Fix automl handling of positional initializer args #373
Conversation
|
|
||
| AUTOML_SETTINGS_REG = { | ||
| 'experiment_timeout_minutes': 1, | ||
| 'experiment_timeout_minutes': 15, |
There was a problem hiding this comment.
15 minutes is the new lowest allowed value for the experiment timeout.
| var = self._get_automated_ml_model(kwarg, key) | ||
| new_args += (var,) | ||
| if isinstance(arg, EconAutoMLConfig): | ||
| arg = self._get_automated_ml_model(arg, f"arg{idx}") |
There was a problem hiding this comment.
What is the prefix actually used for?
There was a problem hiding this comment.
and what if "arg0" is also a keyword argument?
There was a problem hiding this comment.
I believe it is just used to make it easier for a person to identify the experiment when looking through experiments in AzureML studio, and that the service itself adds a unique suffix so that the whole thing will be unambiguous in any case.
| min_samples_leaf=10, | ||
| min_impurity_decrease=0.001, | ||
| verbose=0, min_weight_fraction_leaf=.01) | ||
| est = AutomatedCausalForestDML(model_y=automl_model_reg(), |
There was a problem hiding this comment.
where did the data generation go?
There was a problem hiding this comment.
We were already performing the data generation in the big try block up top, and we made use of this fact in one test but not the others, which I found confusing.
Fixes #371.
Also fixes a minor drlearner issue.