Add a topbasemodel attribute to AutoML

[exalate-issue-sync]

I wonder if topbasemodel should be a property like leader in AutoML objects? it would handy to have that available… Then i could write code like:

{code:r}h2o.varimp_plot(aml@best_basemodel){code}

It's currently very overly complicated to grab the top basemodel id so you can use it. It would be handy to have especially if people dont want SEs for prod, but like to train them anyway as part of the AutoML process.

We can also consider adding a slot for the top of each algo class. Or perhaps a more general object called best_of_family where we have the top model from each class? Here’s how awkward it currently is to get the best GBM model:

{noformat}best_gbm = h2o.get_model(aml.leaderboard[int(aml.leaderboard["model_id"].grep("GBM")[0,:]), "model_id"]){noformat}

If we decide to add {{aml@best_gbm}} and {{aml@best_drf}} etc, we should probably also call the top model slot, {{aml@best_basemodel}} instead of “top”, for consistency.

[exalate-issue-sync]

Erin LeDell commented: Final API looked like this:

Python:

{code:python}aml.get_best_model()
aml.get_best_model(algorithm="gbm")
aml.get_best_model(criterion="auc")
aml.get_best_model(algorithm="gbm", criterion="auc")
aml.get_best_model(criterion="predict_time_per_row_ms"){code}

R:

{code:r}h2o.get_best_model(object = aml)
h2o.get_best_model(object = aml, algorithm = "gbm")
h2o.get_best_model(object = aml, criterion = "auc")
h2o.get_best_model(object = aml, algorithm = "gbm", criterion = "auc")
h2o.get_best_model(object = aml, criterion = "predict_time_per_row_ms"){code}

[h2o-ops]

JIRA Issue Migration Info

Jira Issue: PUBDEV-7820
Assignee: Tomas Fryda
Reporter: Erin LeDell
State: Resolved
Fix Version: 3.32.1.1
Attachments: N/A
Development PRs: Available

Linked PRs from JIRA

#5259