mattia_stats/collect_stats.py

import os
import pandas as pd
import pyarrow.parquet as pq
import json
import numpy as np

# Configuration
base_dir = "/lustre/fsn1/projects/rech/fmr/uft12cr/statistics_corpus_full"
set_dirs = [f"set_{i}" for i in range(1, 11)]  # Sets 1 to 10, excluding filtered sets
test = False  # Set to True to process only a single Parquet file
min_tokens_threshold = 100000  # Minimum tokens required to report certain statistics

# Dictionary to accumulate statistics
stats = {
    "total_tokens_by_language": {},
    "total_words_by_language": {},
    "total_tokens_by_collection": {},
    "total_words_by_collection": {},
    "doc_count_by_language": {},
    "doc_count_by_collection": {},
    "license_distribution": {},
    "compression_rate_by_language": {},
    "avg_doc_length_words_by_language": {},
    "avg_doc_length_tokens_by_language": {},
    "compression_rate_by_collection": {},
    "avg_doc_length_words_by_collection": {},
    "avg_doc_length_tokens_by_collection": {},
}

processed_files = 0
errors_count = 0

# Helper function to update dictionaries for language and collection statistics
def update_stats(df):
    required_columns = {"collection", "language", "token_count", "word_count", "license"}
    if not required_columns.issubset(df.columns):
        print("DataFrame is missing required columns.")
        return

    try:
        df.loc[df['collection'].notna() & df['collection'].str.contains("Github", case=False), 'language'] = 'code'

        for collection_name, col_df in df.groupby("collection"):
            stats["total_tokens_by_collection"].setdefault(collection_name, 0)
            stats["total_tokens_by_collection"][collection_name] += col_df["token_count"].sum()
            stats["total_words_by_collection"].setdefault(collection_name, 0)
            stats["total_words_by_collection"][collection_name] += col_df["word_count"].sum()
            stats["doc_count_by_collection"].setdefault(collection_name, 0)
            stats["doc_count_by_collection"][collection_name] += len(col_df)

            license_counts = col_df["license"].value_counts().to_dict()
            for license_type, count in license_counts.items():
                stats["license_distribution"].setdefault(collection_name, {}).setdefault(license_type, 0)
                stats["license_distribution"][collection_name][license_type] += count

        for language, lang_df in df.groupby("language"):
            stats["total_tokens_by_language"].setdefault(language, 0)
            stats["total_tokens_by_language"][language] += lang_df["token_count"].sum()
            stats["total_words_by_language"].setdefault(language, 0)
            stats["total_words_by_language"][language] += lang_df["word_count"].sum()
            stats["doc_count_by_language"].setdefault(language, 0)
            stats["doc_count_by_language"][language] += len(lang_df)

    except Exception as e:
        print(f"Error updating statistics: {e}")

# Gather all parquet files, with only one if in test mode
all_parquet_files = []
for set_dir in set_dirs:
    set_path = os.path.join(base_dir, set_dir)
    if not os.path.isdir(set_path):
        print(f"Directory {set_path} not found, skipping...")
        continue
    try:
        parquet_files = [os.path.join(set_path, f) for f in os.listdir(set_path) if f.endswith(".parquet")]
        all_parquet_files.extend(parquet_files)
    except Exception as e:
        print(f"Error reading directory {set_path}: {e}")
if test:
    all_parquet_files = all_parquet_files[:1]

# Process all selected parquet files
total_files = len(all_parquet_files)
for file_path in all_parquet_files:
    try:
        df = pq.read_table(file_path).to_pandas()
        update_stats(df)
        processed_files += 1
    except Exception as e:
        print(f"Error processing file {file_path}: {e}")
        errors_count += 1

    if processed_files % 10 == 0 or processed_files == total_files:
        print(f"Processed {processed_files}/{total_files} files with {errors_count} errors.")

# Compute compression rates and average document lengths by language with threshold checks
for language in stats["total_tokens_by_language"]:
    try:
        total_tokens = stats["total_tokens_by_language"][language]
        total_words = stats["total_words_by_language"][language]
        doc_count = stats["doc_count_by_language"][language]

        if total_tokens >= min_tokens_threshold:
            stats["compression_rate_by_language"][language] = total_tokens / total_words if total_words > 0 else None
            stats["avg_doc_length_words_by_language"][language] = total_words / doc_count if doc_count > 0 else None
            stats["avg_doc_length_tokens_by_language"][language] = total_tokens / doc_count if doc_count > 0 else None
        else:
            stats["compression_rate_by_language"][language] = "N/A"
            stats["avg_doc_length_words_by_language"][language] = "N/A"
            stats["avg_doc_length_tokens_by_language"][language] = "N/A"
    except Exception as e:
        print(f"Error calculating stats for language {language}: {e}")

# Compute compression rates and average document lengths by collection with threshold checks
for collection in stats["total_tokens_by_collection"]:
    try:
        total_tokens = stats["total_tokens_by_collection"][collection]
        total_words = stats["total_words_by_collection"][collection]
        doc_count = stats["doc_count_by_collection"][collection]

        if total_tokens >= min_tokens_threshold:
            stats["compression_rate_by_collection"][collection] = total_tokens / total_words if total_words > 0 else None
            stats["avg_doc_length_words_by_collection"][collection] = total_words / doc_count if doc_count > 0 else None
            stats["avg_doc_length_tokens_by_collection"][collection] = total_tokens / doc_count if doc_count > 0 else None
        else:
            stats["compression_rate_by_collection"][collection] = "N/A"
            stats["avg_doc_length_words_by_collection"][collection] = "N/A"
            stats["avg_doc_length_tokens_by_collection"][collection] = "N/A"
    except Exception as e:
        print(f"Error calculating stats for collection {collection}: {e}")

# Convert to native types function
def convert_to_native_types(stats):
    def convert(value):
        if isinstance(value, (np.integer, np.floating)):
            return value.item()
        elif isinstance(value, dict):
            return {k: convert(v) for k, v in value.items()}
        return value
    return {k: convert(v) for k, v in stats.items()}

# Print and save statistics in human-readable format
def print_stats(stats):
    output = []
    output.append("============ Corpus Statistics Overview ============")
    total_tokens = sum(stats["total_tokens_by_collection"].values())
    total_words = sum(stats["total_words_by_collection"].values())
    total_docs = sum(stats["doc_count_by_collection"].values())
    output.append(f"\nTotal Tokens in Corpus: {total_tokens:,}")
    output.append(f"Total Words in Corpus: {total_words:,}")
    output.append(f"Total Documents in Corpus: {total_docs:,}")

    # Display top 10 collections by total tokens
    output.append("\nTop 10 Collections by Total Tokens:")
    for collection, count in sorted(stats["total_tokens_by_collection"].items(), key=lambda x: x[1], reverse=True)[:10]:
        output.append(f"  - {collection}: {count:,}")

    # Display top 10 languages by total tokens
    output.append("\nTop 10 Languages by Total Tokens:")
    for language, count in sorted(stats["total_tokens_by_language"].items(), key=lambda x: x[1], reverse=True)[:10]:
        output.append(f"  - {language}: {count:,}")

    # Display compression rate by language (top 10)
    output.append("\nCompression Rate by Language (Top 10):")
    for language, rate in sorted(stats["compression_rate_by_language"].items(), key=lambda x: x[1] if isinstance(x[1], (int, float)) else 0, reverse=True)[:10]:
        output.append(f"  - {language}: {rate:.2f}" if isinstance(rate, (int, float)) else f"  - {language}: N/A")

    # Display average document length by language (top 10)
    output.append("\nAverage Document Length (Words) by Language (Top 10):")
    for language, avg_len in sorted(stats["avg_doc_length_words_by_language"].items(), key=lambda x: x[1] if isinstance(x[1], (int, float)) else 0, reverse=True)[:10]:
        output.append(f"  - {language}: {avg_len:.2f}" if isinstance(avg_len, (int, float)) else f"  - {language}: N/A")

    # License distribution by collection (top 5 collections)
    output.append("\nLicense Distribution by Collection (Top 5):")
    for collection, licenses in list(sorted(stats["license_distribution"].items(), key=lambda x: sum(x[1].values()), reverse=True))[:5]:
        output.append(f"  - {collection}:")
        for license_type, count in sorted(licenses.items(), key=lambda x: x[1], reverse=True):
            output.append(f"    * {license_type}: {count:,}")

    output.append("====================================================")
    print("\n".join(output))

    # Save to text file
    with open('stats_readable_output.txt', 'w') as f:
        f.write("\n".join(output))

# Print and save the human-readable stats
print_stats(stats)

# Convert to native types and save as JSON
stats_native = convert_to_native_types(stats)

with open('stats_by_language.json', 'w') as f:
    json.dump({
        'total_tokens_by_language': stats_native['total_tokens_by_language'],
        'total_words_by_language': stats_native['total_words_by_language'],
        'doc_count_by_language': stats_native['doc_count_by_language'],
        'compression_rate_by_language': stats_native['compression_rate_by_language'],
        'avg_doc_length_words_by_language': stats_native['avg_doc_length_words_by_language'],
        'avg_doc_length_tokens_by_language': stats_native['avg_doc_length_tokens_by_language'],
    }, f, indent=4)

with open('stats_by_collection.json', 'w') as f:
    json.dump({
        'total_tokens_by_collection': stats_native['total_tokens_by_collection'],
        'total_words_by_collection': stats_native['total_words_by_collection'],
        'doc_count_by_collection': stats_native['doc_count_by_collection'],
        'compression_rate_by_collection': stats_native['compression_rate_by_collection'],
        'avg_doc_length_words_by_collection': stats_native['avg_doc_length_words_by_collection'],
        'avg_doc_length_tokens_by_collection': stats_native['avg_doc_length_tokens_by_collection'],
        'license_distribution': stats_native['license_distribution'],
    }, f, indent=4)