scripts/split.py

import toml
from collections import Counter
import srsly
import typer
from sklearn.model_selection import train_test_split
from pathlib import Path

def count_labels(data):
    label_counter = Counter()
    for annotation in data:
        labels = [span['label'] for span in annotation['spans']]
        label_counter.update(labels)
    return label_counter

def split_data(input_file: Path, train_ratio: float = 0.7, dev_ratio: float = 0.15, 
               train_output: Path = Path("assets/train.jsonl"), 
               dev_output: Path = Path("assets/dev.jsonl"), 
               test_output: Path = Path("assets/test.jsonl"), 
               random_state: int = 1):
    # Read data from JSONL
    data = list(srsly.read_jsonl(input_file))

    # Split data
    test_ratio = 1 - train_ratio - dev_ratio
    train_data, temp_data = train_test_split(data, test_size=(dev_ratio + test_ratio), random_state=random_state)
    dev_data, test_data = train_test_split(temp_data, test_size=test_ratio/(dev_ratio + test_ratio), random_state=random_state)

    # Count labels in each dataset
    train_labels = count_labels(train_data)
    dev_labels = count_labels(dev_data)
    test_labels = count_labels(test_data)

    # Write split data to JSONL files
    srsly.write_jsonl(train_output, train_data)
    srsly.write_jsonl(dev_output, dev_data)
    srsly.write_jsonl(test_output, test_data)

    # Combine label counts into a dictionary
    all_labels = sorted(set(train_labels.keys()) | set(dev_labels.keys()) | set(test_labels.keys()))
    annotations_data = {label: {'Train': train_labels.get(label, 0), 'Dev': dev_labels.get(label, 0), 'Test': test_labels.get(label, 0)} for label in all_labels}

    # Print the table
    print(f"{'Label':<20}{'Train':<10}{'Dev':<10}{'Test':<10}")
    for label in all_labels:
        print(f"{label:<20}{train_labels.get(label, 0):<10}{dev_labels.get(label, 0):<10}{test_labels.get(label, 0):<10}")

    # Save the table in annotations.toml
    with open("annotations.toml", "w") as toml_file:
        toml.dump(annotations_data, toml_file)

    print("Annotations data saved in annotations.toml")

if __name__ == "__main__":
    typer.run(split_data)