Data Dependent Explanation

Experiment initialization and data preparation

from piml import Experiment
from piml.models import XGB2Classifier

exp = Experiment()
exp.data_loader(data="SimuCredit", silent=True)
exp.data_summary(feature_exclude=["Gender", "Race"], silent=True)
exp.data_prepare(target="Approved", task_type="classification", silent=True)

Train Model

exp.model_train(model=XGB2Classifier(n_estimators=100), name="XGB2")

PFI with training data (by default)

exp.model_explain(model="XGB2", show="pfi", figsize=(5, 4))

PFI with testing data

exp.model_explain(model="XGB2", show="pfi", use_test=True, figsize=(5, 4))

H-statistics with training data (use training data by default)

exp.model_explain(model="XGB2", show="hstats",
                  grid_size=5, figsize=(5, 4))

H-statistics with testing data

exp.model_explain(model="XGB2", show="hstats", use_test=True,
                  grid_size=5, figsize=(5, 4))

PDP with training data (use training data by default)

exp.model_explain(model="XGB2", show="pdp", uni_feature="Balance",
                  grid_size=50, figsize=(5, 4))

PDP with testing data

exp.model_explain(model="XGB2", show="pdp", uni_feature="Balance",
                  grid_size=50, use_test=True, figsize=(5, 4))

ICE with training data (use training data by default)

exp.model_explain(model="XGB2", show="ice", uni_feature="Balance",
                  figsize=(5, 4))

ICE with testing data

exp.model_explain(model="XGB2", show="ice", uni_feature="Balance",
                  use_test=True, figsize=(5, 4))

ALE with training data (use training data by default)

exp.model_explain(model="XGB2", show="ale", uni_feature="Balance",
                  grid_size=50, figsize=(5, 4))

ALE with testing data

exp.model_explain(model="XGB2", show="ale", uni_feature="Balance",
                  grid_size=50, use_test=True, figsize=(5, 4))

LIME on training data (use training data by default)

exp.model_explain(model="XGB2", show="lime", sample_id=0, figsize=(5, 4))

LIME on testing data

exp.model_explain(model="XGB2", show="lime", sample_id=0, use_test=True, figsize=(5, 4))

SHAP on training data (use training data by default)

exp.model_explain(model="XGB2", show="shap_waterfall", sample_id=0, figsize=(5, 4))

SHAP on testing data

exp.model_explain(model="XGB2", show="shap_waterfall", sample_id=0, use_test=True, figsize=(5, 4))

Estimated memory usage: 70 MB