Random 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 MoElasticNet
from modeva.models import ModelTuneRandomSearch

Load Dataset

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

ds.scale_numerical(features=("cnt",), method="log1p")
ds.preprocess()

Run random search

param_grid = {"alpha": [0.1, 1.0, 10],
              "l1_ratio": [(i + 1) * 0.1 for i in range(10)]}

model = MoElasticNet()
hpo = ModelTuneRandomSearch(dataset=ds, model=model)
result = hpo.run(param_distributions=param_grid,
                 n_iter=20,
                 metric="MSE",
                 cv=5)
result.table

	alpha	l1_ratio	MSE	MSE_rank	Time
16	0.1	0.2	1.2240	1	0.0053
0	0.1	0.3	1.2558	2	0.0053
19	0.1	0.5	1.3430	3	0.0054
10	0.1	0.6	1.3890	4	0.0052
18	0.1	0.7	1.4156	5	0.0054
12	0.1	0.9	1.4405	6	0.0053
11	1.0	0.7	1.4520	7	0.0087
8	1.0	0.8	1.4528	8	0.0045
2	1.0	0.4	1.4531	9	0.0045
13	1.0	0.5	1.4539	10	0.0045
3	1.0	0.1	1.4540	11	0.0049
7	1.0	0.2	1.4568	12	0.0046
15	10.0	0.1	1.4994	13	0.0045
9	10.0	0.3	1.6682	14	0.0045
14	10.0	0.4	1.8006	15	0.0046
5	10.0	0.5	1.9698	16	0.0044
6	10.0	0.8	2.0789	17	0.0045
4	10.0	0.7	2.0789	17	0.0044
1	10.0	0.9	2.0789	17	0.0044
17	10.0	1.0	2.0789	17	0.0044

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

result.plot(("alpha", "MSE"))

result.plot(("l1_ratio", "MSE"))

Retrain model with best hyperparameter

model_tuned = MoElasticNet(**result.value["params"][0],
                           name="GLM-Tuned")
model_tuned.fit(ds.train_x, ds.train_y)
model_tuned

MoElasticNet(alpha=0.1, copy_X=True, fit_intercept=True,
             l1_ratio=0.30000000000000004, max_iter=1000, positive=False,
             precompute=False, random_state=None, selection='cyclic',
             tol=0.0001, warm_start=False)

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

	MSE	MAE	R2
train	1.1690	0.8863	0.4175
test	1.1805	0.8876	0.4166
GAP	0.0115	0.0013	-0.0009