baserush

Stable base modeling made quick and easy!

baserush is an easy-to-use regression pipeline for preprocessing, optimizing, and summarizing machine learning models within the scikit-learn framework. This package is ideal for efficiently building and comparing stable models from different model types.

Supported Model Types

Linear Models

• LinearRegression
• Lasso
• Ridge
• SGDRegressor

Neighbors Models

• KNeighborsRegressor
• RadiusNeighborsRegressor

CaRT Models

• DecisionTreeRegressor
• ExtraTreeRegressor

Ensemble Models

• RandomForestRegressor
• GradientBoostingRegressor
• ExtraTreesRegressor
• RandomTreesEmbedding

Package Modules

• preprocess: missing values, skewness, standardization, and categorical transformations
• optimize: automatic feature selection; hyperparameter analysis
• summary: training and validation R-Squared, stability tools; model-specific outputs

preprocessing Features

• simputer makes it simple to flag and impute missing values.
• Quickly alleviate skewness with transtorm.
• Use simple_scaler to seamlessly standardize features.
• Efficiently prepare categorical data for modeling with catcoder.

optimize-ation Features

• Base modeling made easy with

  • quick_lm (with automated feature selection)
  • quick_tree, (includes very fast automated hyperparameter tuning)
  • quick_neighbors, (automatically tunes n neighbors)

• Use tuning_results to analyze the top n-models after hyperparameter tuning with GridSeachCV | RandomizedSearchCV.

summary Features

lr_summary, tree_summary, and knn_summary

• Automatically instantiate customizable training and validation sets.
• Generate a dataset of model summaries for easy comparison, including:
  • Model Name
  • Model Class
  • Model Type
  • R-Squared (Training Set)
  • R-Squared (Validation Set)
  • Train-Test Gap
  • Model-Specific:
    • Model Coefficients
    • Feature Importance
    • Hyperparameter Values

Installation

Install using pip:

pip install baserush

Example Usage

print("Examples coming soon.")

License

MIT License. See LICENSE for details.