ridge and lasso

import pandas as pd
import numpy as np
from sklearn.linear_model import Lasso, Ridge
from sklearn.model_selection import RandomizedSearchCV
from sklearn.feature_selection import SelectKBest, f_regression

# Load the training data once into memory
train_file_path = '/Users/scott/Downloads/house-prices/house_train_engineered.csv'
train_df = pd.read_csv(train_file_path)

# Separate features and target in the training data
X_train = train_df.drop(columns=['SalePrice', 'Id'])
y_train = train_df['SalePrice']

# Feature reduction
selector = SelectKBest(score_func=f_regression, k=50)
X_train_reduced = selector.fit_transform(X_train, y_train)

# Define a more targeted range of alpha values
alpha_values_expanded = np.linspace(0.00001, 500, num=1000)

# Additional hyperparameters with expanded range
lasso_max_iter_values = [10000, 50000, 100000]
ridge_max_iter_values = [5000, 10000, 50000]
lasso_tol_values = [1e-4, 1e-5, 1e-6]
ridge_tol_values = [1e-2, 1e-3, 1e-4]

# Initialize models
lasso = Lasso()
ridge = Ridge()


# For Lasso
lasso_alpha_values_expanded = np.linspace(0.000001, 0.0001, num=100)  # Around 0.00001
lasso_max_iter_values_expanded = [48000, 49000, 50000, 51000, 52000]  # Around 50000
lasso_tol_values_expanded = [0.00001, 0.0001, 0.001, 0.0005, 0.00005]  # Around 0.00001

lasso_param_grid_expanded = {
    'alpha': lasso_alpha_values_expanded,
    'max_iter': lasso_max_iter_values_expanded,
    'tol': lasso_tol_values_expanded
}

# For Ridge
ridge_alpha_values_expanded = np.linspace(110, 120, num=100)  # Around 116.6
ridge_max_iter_values_expanded = [9500, 9700, 9900, 10000, 10200, 10400]  # Around 10000
ridge_tol_values_expanded = [9e-05, 1e-04, 1.1e-04, 1.2e-04]  # Around 1e-04

ridge_param_grid_expanded = {
    'alpha': ridge_alpha_values_expanded,
    'max_iter': ridge_max_iter_values_expanded,
    'tol': ridge_tol_values_expanded
}

# RandomizedSearchCV with expanded hyperparameters
lasso_search = RandomizedSearchCV(lasso, param_distributions=lasso_param_grid_expanded, n_iter=1000, cv=5, scoring='neg_mean_squared_error', verbose=1, n_jobs=-1)
ridge_search = RandomizedSearchCV(ridge, param_distributions=ridge_param_grid_expanded, n_iter=1000, cv=5, scoring='neg_mean_squared_error', verbose=1, n_jobs=-1)

# Fit the models to the reduced training set
lasso_search.fit(X_train_reduced, np.log1p(y_train))
ridge_search.fit(X_train_reduced, np.log1p(y_train))

# Print the optimal hyperparameters
print(f"Lasso Optimal Hyperparameters: {lasso_search.best_params_}")
print(f"Ridge Optimal Hyperparameters: {ridge_search.best_params_}")

# Print the mean squared error across CV folds
print(f"Lasso CV Negative Mean Squared Error: {lasso_search.best_score_}")
print(f"Ridge CV Negative Mean Squared Error: {ridge_search.best_score_}")

# Load the test data and preprocess
test_file_path = '/Users/scott/Downloads/house-prices/house_test_engineered.csv'
test_df = pd.read_csv(test_file_path)

# Feature scaling and selection for test data (drop 'Id' and 'SalePrice' if present)
X_test = test_df.drop(columns=['Id', 'SalePrice'], errors='ignore')
X_test_reduced = selector.transform(X_test)

# Generate predictions
lasso_predictions = np.expm1(lasso_search.best_estimator_.predict(X_test_reduced))
ridge_predictions = np.expm1(ridge_search.best_estimator_.predict(X_test_reduced))

# Create submission DataFrames
submission_lasso = pd.DataFrame({'Id': test_df['Id'], 'SalePrice': lasso_predictions})
submission_ridge = pd.DataFrame({'Id': test_df['Id'], 'SalePrice': ridge_predictions})

# Save submission files
submission_lasso.to_csv('submission_lasso.csv', index=False)
submission_ridge.to_csv('submission_ridge.csv', index=False)

print("Submission files generated successfully.")