ml-toolbox

This repo contains various data science strategy and machine learning models to deal with structure as well as unstructured data. It contains module on feature-preprocessing, feature-engineering, machine-learning-models, bayesian-parameter-tuning, etc. Some of these features are collected from the existed libraries such as scikit-learn, keras, h2o, xgboost, lightgbm, catboost, etc. I have also added some technique, which I implemented by following the Research Paper and Data-Scientist advice(on kaggle). There are a lot of feature engineering strategy as well, which I developed during ML-contest and helped me a lot in those contest.

I use this toolbox for my personal usuage. I consistently update it, to make more generic.

Contents:

1. Feature preprocessing
   • cleaning
   • handling null value
   • normalization
   • grouping unknow variable
   • memory optimization
   • text preprocessing
2. Feature Engineering
   • label-encoder/one-hot/binary/hashing
   • binning/quantile-binning
   • target-encoding
   • bayesian-encoding
   • feature-interaction
   • date-time feature
   • time-lag featue(in time series)
   • rounding/decimal value
   • relation feature(aggregation based)
   • text feature using tf-idf, count-vect
   • clustering based feature(linear/non-linear)
   • polynomial feature
   • statistical ferature
3. EDA
   • boxplot/kdeplot/countplot/pairplot
   • heatmap
4. Machine Learning models
   • Tree base model
     • xgboost/lighgbm/catboost
     • sklearn: decision-tree/random-forest/extra-tree/GBM
   • Linear/Non-Linear Model:
     • Logistic-Regression/lasso/Ridge/Passive-Agreesive/SVM
   • Regularized Greedy forest(in progress)
   • field aware factorization machine
   • online learning(vowpal rabbit/follow the regularized leader)(in progress)
   • h2o models (gbm/rf/nn/auto-ml)
5. Deep learning models
   • neural networks(keras/tensorflow)
   • Attention mechanism for LSTM
   • Data augmentation(for image)
   • cyclic-learning-rate
   • keras custom loss-function/metric/callbacks
   • pretrained model
   • word2vec usuage using gensim
   • entity embedding
   • data on fly(efficient training)
   • segmentation
   • graph based neural network
6. Advances
   • gridsearch / bayesian optimization
   • Stacking/blending/rank-average
   • Bert-Model(pretrained text model)
   • parallel-processing for feature-engineering(in progress)