Autoencoder/Unsupervised learning stopping_metric automatically selected as "deviance"

[exalate-issue-sync]

There is a bug in the automatic selection of stopping criteria based on problem type (unsupervised/regression/...).

In the current version, it selects "deviance" which doesn't apply to unsupervised and it never stops.
It should choose MSE as it used to in the prior versions.

The workaround is to specify stopping criteria explicitly to MSE:

ae_model = H2OAutoEncoderEstimator(activation="Tanh",
hidden=[2],
model_id="ae_model",
epochs=2000,
ignore_const_cols=False,
stopping_metric="MSE",
seed=1234)

[h2o-ops-ro]

JIRA Issue Details

Jira Issue: PUBDEV-8537
Assignee: Michal Kurka
Reporter: Arun Aryasomayajula
State: Resolved
Fix Version: 3.36.0.2
Attachments: N/A
Development PRs: Available

[h2o-ops-ro]

Linked PRs from JIRA

#6021