_row['dataset_name']) \
+                    ):
+                    suppress = True
+                    n_suppressed += 1
+                    break
+
+            if not suppress:
+                print('Query {} in {} and {}:\n\n{}\n\n{}\n'.format(
+                    query, row['dataset_name'], previous_row['dataset_name'],
+                    taxonomy_string, previous_row['taxonomy_string']))
+
+            queries_with_multiple_mappings.add(query)
+
+    # ...for each row where we saw this query
+
+    query_to_rows[query].append(i_row)
+
+# ...for each row
+
+print('Found {} queries with multiple mappings ({} occurrences suppressed)'.format(
+    len(queries_with_multiple_mappings),n_suppressed))
+
+
+#%% Verify that nothing "unidentified" maps to a species or subspecies
+
+# E.g., "unidentified skunk" should never map to a specific species of skunk
+
+allowable_unknown_species = [
+    'unknown_tayra' # AFAIK this is a unique species, I'm not sure what's implied here
+]
+
+unk_queries = ['skunk']
+for i_row,row in df.iterrows():
+
+    query = row['query']
+    level = row['taxonomy_level']
+
+    if not isinstance(level,str):
+        assert not isinstance(row['taxonomy_string'],str)
+        continue
+
+    if ( \
+        'unidentified' in query or \
+        ('unk' in query and ('skunk' not in query and 'chipmunk' not in query))\
+        ) \
+        and \
+        ('species' in level):
+
+        if query not in allowable_unknown_species:
+
+            print('Warning: query {}:{} maps to {} {}'.format(
+                row['dataset_name'],
+                row['query'],
+                row['taxonomy_level'],
+                row['scientific_name']
+                ))
+
+
+#%% Make sure there are valid source and level values for everything with a mapping
+
+for i_row,row in df.iterrows():
+    if isinstance(row['scientific_name'],str):
+        if 'source' in row:
+            assert isinstance(row['source'],str)
+        assert isinstance(row['taxonomy_level'],str)
+
+
+#%% Find WCS mappings that aren't species or aren't the same as the input
+
+# WCS used scientific names, so these remappings are slightly more controversial
+# then the standard remappings.
+
+# row = df.iloc[-500]
+for i_row,row in df.iterrows():
+
+    if not isinstance(row['scientific_name'],str):
+        continue
+    if 'WCS' not in row['dataset_name']:
+        continue
+
+    query = row['query']
+    scientific_name = row['scientific_name']
+    common_name = row['common_name']
+    level = row['taxonomy_level']
+    taxonomy_string = row['taxonomy_string']
+
+    common_name_from_taxonomy = taxonomy_string_to_common_name(taxonomy_string)
+    query_string = query.replace(' sp','')
+    query_string = query_string.replace('unknown ','')
+
+    # Anything marked "species" or "unknown" by definition doesn't map to a species,
+    # so ignore these.
+    if (' sp' not in query) and ('unknown' not in query) and \
+        (level != 'species') and (level != 'subspecies'):
+        print('WCS query {} ({}) remapped to {} {} ({})'.format(
+            query,common_name,level,scientific_name,common_name_from_taxonomy))
+
+    if query_string != scientific_name:
+        pass
+        # print('WCS query {} ({}) remapped to {} ({})'.format(
+        #     query,common_name,scientific_name,common_names_from_taxonomy))
+
+
+#%% Download sample images for all scientific names
+
+# You might have to do this:
+#
+# pip install python-magic
+# pip install python-magic-bin
+
+# Takes ~1 minute per 10 rows
+
+remapped_queries = {'papio':'papio+baboon',
+                    'damaliscus lunatus jimela':'damaliscus lunatus',
+                    'mazama':'genus+mazama',
+                    'mirafra':'genus+mirafra'}
+
+import os
+from megadetector.taxonomy_mapping import retrieve_sample_image
+
+scientific_name_to_paths = {}
+image_base = os.path.join(preview_base,'images')
+images_per_query = 15
+min_valid_images_per_query = 3
+min_valid_image_size = 3000
+
+# TODO: parallelize this loop
+#
+# i_row = 0; row = df.iloc[i_row]
+for i_row,row in df.iterrows():
+
+    s = row['scientific_name']
+
+    if (not isinstance(s,str)) or (len(s)==0):
+        continue
+
+    query = s.replace(' ','+')
+
+    if query in remapped_queries:
+        query = remapped_queries[query]
+
+    query_folder = os.path.join(image_base,query)
+    os.makedirs(query_folder,exist_ok=True)
+
+    # Check whether we already have enough images for this query
+    image_files = os.listdir(query_folder)
+    image_fullpaths = [os.path.join(query_folder,fn) for fn in image_files]
+    sizes = [os.path.getsize(p) for p in image_fullpaths]
+    sizes_above_threshold = [x for x in sizes if x > min_valid_image_size]
+    if len(sizes_above_threshold) > min_valid_images_per_query:
+        print('Skipping query {}, already have {} images'.format(s,len(sizes_above_threshold)))
+        continue
+
+    # Check whether we've already run this query for a previous row
+    if query in scientific_name_to_paths:
+        continue
+
+    print('Processing query {} of {} ({})'.format(i_row,len(df),query))
+    paths = retrieve_sample_image.download_images(query=query,
+                                             output_directory=image_base,
+                                             limit=images_per_query,
+                                             verbose=True)
+    print('Downloaded {} images for {}'.format(len(paths),query))
+    scientific_name_to_paths[query] = paths
+
+# ...for each row in the mapping table
+
+
+#%% Rename .jpeg to .jpg
+
+from megadetector.utils import path_utils
+all_images = path_utils.recursive_file_list(image_base,False)
+
+for fn in tqdm(all_images):
+    if fn.lower().endswith('.jpeg'):
+        new_fn = fn[0:-5] + '.jpg'
+        os.rename(fn, new_fn)
+
+
+#%% Choose representative images for each scientific name
+
+# Specifically, sort by size, and take the largest unique sizes. Very small files tend
+# to be bogus thumbnails, etc.
+
+max_images_per_query = 4
+scientific_name_to_preferred_images = {}
+
+# s = list(scientific_name_to_paths.keys())[0]
+for s in list(df.scientific_name):
+
+    if not isinstance(s,str):
+        continue
+
+    query = s.replace(' ','+')
+
+    if query in remapped_queries:
+        query = remapped_queries[query]
+
+    query_folder = os.path.join(image_base,query)
+    assert os.path.isdir(query_folder)
+    image_files = os.listdir(query_folder)
+    image_fullpaths = [os.path.join(query_folder,fn) for fn in image_files]
+    sizes = [os.path.getsize(p) for p in image_fullpaths]
+    path_to_size = {}
+    for i_fp,fp in enumerate(image_fullpaths):
+        path_to_size[fp] = sizes[i_fp]
+    paths_by_size = [x for _, x in sorted(zip(sizes, image_fullpaths),reverse=True)]
+
+    # Be suspicious of duplicate sizes
+    b_duplicate_sizes = [False] * len(paths_by_size)
+
+    for i_path,p in enumerate(paths_by_size):
+        if i_path == len(paths_by_size) - 1:
+            continue
+        if path_to_size[p] == path_to_size[paths_by_size[i_path+1]]:
+            b_duplicate_sizes[i_path] = True
+
+    paths_by_size_non_dup = [i for (i, v) in zip(paths_by_size, b_duplicate_sizes) if not v]
+
+    preferred_paths = paths_by_size_non_dup[:max_images_per_query]
+    scientific_name_to_preferred_images[s] = preferred_paths
+
+# ...for each scientific name
+
+
+#%% Delete unused images
+
+used_images = []
+for images in scientific_name_to_preferred_images.values():
+    used_images.extend(images)
+
+print('Using a total of {} images'.format(len(used_images)))
+used_images_set = set(used_images)
+
+from megadetector.utils import path_utils
+all_images = path_utils.recursive_file_list(image_base,False)
+
+unused_images = []
+for fn in all_images:
+    if fn not in used_images_set:
+        unused_images.append(fn)
+
+print('{} of {} files unused (diff {})'.format(len(unused_images),len(all_images),
+                                               len(all_images) - len(unused_images)))
+
+for fn in tqdm(unused_images):
+    os.remove(fn)
+
+
+#%% Produce HTML preview
+
+with open(html_output_file, 'w', encoding='utf-8') as f:
+
+    f.write('<html><head></head><body>\n')
+
+    names = scientific_name_to_preferred_images.keys()
+    names = sorted(names)
+
+    f.write('<p class="speciesinfo_p" style="font-weight:bold;font-size:130%">'
+            'dataset_name: <b><u>category</u></b> mapped to taxonomy_level scientific_name (taxonomic_common_name) (manual_common_name)</p>\n'
+            '</p>')
+
+    # i_row = 2; row = df.iloc[i_row]
+    for i_row, row in tqdm(df.iterrows(), total=len(df)):
+
+        s = row['scientific_name']
+
+        taxonomy_string = row['taxonomy_string']
+        if isinstance(taxonomy_string,str):
+            taxonomic_match = eval(taxonomy_string)
+            matched_entity = taxonomic_match[0]
+            assert len(matched_entity) == 4
+            common_names = matched_entity[3]
+            if len(common_names) == 1:
+                common_name_string = common_names[0]
+            else:
+                common_name_string = str(common_names)
+        else:
+            common_name_string = ''
+
+        f.write('<p class="speciesinfo_p" style="font-weight:bold;font-size:130%">')
+
+        if isinstance(row.scientific_name,str):
+            output_string = '{}: <b><u>{}</u></b> mapped to {} {} ({}) ({})</p>\n'.format(
+                row.dataset_name, row.query,
+                row.taxonomy_level, row.scientific_name, common_name_string,
+                row.common_name)
+            f.write(output_string)
+        else:
+            f.write('{}: <b><u>{}</u></b> unmapped'.format(row.dataset_name,row.query))
+
+        if s is None or s not in names:
+            f.write('<p class="content_p">no images available</p>')
+        else:
+            image_paths = scientific_name_to_preferred_images[s]
+            basedir = os.path.dirname(html_output_file)
+            relative_paths = [os.path.relpath(p,basedir) for p in image_paths]
+            image_paths = [s.replace('\\','/') for s in relative_paths]
+            n_images = len(image_paths)
+            # image_paths = [os.path.relpath(p, output_base) for p in image_paths]
+            image_width_percent = round(100 / n_images)
+            f.write('<table class="image_table"><tr>\n')
+            for image_path in image_paths:
+                f.write('<td style="vertical-align:top;" width="{}%">'
+                        '<img src="{}" style="display:block; width:100%; vertical-align:top; height:auto;">'
+                        '</td>\n'.format(image_width_percent, image_path))
+            f.write('</tr></table>\n')
+
+    # ...for each row
+
+    f.write('</body></html>\n')
+
+
+#%% Open HTML preview
+
+from megadetector.utils.path_utils import open_file
+open_file(html_output_file)