Add random forest option for partial dependence #13
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
djgagne
requested changes
Jun 22, 2023
There was a problem hiding this comment.
@katiedagon Thank you for adding the random forest functionality. I added a couple of small suggestions that should address the remaining issues with negative output scores.
| @@ -27,6 +28,7 @@ def partial_dep(fn, input_cols, output_col, verbose=0): | |||
| df[param] = le.fit_transform(df[param]) | |||
|
|
|||
| objective = "reg:squarederror" | |||
There was a problem hiding this comment.
Instead of hard coding the model hyperparameters, I would put them in a dictionary called hparams and make that a keyword argument for partial_dep.
There was a problem hiding this comment.
Sounds good - where should I define hparams?
djgagne
approved these changes
Jul 3, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
@djgagne @jsschreck Following our discussion on the MILES slack about the behavior of XGBoost sometimes generating negative values in partial dependence plots, I've attempted to update the code to add a random forest option. I set it up to use RF as the default, but that could be changed. I tested this with an existing echo run, and the PD plots are comparable to those produced with XGB, but still generating negative values which I'm not sure makes sense for these metrics. Here are some example comparisons using a set of metrics from @mariajmolina - noting the vcorr_cust metric is likely out of date, but the valid_mae and vmse_extreme_outp should be relevant (and I believe should not have negative values).
I welcome comments/suggestions on any of the above, plus how best to let users choose between the two options, in addition to general comments about how the RF model is configured (e.g., hyperparameters).