src/05_predict_outcomes_MK14.py

import pandas as pd
import numpy as np
import pyarrow as pa
import pyarrow.parquet as pq
import h2o
from h2o.automl import H2OAutoML
from h2o.frame import H2OFrame
import pickle
import os
import matplotlib.pyplot as plt

# Define the target
target = "MK14"  # Replace with WEE1 or CXCR4 as needed

# Start the H2O cluster
h2o.init()

# Load datasets
data_train = pq.read_table("intermediate_data/data_train_joined.parquet").to_pandas()
data_test = pq.read_table("intermediate_data/data_test_joined.parquet").to_pandas()

# Filter data for the selected target
data_train = data_train[data_train["Target Name"] == target]
data_test = data_test[data_test["Target Name"] == target]

# Define target column
target_column = " RMSD_Energy"

# Identify feature columns: all columns after "LF_score" (inclusive)
start_idx = list(data_train.columns).index("LF_score")
feature_columns = data_train.columns[start_idx:].tolist()
feature_columns = [col for col in feature_columns if col != " RMSD_Energy"]

# Convert to H2OFrame
train_h2o = H2OFrame(data_train)
test_h2o = H2OFrame(data_test)

# train_h2o[target_column] = train_h2o[target_column].asnumeric()
# test_h2o[target_column] = test_h2o[target_column].asnumeric()

# train_h2o[target_column] = train_h2o[target_column].asfactor()
# test_h2o[target_column] = test_h2o[target_column].asfactor()

# Train AutoML model
aml = H2OAutoML(max_models=1, seed=42)
aml.train(x=feature_columns, y=target_column, training_frame=train_h2o)

# Save the trained model
top_model = aml.leader
model_path = h2o.save_model(top_model, path = f"intermediate_data/top_model{target}", force=True)

# Generate predictions
for dataset_name, dataset_h2o, dataset_df in [
    ("train", train_h2o, data_train),
    ("test", test_h2o, data_test),
]:
    predictions = aml.leader.predict(dataset_h2o).as_data_frame()
    dataset_df["predictions"] = predictions["predict"]

    # Save predictions
    output_path = f"intermediate_data/data_{dataset_name}_{target}_pred.parquet"
    pq.write_table(pa.Table.from_pandas(dataset_df), output_path)

test_h2o_no_smiles = test_h2o.drop("SMILES")

# Analysis
top_model.explain(test_h2o_no_smiles)
plt.savefig(
	f"product/model_summary_{target}_{parameters["date_code"]}.pdf",
        format = "pdf", bbox_inches = "tight")

# Shutdown H2O cluster
h2o.shutdown(prompt=False)

print(f"Completed training and predictions for {target}")