"""
This script loads sanitized molecular fingerprint data, runs H2O AutoML
to predict Ki values (nM) for serotonin receptor ligands, and saves the
test set predictions.

Expected input file: fingerprints_with_ki.csv
Output: h2o_test_predictions.csv
"""

import h2o
import pandas as pd
from h2o.automl import H2OAutoML

# Initialize H2O cluster
h2o.init(max_mem_size="8G")

# Load fingerprinted dataset (must be in same directory or provide full path)
df = pd.read_csv("fingerprints_with_ki.csv")

# Convert to H2OFrame
df_h2o = h2o.H2OFrame(df)

# Ensure target column is numeric
df_h2o["Ki_nM"] = df_h2o["Ki_nM"].asnumeric()

# Split into training and test sets
train, test = df_h2o.split_frame(ratios=[0.8], seed=42)

# Define predictors and response
x = df_h2o.columns[:-1]  # all fingerprint columns
y = "Ki_nM"

# Run AutoML
aml = H2OAutoML(
    max_models=20,
    max_runtime_secs=600,
    seed=1,
    sort_metric="RMSE"
)
aml.train(x=x, y=y, training_frame=train)

# Evaluate model
perf = aml.leader.model_performance(test_data=test)
print("\n✅ Test Set Performance:")
print("R² Score:", perf.r2())
print("RMSE:", perf.rmse())

# Save predictions
preds = aml.leader.predict(test)
h2o.export_file(preds, path="h2o_test_predictions.csv", force=True)