datamule 0.415__cp39-cp39-macosx_10_9_universal2.whl

datamule/dataset_builder/dataset_builder.py

@@ -0,0 +1,259 @@
+import pandas as pd
+import json
+import os
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from tqdm import tqdm
+import google.generativeai as genai
+import time
+import psutil
+from threading import Lock
+
+class RateLimiter:
+    def __init__(self, max_rpm):
+        self.min_delay = 62.0 / max_rpm # 58 seconds to allow for some buffer WIP
+        self.last_request = time.time()
+        self.lock = Lock()
+        self.request_count = 0
+    
+    def acquire(self):
+        with self.lock:
+            now = time.time()
+            time_since_last = now - self.last_request
+            delay_needed = self.min_delay - time_since_last
+            self.last_request = now + max(0, delay_needed)  # Update based on expected completion
+            self.request_count += 1
+            count = self.request_count
+        
+        # Sleep outside the lock
+        if delay_needed > 0:
+            time.sleep(delay_needed)
+        
+        return count
+
+class DatasetBuilder:
+    def __init__(self):
+        self.base_prompt = None
+        self.response_schema = None
+        self.input_path = None
+        self.output_path = None
+        self.failed_path = None
+        self.max_rpm = 1450
+        self.max_workers = 30
+        self.save_frequency = 100
+        self.output_columns = None
+        self.buffer = []
+        self.buffer_lock = Lock()
+        self.failed_ids = set()
+        self.failed_lock = Lock()
+        self.model_name = "gemini-1.5-flash-8b"  # Default model
+        self.model_config = {}  # Additional model configuration
+        self.api_key = None
+
+    def set_api_key(self, api_key):
+        """Set the API key for Google's Generative AI."""
+        self.api_key = api_key
+        genai.configure(api_key=api_key)
+        return self
+
+    def set_paths(self, input_path, output_path, failed_path):
+        """Set input and output file paths."""
+        self.input_path = input_path
+        self.output_path = output_path
+        self.failed_path = failed_path
+        return self
+
+    def set_base_prompt(self, prompt):
+        """Set the base prompt for LLM processing."""
+        self.base_prompt = prompt
+        return self
+
+    def set_response_schema(self, schema):
+        """Set the response schema and derive output columns."""
+        self.response_schema = schema
+        # Derive output columns from schema
+        if schema and 'items' in schema and 'properties' in schema['items']:
+            properties = schema['items']['properties']
+            self.output_columns = ['accession_number'] + list(properties.keys())
+        return self
+
+    def set_rpm(self, max_rpm=1450):
+        """Set the maximum requests per minute."""
+        self.max_rpm = max_rpm
+        return self
+
+    def set_max_workers(self, max_workers=30):
+        """Set the maximum number of concurrent workers."""
+        self.max_workers = max_workers
+        return self
+
+    def set_save_frequency(self, frequency=100):
+        """Set how often to save progress."""
+        self.save_frequency = frequency
+        return self
+
+    def set_model(self, model_name="gemini-1.5-flash-8b", **model_config):
+        """Set the model name and configuration."""
+        self.model_name = model_name
+        self.model_config = model_config
+        return self
+
+    def validate_config(self):
+        """Validate that all required configurations are set."""
+        if not all([self.base_prompt, self.response_schema, self.input_path, 
+                   self.output_path, self.failed_path, self.api_key]):
+            raise ValueError("""Missing required configuration. Please ensure you have set:
+                           - API key
+                           - Paths (input_path, output_path, failed_path)
+                           - Base prompt
+                           - Response schema""")
+
+    def get_processed_ids(self):
+        """Get set of processed accession numbers from output file."""
+        if not os.path.exists(self.output_path):
+            return set()
+        
+        try:
+            # Read only the accession_number column for memory efficiency
+            df = pd.read_csv(self.output_path, usecols=['accession_number'])
+            return set(df['accession_number'])
+        except Exception as e:
+            print(f"Warning: Error reading processed IDs: {e}")
+            return set()
+
+    def save_data(self, df_new):
+        """Append new data to existing CSV."""
+        df_new.to_csv(self.output_path, mode='a', header=not os.path.exists(self.output_path), index=False)
+
+    def save_failed_ids(self):
+        """Save failed accession numbers to file."""
+        with open(self.failed_path, 'w') as f:
+            for acc in self.failed_ids:
+                f.write(f"{acc}\n")
+
+    def process_text(self, args):
+        """Process a single text entry through the model."""
+        model, text, accession_number, rate_limiter = args
+        
+        current_requests = rate_limiter.acquire()
+        
+        full_prompt = self.base_prompt + "\n\nINFORMATION:\n" + text
+
+        try:
+            generation_config = genai.GenerationConfig(
+                response_mime_type="application/json",
+                response_schema=self.response_schema,
+                **self.model_config
+            )
+            
+            response = model.generate_content(
+                full_prompt,
+                generation_config=generation_config
+            )
+            results = json.loads(response.text)
+            
+            for result in results:
+                result['accession_number'] = accession_number
+            
+            with self.buffer_lock:
+                self.buffer.extend(results)
+            
+            return True, current_requests
+        except Exception as e:
+            with self.failed_lock:
+                self.failed_ids.add(accession_number)
+            return False, f"Error processing {accession_number}: {str(e)}"
+
+    def build(self):
+        """Main processing method to build the dataset."""
+        self.validate_config()
+
+        # Initialize model and rate limiter
+        model = genai.GenerativeModel(self.model_name)
+        rate_limiter = RateLimiter(self.max_rpm)
+        
+        # Load data
+        print("Loading data...")
+        df_input = pd.read_csv(self.input_path)
+        processed_ids = self.get_processed_ids()
+        df_to_process = df_input[~df_input['accession_number'].isin(processed_ids)]
+        
+        total_in_dataset = len(df_input)
+        already_processed = len(processed_ids)
+        to_process = len(df_to_process)
+        
+        print(f"Total entries in dataset: {total_in_dataset}")
+        print(f"Already processed: {already_processed}")
+        print(f"New entries to process: {to_process}")
+        
+        if len(df_to_process) == 0:
+            print("All entries already processed!")
+            return
+
+        work_items = [
+            (model, row['text'], row['accession_number'], rate_limiter) 
+            for _, row in df_to_process.iterrows()
+        ]
+        
+        start_time = time.time()
+        last_save_time = time.time()
+        processed_count = 0
+        
+        with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
+            futures = {executor.submit(self.process_text, item): item for item in work_items}
+            
+            with tqdm(total=total_in_dataset, initial=already_processed, desc="Processing entries") as pbar:
+                for future in as_completed(futures):
+                    success, result = future.result()
+                    
+                    if not success:
+                        print(f"\n{result}")
+                    
+                    processed_count += 1
+                    pbar.update(1)
+                    
+                    elapsed = time.time() - start_time
+                    rpm = processed_count / (elapsed / 60)
+                    memory_usage = psutil.Process(os.getpid()).memory_info().rss / 1024 / 1024
+                    
+                    current_progress = already_processed + processed_count
+                    pbar.set_description(
+                        f"Processed {current_progress}/{total_in_dataset} | {rpm:.0f} RPM | Mem: {memory_usage:.0f}MB"
+                    )
+                    
+                    # Save periodically using append
+                    if len(self.buffer) >= self.save_frequency:
+                        with self.buffer_lock:
+                            df_new = pd.DataFrame(self.buffer)
+                            self.buffer = []
+                        
+                        if not df_new.empty:
+                            self.save_data(df_new)
+                            last_save_time = time.time()
+                    
+                    # Save failed IDs periodically
+                    if self.failed_ids and time.time() - last_save_time > 300:
+                        self.save_failed_ids()
+                        last_save_time = time.time()
+        
+        # Save any remaining results
+        if self.buffer:
+            with self.buffer_lock:
+                df_new = pd.DataFrame(self.buffer)
+                self.buffer = []
+            
+            if not df_new.empty:
+                self.save_data(df_new)
+        
+        if self.failed_ids:
+            self.save_failed_ids()
+        
+        # Print final statistics
+        elapsed = time.time() - start_time
+        final_rpm = processed_count / (elapsed / 60)
+        
+        print(f"\nProcessing complete:")
+        print(f"Total processed in this run: {processed_count}")
+        print(f"Average speed: {final_rpm:.0f} RPM")
+        print(f"Failed entries: {len(self.failed_ids)}")
+        if self.failed_ids:
+            print(f"Failed entries saved to: {self.failed_path}")

datamule/document.py

@@ -0,0 +1,130 @@
+import json
+import csv
+from .parser.document_parsing.sec_parser import Parser
+from .helper import convert_to_dashed_accession
+
+# we need to modify parse filing to take option in memory
+
+parser = Parser()
+
+class Document:
+    def __init__(self, type, filename):
+        self.type = type
+        self.filename = filename
+
+        self.data = None
+
+    def parse(self):
+        self.data = parser.parse_filing(self.filename, self.type)
+        return self.data
+    
+    def write_json(self, output_filename=None):
+        if not self.data:
+            raise ValueError("No data to write. Parse filing first.")
+            
+        if output_filename is None:
+            output_filename = f"{self.filename.rsplit('.', 1)[0]}.json"
+            
+        with open(output_filename, 'w') as f:
+            json.dump(self.data, f, indent=2)
+
+    def write_csv(self, output_filename=None, accession_number=None):
+        if self.data is None:
+            raise ValueError("No data available. Please call parse_filing() first.")
+
+        if output_filename is None:
+            output_filename = f"{self.filename.rsplit('.', 1)[0]}.csv"
+
+        with open(output_filename, 'w', newline='') as csvfile:
+            if not self.data:
+                return output_filename
+
+            has_document = any('document' in item for item in self.data)
+            
+            if has_document and 'document' in self.data:
+                writer = csv.DictWriter(csvfile, ['section', 'text'], quoting=csv.QUOTE_ALL)
+                writer.writeheader()
+                flattened = self._flatten_dict(self.data['document'])
+                for section, text in flattened.items():
+                    writer.writerow({'section': section, 'text': text})
+            else:
+                fieldnames = list(self.data[0].keys())
+                if accession_number:
+                    fieldnames.append('Accession Number')
+                writer = csv.DictWriter(csvfile, fieldnames, quoting=csv.QUOTE_ALL)
+                writer.writeheader()
+                for row in self.data:
+                    if accession_number:
+                        row['Accession Number'] = convert_to_dashed_accession(accession_number)
+                    writer.writerow(row)
+
+        return output_filename
+    
+    def _document_to_section_text(self, document_data, parent_key=''):
+        items = []
+        
+        if isinstance(document_data, dict):
+            for key, value in document_data.items():
+                # Build the section name
+                section = f"{parent_key}_{key}" if parent_key else key
+                
+                # If the value is a dict, recurse
+                if isinstance(value, dict):
+                    items.extend(self._document_to_section_text(value, section))
+                # If it's a list, handle each item
+                elif isinstance(value, list):
+                    for i, item in enumerate(value):
+                        if isinstance(item, dict):
+                            items.extend(self._document_to_section_text(item, f"{section}_{i+1}"))
+                        else:
+                            items.append({
+                                'section': f"{section}_{i+1}",
+                                'text': str(item)
+                            })
+                # Base case - add the item
+                else:
+                    items.append({
+                        'section': section,
+                        'text': str(value)
+                    })
+        
+        return items
+
+    def _flatten_dict(self, d, parent_key=''):
+        items = {}
+        
+        if isinstance(d, list):
+            return [self._flatten_dict(item) for item in d]
+                
+        for k, v in d.items():
+            new_key = f"{parent_key}_{k}" if parent_key else k
+            
+            if isinstance(v, dict):
+                items.update(self._flatten_dict(v, new_key))
+            else:
+                items[new_key] = str(v)
+                    
+        return items
+   
+    def __iter__(self):
+        if not self.data:
+            self.parse()
+
+        if self.type == 'INFORMATION TABLE':
+            return iter(self.data)
+        elif self.type == '8-K':
+            return iter(self._document_to_section_text(self.data['document']))
+        elif self.type == '10-K':
+            return iter(self._document_to_section_text(self.data['document']))
+        elif self.type == '10-Q':
+            return iter(self._document_to_section_text(self.data['document']))
+        elif self.type in ['3', '4', '5']:
+            return iter(self._flatten_dict(self.data['holdings']))
+        elif self.type == 'D':
+            return iter(self._flatten_dict(self.data['document']['relatedPersonsList']['relatedPersonInfo']))
+        elif self.type == 'NPORT-P':
+            return iter(self._flatten_dict(self.data['document']['formData']['invstOrSecs']['invstOrSec']))
+        elif self.type == 'SC 13D':
+            return iter(self._document_to_section_text(self.data['document']))
+        elif self.type == 'SC 13G':
+            return iter(self._document_to_section_text(self.data['document']))

datamule/helper.py

@@ -0,0 +1,123 @@
+import requests
+import os
+from tqdm import tqdm
+import zipfile
+from pkg_resources import resource_filename
+import csv
+import re
+
+# Unused in current implementation.
+def construct_primary_doc_url(cik, accession_number,primary_doc_url):
+    accession_number = accession_number.replace("-", "")
+    return f"https://www.sec.gov/Archives/edgar/data/{cik}/{accession_number}/{primary_doc_url}"
+
+# DONE
+def _download_from_dropbox(url, output_path):
+    headers = {'user-agent': 'Wget/1.16 (linux-gnu)'}
+    r = requests.get(url, stream=True, headers=headers)
+    total_size = int(r.headers.get('content-length', 0))
+    
+    with open(output_path, 'wb') as f, tqdm(
+        desc="Downloading " + os.path.basename(output_path),
+        total=total_size,
+        unit='iB',
+        unit_scale=True,
+        unit_divisor=1024,
+    ) as progress_bar:
+        for chunk in r.iter_content(chunk_size=1024):
+            size = f.write(chunk)
+            progress_bar.update(size)
+    
+    # Check if the downloaded file is a zip file
+    if zipfile.is_zipfile(output_path):
+        extract_path = os.path.dirname(output_path)
+        with zipfile.ZipFile(output_path, 'r') as zip_ref:
+            for file_info in zip_ref.infolist():
+                extract_file_path = os.path.join(extract_path, file_info.filename)
+                with zip_ref.open(file_info) as file_in_zip, \
+                    open(extract_file_path, 'wb') as output_file, \
+                    tqdm(total=file_info.file_size, unit='B', unit_scale=True, 
+                         desc=f"Extracting {file_info.filename}") as pbar:
+                    while True:
+                        chunk = file_in_zip.read(8192)
+                        if not chunk:
+                            break
+                        output_file.write(chunk)
+                        pbar.update(len(chunk))
+        
+        # Remove the zip file after extraction
+        os.remove(output_path)
+        print(f"Extracted contents to {extract_path}")
+    else:
+        print(f"Downloaded file is not a zip. Saved to {output_path}")
+
+# May generalize to load any package resource
+def load_package_csv(name):
+    """Load package CSV files"""
+    csv_path = resource_filename('datamule', f'data/{name}.csv')
+    company_tickers = []
+    
+    with open(csv_path, 'r') as csvfile:
+        csv_reader = csv.DictReader(csvfile)
+        for row in csv_reader:
+            company_tickers.append(row)
+    
+    return company_tickers
+
+def load_package_dataset(dataset):
+    if dataset == 'company_tickers':
+        return load_package_csv('company_tickers')
+    elif dataset =='company_former_names':
+        return load_package_csv('company_former_names')
+    elif dataset =='company_metadata':
+        return load_package_csv('company_metadata')
+    elif dataset == 'sec_glossary':
+        return load_package_csv('sec-glossary')
+    elif dataset == 'xbrl_descriptions':
+        return load_package_csv('xbrl_descriptions')
+
+# DONE
+def identifier_to_cik(ticker):
+    """Convert company tickers to CIK codes"""
+    company_tickers = load_package_csv('company_tickers')
+    if ticker:
+        if isinstance(ticker, list):
+            cik = []
+            for t in ticker:
+                cik.extend([company['cik'] for company in company_tickers if t == company['ticker']])
+        else:
+            cik = [company['cik'] for company in company_tickers if ticker == company['ticker']]
+
+    if not cik:
+        raise ValueError("No matching companies found")
+
+    return cik
+
+
+def fix_filing_url(url):
+    match_suffix = re.search(r'/(\d{4})\.(.+?)$', url)
+    if match_suffix:
+        suffix_number = match_suffix.group(1)
+        file_ext = match_suffix.group(2)
+        match_accession = re.search(r'/(\d{18})/', url)
+        if match_accession:
+            accession_number = match_accession.group(1)
+            formatted_accession_number = f"{accession_number[:10]}-{accession_number[10:12]}-{accession_number[12:]}"
+            new_url = url.rsplit('/', 1)[0] + f'/{formatted_accession_number}-{suffix_number}.{file_ext}'
+            return new_url
+    return url
+
+def convert_to_dashed_accession(accession):
+    # Remove any existing dashes or whitespace
+    cleaned = ''.join(accession.split())
+    
+    # Check if the cleaned string has 18 characters
+    if len(cleaned) != 18:
+        raise ValueError("Invalid accession number format. Expected 18 characters.")
+    
+    # Insert dashes at the correct positions
+    dashed = f"{cleaned[:10]}-{cleaned[10:12]}-{cleaned[12:]}"
+    
+    return dashed
+
+headers = {'User-Agent': 'John Smith johnsmith@gmail.com'}