Particle Swarm Optimization Search

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
from modeva.models import ModelTunePSO

Load Dataset

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

Run PSO search

param_bounds = {"max_depth": [1, 4],
                "learning_rate": [0.01, 1.0]}
param_types = {"max_depth": "int"}

model = MoLGBMClassifier(verbose=-1)
hpo = ModelTunePSO(dataset=ds, model=model)
result = hpo.run(param_bounds=param_bounds,
                 param_types=param_types,
                 n_iter=2,
                 n_particles=10,
                 metric=("AUC", "LogLoss"),
                 cv=5)
result.table

	max_depth	learning_rate	AUC	LogLoss	AUC_rank	LogLoss_rank	mean_fit_time
38	1	0.8343	0.7751	0.4367	1	1	0.4229
18	1	0.8343	0.7751	0.4367	1	1	0.6503
8	1	0.8343	0.7751	0.4367	1	1	0.2331
27	1	0.6666	0.7748	0.4368	4	4	0.1835
32	2	0.6494	0.7706	0.4407	5	8	0.2291
2	2	0.6494	0.7706	0.4407	5	8	1.3934
12	2	0.6494	0.7706	0.4407	5	8	0.2840
17	1	0.0963	0.7696	0.4401	8	5	0.2097
37	1	0.0963	0.7696	0.4401	8	5	0.1872
7	1	0.0963	0.7696	0.4401	8	5	0.1762
15	3	0.5336	0.7655	0.4479	11	11	0.6640
35	3	0.5336	0.7655	0.4479	11	11	0.2572
5	3	0.5336	0.7655	0.4479	11	11	3.0980
11	3	0.5494	0.7641	0.4494	14	14	0.3544
1	3	0.5494	0.7641	0.4494	14	14	2.3557
31	3	0.5494	0.7641	0.4494	14	14	0.2706
14	4	0.3896	0.7641	0.4504	17	20	2.2246
34	4	0.3896	0.7641	0.4504	17	20	0.4668
4	4	0.3896	0.7641	0.4504	17	20	3.0105
33	2	0.8929	0.7633	0.4497	20	17	0.1539
3	2	0.8929	0.7633	0.4497	20	17	1.4171
13	2	0.8929	0.7633	0.4497	20	17	0.3990
26	2	1.0	0.7632	0.4510	23	23	0.1985
22	2	1.0	0.7632	0.4510	23	23	2.3360
23	2	1.0	0.7632	0.4510	23	23	0.4037
25	2	1.0	0.7632	0.4510	23	23	0.1830
28	2	1.0	0.7632	0.4510	23	23	0.5747
24	3	0.7842	0.7559	0.4676	28	31	0.2910
0	3	0.718	0.7557	0.4634	29	28	2.1687
10	3	0.718	0.7557	0.4634	29	28	1.0073
30	3	0.718	0.7557	0.4634	29	28	0.9453
19	3	0.8713	0.7512	0.4758	32	32	1.7386
9	3	0.8713	0.7512	0.4758	32	32	0.6181
39	3	0.8713	0.7512	0.4758	32	32	0.3238
21	3	1.0	0.7464	0.4915	35	38	8.1627
20	3	1.0	0.7464	0.4915	35	38	2.2119
16	3	0.9263	0.7433	0.4889	37	35	0.8553
36	3	0.9263	0.7433	0.4889	37	35	0.9300
6	3	0.9263	0.7433	0.4889	37	35	0.6379
29	4	1.0	0.7256	0.5610	40	40	0.7462

result.plot("parallel", figsize=(8, 6))

result.plot(("max_depth", "AUC"))

result.plot(("learning_rate", "AUC"))

Retrain model with best hyperparameter

model_tuned = MoLGBMClassifier(**result.value["params"][0],
                             name="LGBM-Tuned",
                             verbose=-1)
model_tuned.fit(ds.train_x, ds.train_y)
model_tuned

MoLGBMClassifier(boosting_type='gbdt', class_weight=None, colsample_bytree=1.0,
                 importance_type='split', learning_rate=0.7180374727086953,
                 max_depth=3, 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=None,
                 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.

Diagnose the tuned model

ts = TestSuite(ds, model_tuned)
result = ts.diagnose_accuracy_table()
result.table

	AUC	ACC	F1	LogLoss	Brier
train	0.8555	0.8412	0.5539	0.3697	0.1157
test	0.7646	0.8158	0.4633	0.4441	0.1379
GAP	-0.0909	-0.0254	-0.0906	0.0744	0.0222