kaggle_learn

This module is a collection for useful code in kaggle (or more data science general), includes general utility functions, feature engineering functions, model training / cross-validation / model stacking helper functions.

Modules

Preprocessing

Numerical / categorical (tabular data) and text data preprocessing.

Feature Engineering

Some commonly used feature engineering methods used in kaggle competitions:

• group by on one / multiple categorical features and get the summary statistics of numerical features
• text meta features
• text similarity features
• time meta features
• ...

Models

Models not implemented in sklearn:

• NB-LR (Logistic Regression with Naive Bayes features)

Neural network templates for different tasks:

• text classification (RNN / CNN based)

Model Runner

Functions / classes that takes data and run specified models (with cross validation).

Feature Importance

• Permutation feature importance

Metrics

• customized keras callbacks

Utilities

Utility functions:

• reduce pandas dataframe memory
• timer / logger
• ...

Usage

Clone this repository:

git clone https://github.com/bangdasun/kaggle_learn.git

Create environment to install required packages:

cd kaggle_learn
conda env create -f standard_env.yaml