fast-explain

FastExplain

Fit Fast, Explain Fast

Installing

pip install fast-explain

Clean Data, Fit ML Models and Explore Results all in one line.

FastExplain provides an out-of-the-box tool for analysts to quickly explore data, train and interpret models, with flexibility to fine-tune if needed.

• Automated cleaning and fitting of machine learning models with hyperparameter search
• Aesthetic display of explanatory methods ready for reporting
• Connected interface for all data, models and related explanatory methods

Quickstart

Automated Cleaning and Fitting

from FastExplain import *
df = load_titanic_data()
classification = model_data(df, dep_var="Survived", model="ebm")

Aesthetic Display

feature_correlation(classification.data.df)

plot_one_way_analysis(classification.data.df, "Age", "Survived", filter = "Sex == 1")

plot_ebm_explain(classification.m, classification.data.df, "Age")

plot_ale(classification.m, classification.data.xs, "Age", filter = "Sex == 1", dep_name = "Survived")

classification_1 = model_data(df, dep_var="Survived", model="rf", hypertune=True, cont_names=['Age'], cat_names = [], hypertune=True)
models = [classification.m, classification_1.m]
data = [classification.data.xs, classification_1.data.xs]
plot_ale(models, data, 'Age', dep_name = "Survived")

Connected Interface

classification_1.plot_one_way_analysis("Age", filter = "Sex == 1")
classification_1.plot_ale("Age", filter = "Sex == 1")

classification_1.shap_dependence_plot("Age", filter = "Sex == 1")

classification_1.error
# {'auc': {'model': {'train': 0.9934332941166654,
# 'val': 0.8421607378129118,
# 'overall': 0.9665739941840028}},
# 'cross_entropy': {'model': {'train': 0.19279692001978943,
# 'val': 0.4600233891109683,
# 'overall': 0.24648214781700722}}}

Models Supported

• Random Forest
• XGBoost
• Explainable Boosting Machine
• ANY Model Class with fit and predict attributes

pip install lightgbm

from lightgbm import LGBMClassifier
custom_model = model_data(df, 'Survived', model=LGBMClassifier)
custom_model.plot_ale("Age")
custom_model.shap_dependence_plot("Age")

Exploratory Methods Supported:

• One-way Analysis
• Two-way Analysis
• Feature Importance Plots
• ALE Plots
• Explainable Boosting Methods
• SHAP Values
• Partial Dependence Plots
• Sensitivity Analysis