mlp_no_pca.py

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.neural_network import MLPRegressor
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score

# 1. Load Local Dataset
excel_path = r"C:\Users\bhave\OneDrive\Bhaveshya\bio glass research\Book2_grouped_train_test_split.xlsx"
df = pd.read_excel(excel_path, sheet_name='Sheet1')

# 2. Convert specified columns to categorical (string) format
CAT_COLS = [
    'SBF_pH', 'Post_Ca_P_Ratio', 'Cell_Viability_Pct',
    'Soaking_Period_Days', 'Cell_Extract_uL'
]
for col in CAT_COLS:
    if col in df.columns:
        df[col] = df[col].astype(str)

# 3. One-Hot Encoding
# Use the full dataset before train/val split to keep consistent feature columns
# across all targets when the same preprocessing is applied.
df_encoded = pd.get_dummies(df, drop_first=True)

# 4. Target List & General Exclusions
TARGETS = ['Density_kg_m3', 'Microhardness_GPa', 'Youngs_Modulus_GPa', 'Bulk_Modulus_GPa']
EXCLUDE_COLS = [
    'Sample_ID', 'Paper_Series', 'Formulation_Group', 'Composition_Group', 'Data_Split',
    'Density_kg_m3', 'Microhardness_GPa', 'Youngs_Modulus_GPa', 'Bulk_Modulus_GPa'
]

# Run MLP Regressor for each target
for TARGET in TARGETS:
    print(f"\n==========================================")
    print(f" MLP Regressor (No PCA) - Target: {TARGET}")
    print(f"==========================================")

    feature_cols = [
        col for col in df_encoded.columns
        if col not in EXCLUDE_COLS and not any(col.startswith(ex) for ex in EXCLUDE_COLS)
    ]

    train_data = df_encoded[df['Data_Split'] == 'Training'].copy()
    X = train_data[feature_cols]
    y = train_data[TARGET]

    X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.10, random_state=42)

    scaler = StandardScaler()
    X_train_scaled = scaler.fit_transform(X_train)
    X_val_scaled = scaler.transform(X_val)

    model = MLPRegressor(
        hidden_layer_sizes=(64, 32),
        activation='relu',
        solver='adam',
        learning_rate_init=0.01,
        max_iter=1,
        warm_start=True,
        random_state=42,
    )

    epochs = 150
    train_loss, val_loss = [], []

    for _ in range(epochs):
        model.fit(X_train_scaled, y_train)
        train_pred = model.predict(X_train_scaled)
        val_pred = model.predict(X_val_scaled)
        train_loss.append(mean_squared_error(y_train, train_pred))
        val_loss.append(mean_squared_error(y_val, val_pred))

    y_val_pred = model.predict(X_val_scaled)
    val_mse = mean_squared_error(y_val, y_val_pred)
    val_mae = mean_absolute_error(y_val, y_val_pred)
    val_r2 = r2_score(y_val, y_val_pred)
    val_rmse = np.sqrt(val_mse)

    print(f"Validation MSE:  {val_mse:.4f}")
    print(f"Validation MAE:  {val_mae:.4f}")
    print(f"Validation RMSE: {val_rmse:.4f}")
    print(f"Validation R²:   {val_r2:.4f}")

    plt.figure(figsize=(8, 5))
    plt.plot(range(1, epochs + 1), train_loss, label='Train Loss (MSE)', color='blue', linewidth=2)
    plt.plot(range(1, epochs + 1), val_loss, label='Validation Loss (MSE)', color='orange', linewidth=2, linestyle='--')
    plt.title(f'MLP Regressor (No PCA): Train vs Validation Loss ({TARGET})', fontsize=12, fontweight='bold')
    plt.xlabel('Epochs', fontsize=10)
    plt.ylabel('Loss (MSE)', fontsize=10)
    plt.legend(fontsize=10)
    plt.grid(True, linestyle=':', alpha=0.6)
    plt.tight_layout()
    plt.savefig(f'{TARGET}_mlp_no_pca_loss.png')
    plt.show()
    plt.close()