GAM creates multiple knots at the same coordinates

[exalate-issue-sync]

When quantile-based knots generation (currently equidistant) is used together with an ordinal value of low cardinality (10-12), multiple knots (at least two) might hit he same value.

{noformat}import h2o
import numpy as np
from h2o.estimators.gam import H2OGeneralizedAdditiveEstimator
h2o.init()

load and prepare California housing dataset

np.random.seed(1234)
data = h2o.H2OFrame(
python_obj={'C1': list(np.random.randint(0, 9, size=1000)),
'target': list(np.random.randint(0, 2, size=1000))
})

use only 3 features and transform into classification problem

feature_names = ['C1']
data['target'] = data['target'].asfactor()

split into train and validation sets

train, test = data.split_frame([0.8], seed=1234)

build the GAM model

h2o_model = H2OGeneralizedAdditiveEstimator(family='binomial',
gam_columns=feature_names,
scale=[1],
num_knots=[10],
)
h2o_model.train(x=feature_names, y='target', training_frame=train){noformat}

!image (1).png|width=2068,height=1288!

Discovered by [~accountid:5f484754fe0a46004dc38d32]

[h2o-ops]

JIRA Issue Migration Info

Jira Issue: PUBDEV-7798
Assignee: Karthik Murthy
Reporter: Pavel Pscheidl
State: Resolved
Fix Version: 3.32.0.5
Attachments: Available (Count: 1)
Development PRs: Available

Linked PRs from JIRA

#5347
#5397

Attachments From Jira

Attachment Name: image (1).png
Attached By: Pavel Pscheidl
File Link:https://h2o-3-jira-github-migration.s3.amazonaws.com/PUBDEV-7798/image (1).png