Skip to content

Latest commit

 

History

History
160 lines (116 loc) · 4.92 KB

README.md

File metadata and controls

160 lines (116 loc) · 4.92 KB

Regression Model Implementation and Optimization

Overview

This project implements a variety of regression models, both linear and ensemble-based, to predict numerical outcomes. The models are optimized using Grid Search and Randomized Grid Search techniques to find the best hyperparameters for each model.

Contents

  1. Linear Models
  2. Ensemble Models
  3. Hyperparameter Optimization
    • Grid Search
    • Randomized Grid Search
  4. How to Use
  5. Dependencies

Linear Models

The following linear regression models are included:

1. LinearRegression

A basic linear regression model that minimizes the mean squared error.

2. Ridge

Ridge regression adds L2 regularization to the loss function to prevent overfitting.

3. SGDRegressor

Stochastic Gradient Descent is used to minimize the loss function iteratively.

4. Lasso

Lasso regression adds L1 regularization to the loss function, which performs feature selection by driving some coefficients to zero.

5. ElasticNetCV

Combines L1 and L2 regularization with cross-validation to find the best combination of hyperparameters.

6. HuberRegressor

A robust regression method that minimizes the influence of outliers on the model.

7. QuantileRegressor

Predicts a specific quantile (e.g., median) instead of the mean of the response variable.

8. RANSACRegressor

A robust regression model that iteratively fits the model to subsets of the data and identifies inliers.

9. PoissonRegressor

Used for count data regression assuming the response variable follows a Poisson distribution.

10. TweedieRegressor

Handles distributions from the Tweedie family, including compound Poisson-gamma.

11. GammaRegressor

Models data that follow a gamma distribution, useful for positively skewed data.

Ensemble Models

The following ensemble-based regression models are included:

1. DecisionTreeRegressor

A non-linear model that uses a tree structure to split data based on feature values.

2. RandomForestRegressor

An ensemble of decision trees where each tree is trained on a random subset of the data.

3. GradientBoostingRegressor

An iterative method that combines weak learners (trees) to minimize the loss function.

4. XGBoost

An optimized gradient boosting implementation designed for speed and performance.

5. LightGBM

A gradient boosting framework optimized for efficiency, capable of handling large datasets.

6. CatBoost

A gradient boosting algorithm that handles categorical features natively.

Hyperparameter Optimization

Grid Search

This exhaustive search method evaluates all possible combinations of hyperparameter values specified in the search space.

Example:

from sklearn.model_selection import GridSearchCV
from sklearn.linear_model import Ridge

param_grid = {
    'alpha': [0.1, 0.5, 1.0],
    'max_iter': [1000, 2000, 3000]
}
grid_search = GridSearchCV(estimator=Ridge(), param_grid=param_grid, scoring='neg_mean_squared_error', cv=5)
grid_search.fit(X_train, y_train)

print("Best Parameters:", grid_search.best_params_)
print("Best Score:", grid_search.best_score_)

Randomized Grid Search

This method evaluates a random subset of the hyperparameter search space, making it faster for large datasets and complex models.

Example:

from sklearn.model_selection import RandomizedSearchCV
from sklearn.ensemble import GradientBoostingRegressor

param_dist = {
    'n_estimators': [50, 100, 200],
    'max_depth': [None, 10, 20],
    'learning_rate': [0.01, 0.1, 0.2]
}
random_search = RandomizedSearchCV(estimator=GradientBoostingRegressor(), param_distributions=param_dist, scoring='r2', n_iter=10, cv=5, random_state=42)
random_search.fit(X_train, y_train)

print("Best Parameters:", random_search.best_params_)
print("Best Score:", random_search.best_score_)

How to Use

  1. Import Models: The models are defined in dictionaries (regressors and ensemble_models). Use the key to access a specific model.

    from sklearn.linear_model import LinearRegression

model = regressors['LinearRegression']
model.fit(X_train, y_train)
predictions = model.predict(X_test)

  2. Perform Optimization: Use either GridSearchCV or RandomizedSearchCV to optimize hyperparameters.

  3. Evaluate Models: Evaluate the model using metrics such as Mean Squared Error (MSE), R2 Score, or Mean Absolute Error (MAE).

    from sklearn.metrics import mean_squared_error, r2_score

mse = mean_squared_error(y_test, predictions)
r2 = r2_score(y_test, predictions)
print(f"MSE: {mse}, R2: {r2}")

Dependencies

  • Python 3.7+
  • scikit-learn
  • xgboost
  • lightgbm
  • catboost

Install required libraries:

pip install scikit-learn xgboost lightgbm catboost