Add objective base classes to API reference #736
Conversation
Codecov Report
@@ Coverage Diff @@
## master #736 +/- ##
=======================================
Coverage 99.29% 99.29%
=======================================
Files 140 140
Lines 4981 4981
=======================================
Hits 4946 4946
Misses 35 35
Continue to review full report at Codecov.
|
| :template: class.rst | ||
| :nosignatures: | ||
|
|
||
| ObjectiveBase |
There was a problem hiding this comment.
Looks like ObjectiveBase is missing the classmethod+property+abstractmethod fields: name, greater_is_better, score_needs_proba
@jeremyliweishih you were talking about this last week, right? In relation to the classproperty decorator (even though we don't use that in ObjectiveBase). Do we have an issue filed for that yet?
There was a problem hiding this comment.
Not exactly the same issue - to fix this problem it needs templating similar to whats in pipeline_class.rst and other templates with class properties. This will show the attribute and value but not the docstring. The issue with classproperty is that we can't access the __doc__ attribute which holds the doc strings. #652 is tracking that.
There was a problem hiding this comment.
Ah, got it. Here's pipeline_class.rst. So we'd have to create an objective_class.rst which mirrors that but with the new field names?
There was a problem hiding this comment.
yeah! I was trying to find a more modular solution but I couldn't find one so the best we have now is to create an additional template.
There was a problem hiding this comment.
@jeremyliweishih Hmm, I'm trying some things out right now, but I can't get what we did for pipeline_class.rst to work in this case, since that shows the attribute and value but not the docstring... yet these functions only have docstrings?
ex of function:
@classmethod
@abstractmethod
def name(cls):
"""Returns a name describing the objective."""
This is similar to component_graph on PipelineBase, which I noticed we also don't expose right now?
There was a problem hiding this comment.
oh yes you're correct! #652 tracks to fix that but I haven't found a good solution yet. Could you can add to that issue to show that we need to add for this as well? I’ll try to take a look when I can!
There was a problem hiding this comment.
Yah, will do! In this case @jeremyliweishih, @dsherry I'm going to address standardizing our language around problem_type, merge this PR, and file an issue to track the missing fields, and then link to #652 :)
Edit: filed #738 for this!
There was a problem hiding this comment.
Ah dang, got it. Sounds good!
| :nosignatures: | ||
|
|
||
| ObjectiveBase | ||
| BinaryClassificationObjective |
There was a problem hiding this comment.
BinaryClassificationObjective is missing can_optimize_threshold
| ObjectiveBase | ||
| BinaryClassificationObjective | ||
| MulticlassClassificationObjective | ||
| RegressionObjective |
There was a problem hiding this comment.
I noticed binary/multi/reg each have different text in the docs for problem_type. Can we standardize that? I do like how in RegressionObjective we mention that its set to ProblemTypes.REGRESSION, so that's what I'd go with.
There was a problem hiding this comment.
@angela97lin thanks for doing this! This is great.
I left one comment about standardizing our language around problem_type.
Also, there are some fields missing from ObjectiveBase and BinaryClassificationObjective. I know @jeremyliweishih was dealing with a similar issue in the pipeline doc recently so maybe he has pointers. My suggestion would be to merge this PR and file an issue to track the missing fields.
Update: Jeremy posted his advice on how to add the missing fields, so if you decide to add that in this PR please ping me and I'll re-review!
Closes #714
Updated docs here: https://evalml.featurelabs.com/en/714_obj_api/api_reference.html#objective-functions