import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error, r2_score

train_df = pd.read_parquet("./ml_input_data/train_split.parquet")
valid_df=pd.read_parquet("./ml_input_data/validation_split.parquet")
test_df=pd.read_parquet("./ml_input_data/test_split.parquet")

chem_properties=list(train_df.keys()[1:9])
ecfp=list(train_df.keys()[24:])
target="LUMO_HOMO_GAP"
feature_cols = chem_properties + ecfp


X_train, y_train = train_df[feature_cols], train_df[target]
X_valid, y_valid = valid_df[feature_cols], valid_df[target]
X_test,  y_test  = test_df[feature_cols],  test_df[target]

# train simple base line model
rf_10 = RandomForestRegressor(
    n_estimators=10,
    max_depth=None,
    random_state=42,
)
rf_10.fit(X_train, y_train)
y_pred = rf_10.predict(X_test)
mse = mean_squared_error(y_test, y_pred)
r2  = r2_score(y_test,  y_pred)

print(f"Test  MSE: {mse:.4f}")
print(f"Test  R² : {r2:.4f}") 

#test a subset of molecules with the highest 10% energy gaps
all_data="./ml_input_data/all_data_no_split.parquet"
df=pd.read_parquet(all_data)
top_10_percent = df.sort_values(by='LUMO_HOMO_GAP', ascending=False).head(int(0.1 * len(df)))
top_10_percent_X_test,  top_10_percent_y_test  = top_10_percent[feature_cols],  top_10_percent[target]
top_10_percent_y_pred = rf_10.predict(top_10_percent_X_test)
mse = mean_squared_error(top_10_percent_y_test, top_10_percent_y_pred)
r2  = r2_score(top_10_percent_y_test,  top_10_percent_y_pred)
print(f"Top 10% Test  MSE: {mse:.4f}")
print(f"Top 10% Test  R² : {r2:.4f}")