Thomas J. Fan

I think `raw_predict` makes sense. I'm unsure of adding expose the loss objects through `estimator.loss`. Can the link functions be a regular functions and in the docstring of `raw_predict` we...

I think adding three new methods estimators is pretty big API update. I'm trying to find a path to get some momentum, like standardizing on `raw_predict`. After `raw_predict` is out,...

In the short term, I'm thinking of having the inverse link function in the docstring, so the `raw_predict` is meaningful: ```python def raw_predict(self, X): """Return the raw prediction. `predict`'s output...

This goes down the line of "how to add additional arguments that is builder specific". The API options I see are: ``` # Your proposal image_spec = ImageSpec(builder="default", builder_config=DefaultBuilderConfig(...)) #...

I like builder object with `builder=` the most: ```python image_spec = ImageSpec(builder=DefaultBuilder(buildx_builder="...")) ``` Mostly because we do not need to handle the two edge cases: ```python # incompatible builder image_spec...

We'll have to error if someone writes: ```python image_spec = ImageSpec( builder="envd", platform="linux/amd64,linux/arm64", ... ) ``` > Could we just use buildx by default if a user specifies two architectures...

> envd already supports multi-arch, no? If we add multiarch support to the `envd` builder, then it's okay. > Why shouldn't they write the builder name? I'm okay with them...

> An alternative would be to extract _validate_data from the estimator, move it to utils, and have users call validate_data(estimator, X, y, ...) which mutates the estimator itself. WDYT? This...

@ebsmothers Thanks for catching this! There was a bug in my script for finding the recipes that support gradient clipping. I updated this PR to fix the issue.

Are there any papers that study using the RandomForest's OOB for cross validation?