import pandas as pd import matplotlib.pyplot as plt import seaborn as sns from pathlib import Path import glob from concurrent.futures import ProcessPoolExecutor from tqdm import tqdm import time import argparse import numpy as np import gc class VisualizeWikipedia: def __init__(self, parquet_dir: str | Path, sample: bool = False, batch_size: int = 10): self.parquet_dir = Path(parquet_dir) self.sample = sample self.batch_size = batch_size self.parquet_files = glob.glob(str(self.parquet_dir / "*.parquet")) print(f"Found {len(self.parquet_files)} parquet files") @staticmethod def read_parquet_file(file): return pd.read_parquet(file) def process_batch(self, files): with ProcessPoolExecutor() as executor: dfs = list(executor.map(self.read_parquet_file, files)) return pd.concat(dfs, ignore_index=True) def visualize(self): if self.sample: sample_files = self.parquet_files[::10] else: sample_files = self.parquet_files print(f"Using {len(sample_files)} files for visualization") # Initialize statistics total_rows = 0 total_sentences = 0 text_lengths = [] sentence_counts = [] date_counts = {} sentence_lengths = [] # Process files in batches start_time = time.time() for i in tqdm(range(0, len(sample_files), self.batch_size), desc="Processing batches"): try: batch_files = sample_files[i:i + self.batch_size] batch_df = self.process_batch(batch_files) # Update statistics total_rows += len(batch_df) if 'text_sentences' in batch_df.columns: batch_sentence_lists = batch_df['text_sentences'].dropna() batch_sentence_counts = batch_sentence_lists.apply(len) total_sentences += batch_sentence_counts.sum() sentence_counts.extend(batch_sentence_counts.tolist()) # Flatten and get sentence lengths for sentence_list in batch_sentence_lists: sentence_lengths.extend([len(s) for s in sentence_list if isinstance(s, str)]) if 'text' in batch_df.columns: text_lengths.extend(batch_df['text'].str.len().tolist()) if 'timestamp' in batch_df.columns: batch_df['timestamp'] = pd.to_datetime(batch_df['timestamp']) batch_df['date'] = batch_df['timestamp'].dt.date for date, count in batch_df['date'].value_counts().items(): date_counts[date] = date_counts.get(date, 0) + count # Clear memory after each batch del batch_df gc.collect() except Exception as e: print(f"\nError processing batch {i//self.batch_size + 1}: {str(e)}") continue print(f"\nProcessing took {time.time() - start_time:.2f} seconds") # Display basic information print("\nDataset Information:") print("=" * 50) print(f"Total number of rows: {total_rows:,}") print(f"Total number of sentences: {total_sentences:,}") # Get a sample of the data for detailed statistics sample_df = pd.read_parquet(sample_files[0]) print("\nColumns in the dataset:") print(sample_df.columns.tolist()) print("\nData types:") print(sample_df.dtypes) print("\nSample rows:") print("=" * 50) print(sample_df.head()) del sample_df gc.collect() # Create visualizations fig, axes = plt.subplots(2, 2, figsize=(24, 12)) # 2 rows, 2 columns axes = axes.flatten() # Flatten the 2x2 array for easier indexing # 1. Text length distribution if text_lengths: sns.histplot(data=text_lengths, bins=range(0, 30001, 250), ax=axes[0]) axes[0].set_title('Text Lengths') axes[0].set_xlabel('Characters') axes[0].set_ylabel('Count') axes[0].set_xlim(0, 30000) # 2. Article date distribution if date_counts: dates = sorted(date_counts.keys()) counts = [date_counts[date] for date in dates] axes[1].plot(dates, counts) axes[1].set_title('Articles Over Time') axes[1].set_xlabel('Date') axes[1].set_ylabel('Count') axes[1].tick_params(axis='x', rotation=45) # 3. Sentence count per article if sentence_counts: sns.histplot(data=sentence_counts, bins=range(0, 1001, 10), ax=axes[2]) axes[2].set_title('Sentence Counts per Article') axes[2].set_xlabel('Sentence Count') axes[2].set_ylabel('Count') axes[2].set_xlim(0, 1000) # 4. Sentence length if sentence_lengths: sns.histplot(data=sentence_lengths, bins=range(0, 301, 2), ax=axes[3]) axes[3].set_title('Sentence Lengths') axes[3].set_xlabel('Characters') axes[3].set_ylabel('Count') axes[3].set_xlim(0, 300) # Save the plot plt.tight_layout() plt.savefig('src/_TEMP/wikipedia_dataset_visualization.png') print("\nVisualization saved as 'wikipedia_dataset_visualization.png'") # Display basic statistics print("\nBasic Statistics:") print("=" * 50) if text_lengths: print(f"Text length - Mean: {np.mean(text_lengths):.2f}, Median: {np.median(text_lengths):.2f}") print(f"Text length - Min: {min(text_lengths):,}, Max: {max(text_lengths):,}") print(f"Text length - 25th percentile: {np.percentile(text_lengths, 25):.2f}") print(f"Text length - 75th percentile: {np.percentile(text_lengths, 75):.2f}") if sentence_counts: print(f"\nSentence count - Mean: {np.mean(sentence_counts):.2f}, Median: {np.median(sentence_counts):.2f}") print(f"Sentence count - Min: {min(sentence_counts):,}, Max: {max(sentence_counts):,}") print(f"Sentence count - 25th percentile: {np.percentile(sentence_counts, 25):.2f}") print(f"Sentence count - 75th percentile: {np.percentile(sentence_counts, 75):.2f}") if sentence_lengths: print(f"\nSentence length - Mean: {np.mean(sentence_lengths):.2f}, Median: {np.median(sentence_lengths):.2f}") print(f"Sentence length - Min: {min(sentence_lengths):,}, Max: {max(sentence_lengths):,}") print(f"Sentence length - 25th percentile: {np.percentile(sentence_lengths, 25):.2f}") print(f"Sentence length - 75th percentile: {np.percentile(sentence_lengths, 75):.2f}") if __name__ == "__main__": parser = argparse.ArgumentParser(description="Visualize Wikipedia dataset") parser.add_argument("-d", "--parquet_dir", type=str, required=True, help="Directory containing the parquet files") parser.add_argument("-s", "--sample", action="store_true", help="Use a sample of files for faster processing") parser.add_argument("-b", "--batch_size", type=int, default=10, help="Number of files to process at once") args = parser.parse_args() visualize_wikipedia = VisualizeWikipedia(parquet_dir=args.parquet_dir, sample=args.sample, batch_size=args.batch_size) visualize_wikipedia.visualize()