src/multivariate_stratification.py

import pandas as pd
import numpy as np

# Load the data
df = pd.read_csv("~/Desktop/sabdab_summary_with_flags.tsv", sep="\t")

# Create the strata column
strata_cols = ["antigen_type", "heavy_species", "method", "scfv",
               "engineered", "light_ctype", "in_nr_set", "curated_quality_dataset"]
df["strata"] = df[strata_cols].astype(str).agg("_".join, axis=1)

# Initialize split column
df["split"] = ""

# Define fractions
train_frac = 0.8
val_frac = 0.1
test_frac = 0.1

# Set random seed for reproducibility
np.random.seed(42)

# Get all unique strata
all_strata = df["strata"].unique()

for s in all_strata:
    idx = df[df["strata"] == s].index.to_list()
    np.random.shuffle(idx)
    n = len(idx)
    n_train = max(1, int(n * train_frac))
    n_val = max(1, int(n * val_frac))
    n_test = n - n_train - n_val  # whatever remains

    # Adjust in case rounding caused n_test < 1
    if n_test < 1:
        n_test = 1
        if n_val > 1:
            n_val -= 1
        else:
            n_train -= 1

    # Assign
    df.loc[idx[:n_train], "split"] = "train"
    df.loc[idx[n_train:n_train+n_val], "split"] = "validation"
    df.loc[idx[n_train+n_val:], "split"] = "test"

# Drop temporary strata column
df.drop(columns=["strata"], inplace=True)

# Check counts
print(df["split"].value_counts(normalize=True) * 100)

cols = ["antigen_type", "heavy_species", "method", "scfv",
        "engineered", "light_ctype", "in_nr_set", "curated_quality_dataset"]

for col in cols:
    print(f"\n=== {col} ===")
    # Cross-tab of counts by split
    ct = pd.crosstab(df[col], df["split"])
    # Convert counts to percentages **row-wise** so each category sums to 100%
    ct_percent = ct.div(ct.sum(axis=1), axis=0) * 100
    print(ct_percent.round(1))

# Save to TSV
df.to_csv("~/Desktop/sabdab_summary_with_splits.tsv", sep="\t", index=False)
print("Saved updated DataFrame with split column to ~/Desktop/sabdab_summary_with_splits.tsv")