Scikit-Learn Library

Sci-kit Learn

Data and Feature Processing

• categorical encoders

Modelling

3 Steps

1. Instantiate model
2. model.fit
3. model.predict

Supervised Learning

Supervised Learning

• linear_model
  • LogisticRegression
• tree
• ensemble

Unsupervised Learning

unsupervised learning

• neighbors.
  • KNeighborsClassifier(3, weights='distance')
• cluster
  • KMeans
• decomposition

Model Training and Model Selection

sklearn.

• model_selection
  • train_test_split(X, y, test_size, [random_state, stratify])
  • GridSearchCV
    • tune hyper parameters
• metrics
  • roc_curve, auc
  • confusion_matrix, precision_score, recall_score, classification_report

Pasty

Pasty

It is closely inspired by and compatible with the formula mini-language used in R and S

pasty.

• dmatrices('y~x0+x1[+0])
  • returns nd array with additional info
    • X.design_info
  • can use standardize(x), center(x), C(x)
    • C(x) - categorical data
      • treat like dummy variable automatically
• build_design_matrices(<design_info>, new data)

pasty objects can be taken directly to methods like

numpy.linalg

• lstsq

StatsModel

StatsModels include classical frequentists statistical models like

• Linear Models, generalized linear models
• Linear Mixed Effects Models
• Analysis of Variance methods
• Time Series Processing and State Space Models
• Generalized Methods of Moments

Basic Usage

statsmodels.api - array based model api

sm.

• add_constant()
• OLS
  • yields a model
• tsa

statsmodels.formula.api - formula (pasty-like) based model api

smf.

• ols

model.

• fit()
• predict()

Additional Libraries

Boosting Trees

XGBoost

LightGBM

Examples

• Examples of Gradient Boosting

H2O

H2O is a Java-based software for data modeling and general computing. The H2O software is many things, but the primary purpose of H2O is as a distributed (many machines), parallel (many CPUs), in memory (several hundred GBs Xmx) processing engine.

H2O algorithms

Vowpal Wabbit

Vowpal Wabbit