h2o/sklearn/wrapper.py#params_as_h2o_frames fails to pass on **fit_params correctly to decorated function

[exalate-issue-sync]

When calling the fit method of a class like h2o.sklearn.H2OAutoMLRegressor and passing not only standard params X and y but also additional non-standard custom params which are bound to variable keyword param **fit_params, those fit_params will not be passed on correctly to the actual fit function by the fit function's decorator that has been created beforehand by h2o/sklearn/wrapper.py#params_as_h2o_frames. Instead those fit_params will end up one level too deeply nested in an envelope of type dictionary so that the decorated fit function cannot bind them anymore.

I have been able to fix the problem in my copy of the code by putting the following snippet into function params_as_h2o_frames:

{{
# PATCH BEGIN
# this patch is to ensure that custom keyword params bound to a var keyword parameter like fit_params
# are forwarded to the wrapped function fn in an unpacked way rather than as items of a dictionary
# which would not be recognised by the function fn
# result = fn(_args) # this is the original line of code being replaced by the modified version to follow
var_keyword_param = {}
for param_name in sig.parameters.keys():
if sig.parameters[param_name].kind == Parameter.VAR_KEYWORD:
var_keyword_param = _args.pop(param_name)
break # there can only be at most one parameter of kind VAR_KEYWORD in any function signature
result = fn(**_args, **var_keyword_param)
# PATCH END
}}

[h2o-ops-ro]

JIRA Issue Details

Jira Issue: PUBDEV-8522
Assignee: Sebastien Poirier
Reporter: N/A
State: Resolved
Fix Version: 3.36.1.3
Attachments: N/A
Development PRs: Available

[h2o-ops-ro]

Linked PRs from JIRA

#6189
#6205