split_data.py

import pandas as pd
from sklearn.model_selection import train_test_split

# Load all 100 examples
df = pd.read_csv("data.csv")

# First split: 80 training examples, 20 remaining
train_df, remaining_df = train_test_split(
    df,
    test_size=0.2,
    stratify=df["label"],
    random_state=42
)

# Second split: divide the remaining 20 equally
validation_df, test_df = train_test_split(
    remaining_df,
    test_size=0.5,
    stratify=remaining_df["label"],
    random_state=42
)

train_df.to_csv("train.csv", index=False)
validation_df.to_csv("validation.csv", index=False)
test_df.to_csv("test.csv", index=False)

print(f"Training examples:   {len(train_df)}")
print(f"Validation examples: {len(validation_df)}")
print(f"Test examples:       {len(test_df)}")