Tree Ensemble Models (Classification)

Installation

# To install the required package, use the following command:
# !pip install modeva

Authentication

# To get authentication, use the following command: (To get full access please replace the token to your own token)
# from modeva.utils.authenticate import authenticate
# authenticate(auth_code='eaaa4301-b140-484c-8e93-f9f633c8bacb')

Import required modules

from modeva import DataSet
from modeva import TestSuite
from modeva.models import (MoLGBMClassifier,
                           MoXGBClassifier,
                           MoCatBoostClassifier,
                           MoGradientBoostingClassifier,
                           MoRandomForestClassifier)

Load and prepare dataset

ds = DataSet()
ds.load(name="TaiwanCredit")
ds.set_random_split()
ds.set_target("FlagDefault")

Train model

You may replace the model by anyone of the following, including MoGradientBoostingClassifier, MoRandomForestClassifier, MoXGBClassifier, MoCatBoostClassifier

model = MoLGBMClassifier(max_depth=2, verbose=-1, random_state=0)
model.fit(ds.train_x, ds.train_y.ravel())

MoLGBMClassifier(boosting_type='gbdt', class_weight=None, colsample_bytree=1.0,
                 importance_type='split', learning_rate=0.1, max_depth=2,
                 min_child_samples=20, min_child_weight=0.001,
                 min_split_gain=0.0, n_estimators=100, n_jobs=None,
                 num_leaves=31, objective=None, random_state=0, reg_alpha=0.0,
                 reg_lambda=0.0, subsample=1.0, subsample_for_bin=200000,
                 subsample_freq=0, verbose=-1)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Basic accuracy analysis

ts = TestSuite(ds, model)
results = ts.diagnose_accuracy_table()
results.table

	AUC	ACC	F1	LogLoss	Precision	Recall	Brier
train	0.7896	0.8205	0.4726	0.4255	0.6825	0.3615	0.1334
test	0.7803	0.8300	0.4812	0.4216	0.7070	0.3647	0.1311
GAP	-0.0093	0.0095	0.0085	-0.0039	0.0246	0.0032	-0.0023

Feature importance analysis

results = ts.interpret_fi()
results.plot(n_bars=10)

Effect importance analysis

results = ts.interpret_ei()
results.plot(n_bars=10)

Local feature importance analysis

results = ts.interpret_local_fi(dataset='train', sample_index=1, centered=True)
results.plot(n_bars=10)

Local effect importance analysis

results = ts.interpret_local_ei(dataset='train', sample_index=1)
results.plot(n_bars=10)

Main effect plot

For numerical feature

results = ts.interpret_effects(features="PAY_1")
results.plot()

Extract the detail information of the effect, e.g., split points and values.

results.value["Details"]

{'PAY_1': {'fidx': (5,), 'values': array([-1.0534514 , -1.16997403, -0.61438626,  0.38886216,  0.33472527,
        0.24835294]), 'splits': [array([           -inf, -1.00000002e-35,  1.00000002e-35,  1.50000000e+00,
        3.50000000e+00,  4.50000000e+00,             inf])], 'right_inclusive': True}, 'PAY_6': {'fidx': (10,), 'values': array([-0.01286074,  0.0764496 , -0.05644409]), 'splits': [array([          -inf, 1.00000002e-35, 3.50000000e+00,            inf])], 'right_inclusive': True}, 'LIMIT_BAL': {'fidx': (0,), 'values': array([ 0.32395616,  0.2880562 ,  0.23693366,  0.24081257,  0.1242532 ,
        0.11563613,  0.01055193, -0.02299061, -0.03081845, -0.05087281]), 'splits': [array([   -inf,  25000.,  45000.,  75000.,  95000., 125000., 145000.,
       155000., 285000., 305000.,     inf])], 'right_inclusive': True}, 'BILL_AMT1': {'fidx': (11,), 'values': array([-0.03576016, -0.00935902,  0.09906073,  0.0790521 ,  0.13835849,
        0.10284346,  0.14139967,  0.12624014,  0.10861737,  0.07145943,
        0.04787676,  0.04199568,  0.02953576,  0.00354041,  0.01641528,
        0.05408157,  0.0590748 ]), 'splits': [array([           -inf, -1.00000002e-35,  2.77488155e+00,  2.89376030e+00,
        2.99629270e+00,  3.02180920e+00,  3.34879130e+00,  3.84176620e+00,
        3.93374030e+00,  3.94615745e+00,  4.18843615e+00,  4.21968930e+00,
        4.27368400e+00,  4.29991030e+00,  5.03414900e+00,  5.04367255e+00,
        5.40019150e+00,             inf])], 'right_inclusive': True}, 'BILL_AMT2': {'fidx': (12,), 'values': array([ 0.10091232, -0.04900599, -0.00214606,  0.04257239,  0.07686822]), 'splits': [array([       -inf, -2.0588013 ,  4.6097598 ,  5.20135065,  5.3489307 ,
               inf])], 'right_inclusive': True}, 'BILL_AMT3': {'fidx': (13,), 'values': array([-0.04677684, -0.01265278, -0.04687064, -0.03267184,  0.03566952,
        0.04341139,  0.08114306,  0.09737877,  0.12260411,  0.15647769,
        0.27079123,  0.30592314]), 'splits': [array([          -inf, 1.00000002e-35, 4.19129730e+00, 4.27521925e+00,
       4.67514110e+00, 4.77373500e+00, 4.77944850e+00, 5.04415915e+00,
       5.33446600e+00, 5.35396525e+00, 5.39070830e+00, 5.51115655e+00,
                  inf])], 'right_inclusive': True}, 'PAY_AMT1': {'fidx': (17,), 'values': array([ 0.07084209,  0.04034364,  0.02922558,  0.00996485, -0.01358633,
       -0.00676261, -0.0349187 ,  0.07045835,  0.01075141, -0.02214641,
       -0.04500528, -0.0578455 , -0.07412234,  0.0728663 ,  0.01136341,
       -0.0287415 , -0.06786967]), 'splits': [array([          -inf, 1.00000002e-35, 8.74094000e-01, 2.94448235e+00,
       3.00881300e+00, 3.17623590e+00, 3.20452690e+00, 3.26611415e+00,
       3.47719370e+00, 3.66048595e+00, 3.69050610e+00, 3.75308490e+00,
       3.86308480e+00, 4.19509665e+00, 4.38260215e+00, 4.40198600e+00,
       4.83255660e+00,            inf])], 'right_inclusive': True}, 'PAY_AMT2': {'fidx': (18,), 'values': array([ 0.14818567,  0.0977724 ,  0.07496113,  0.04027566,  0.03114211,
        0.00813082, -0.02643381, -0.15991107, -0.18238135, -0.24355615]), 'splits': [array([      -inf, 3.1765252 , 3.2149762 , 3.30135555, 3.69910025,
       3.77103605, 4.00889805, 4.14935775, 4.17617815, 4.2861982 ,
              inf])], 'right_inclusive': True}, 'PAY_AMT3': {'fidx': (19,), 'values': array([ 0.08127522,  0.07458042,  0.05490845,  0.0198504 ,  0.01355191,
        0.00178131, -0.01815329, -0.02270906, -0.0345931 , -0.06555223,
       -0.08572601]), 'splits': [array([          -inf, 1.00000002e-35, 2.65801040e+00, 2.83600735e+00,
       2.90417395e+00, 2.93043925e+00, 2.95640850e+00, 3.38810125e+00,
       3.39820050e+00, 3.65340550e+00, 3.84522210e+00,            inf])], 'right_inclusive': True}, 'EDUCATION': {'fidx': (2,), 'values': array([-0.45725397,  0.00387936, -0.01658904]), 'splits': [array([          -inf, 1.00000002e-35, 2.50000000e+00,            inf])], 'right_inclusive': True}, 'PAY_AMT4': {'fidx': (20,), 'values': array([ 0.05424298,  0.02541088, -0.00441816, -0.03026423]), 'splits': [array([          -inf, 1.00000002e-35, 3.00065090e+00, 3.31249470e+00,
                  inf])], 'right_inclusive': True}, 'MARRIAGE': {'fidx': (3,), 'values': array([ 0.0423956 , -0.12718679]), 'splits': [array([-inf,  1.5,  inf])], 'right_inclusive': True}, 'AGE': {'fidx': (4,), 'values': array([-0.08920078, -0.03964159,  0.01700743,  0.04028322]), 'splits': [array([-inf, 35.5, 39.5, 45.5,  inf])], 'right_inclusive': True}, 'PAY_5': {'fidx': (9,), 'values': array([-0.15075029,  0.01901355]), 'splits': [array([          -inf, 1.00000002e-35,            inf])], 'right_inclusive': True}, 'BILL_AMT4': {'fidx': (14,), 'values': array([-0.01398571,  0.02474484,  0.03286639,  0.04270729,  0.05822533,
        0.08875921,  0.18554294]), 'splits': [array([      -inf, 4.09817635, 4.69402145, 4.7208288 , 5.1276295 ,
       5.33973635, 5.4496842 ,        inf])], 'right_inclusive': True}, 'SEX': {'fidx': (1,), 'values': array([ 0.04318783, -0.04318783]), 'splits': [array([-inf,  1.5,  inf])], 'right_inclusive': True}, 'BILL_AMT5': {'fidx': (15,), 'values': array([-0.00471474, -0.02418091, -0.00817011,  0.13564104]), 'splits': [array([     -inf, 3.8332428, 4.9599828, 5.390088 ,       inf])], 'right_inclusive': True}, 'BILL_AMT6': {'fidx': (16,), 'values': array([-0.01883298,  0.04975605,  0.05801912]), 'splits': [array([      -inf, 2.89070025, 4.477859  ,        inf])], 'right_inclusive': True}, 'PAY_AMT6': {'fidx': (22,), 'values': array([ 0.04949369,  0.09821501,  0.07463474,  0.01185624, -0.05536766,
       -0.17904737, -0.23254459]), 'splits': [array([      -inf, 0.874094  , 2.9355069 , 3.98238435, 3.99578865,
       4.297246  , 4.7525315 ,        inf])], 'right_inclusive': True}, 'PAY_2': {'fidx': (6,), 'values': array([-0.07558263,  0.1635372 ,  0.09918538, -0.02073921]), 'splits': [array([-inf,  1.5,  2.5,  3.5,  inf])], 'right_inclusive': True}, 'PAY_3': {'fidx': (7,), 'values': array([-0.26863568, -0.19129001,  0.13112096,  0.06531028]), 'splits': [array([           -inf, -1.00000002e-35,  1.50000000e+00,  3.50000000e+00,
                   inf])], 'right_inclusive': True}, 'PAY_AMT5': {'fidx': (21,), 'values': array([-0.01371426,  0.05253822]), 'splits': [array([     -inf, 4.4334417,       inf])], 'right_inclusive': True}, 'PAY_4': {'fidx': (8,), 'values': array([-0.1752667 , -0.12244993,  0.02204623]), 'splits': [array([           -inf, -1.00000002e-35,  1.00000002e-35,             inf])], 'right_inclusive': True}}

Main effect plot for categorical feature

For categorical feature

results = ts.interpret_effects(features="EDUCATION")
results.plot()

For 2 features

results = ts.interpret_effects(features=("PAY_1", "PAY_2"))
results.plot()