GAMI-Net Classification (Taiwan Credit)

Experiment initialization and data preparation

from piml import Experiment
from piml.models import GAMINetClassifier

exp = Experiment()
exp.data_loader(data="TaiwanCredit", silent=True)
exp.data_summary(feature_exclude=["LIMIT_BAL", "SEX", "EDUCATION", "MARRIAGE", "AGE"], silent=True)
exp.data_prepare(target="FlagDefault", task_type="classification", silent=True)

Train Model

exp.model_train(model=GAMINetClassifier(), name="GAMI-Net")

Train Model with monotonicity constraints on PAY_1

exp.model_train(model=GAMINetClassifier(mono_increasing_list=("PAY_1", )), name="Mono-GAMI-Net")

Evaluate predictive performance of GAMI-Net

exp.model_diagnose(model="GAMI-Net", show='accuracy_table')

          ACC      AUC      F1  LogLoss    Brier

Train  0.8169   0.7734  0.4465   0.4352   0.1366
Test   0.8228   0.7672  0.4478   0.4292   0.1334
Gap    0.0060  -0.0063  0.0012  -0.0060  -0.0032

Evaluate predictive performance of Mono-GAMI-Net

exp.model_diagnose(model="Mono-GAMI-Net", show='accuracy_table')

          ACC      AUC      F1  LogLoss    Brier

Train  0.8161   0.7713  0.4483   0.4373   0.1373
Test   0.8243   0.7650  0.4573   0.4310   0.1338
Gap    0.0082  -0.0063  0.0090  -0.0063  -0.0034

Global effect plot for PAY_1

exp.model_interpret(model="GAMI-Net", show="global_effect_plot", uni_feature="PAY_1",
                    original_scale=True, figsize=(5, 4))

Global effect plot for PAY_1 of Mono-GAMI-Net

exp.model_interpret(model="Mono-GAMI-Net", show="global_effect_plot", uni_feature="PAY_1",
                    original_scale=True, figsize=(5, 4))

Effect importance of Mono-GAMI-Net

exp.model_interpret(model="Mono-GAMI-Net", show="global_ei", figsize=(5, 4))

Feature importance of Mono-GAMI-Net

exp.model_interpret(model="Mono-GAMI-Net", show="global_fi", figsize=(5, 4))

Local interpretation by effect of Mono-GAMI-Net

exp.model_interpret(model="Mono-GAMI-Net", show="local_ei", sample_id=0, original_scale=True, figsize=(5, 4))

Local interpretation by feature of Mono-GAMI-Net

exp.model_interpret(model="Mono-GAMI-Net", show="local_fi", sample_id=0, original_scale=True, figsize=(5, 4))