Added normalization for predictions. #91
Conversation
There was a problem hiding this comment.
The aleatoric uncertainties would also need to be denormed
There was a problem hiding this comment.
I think line 108 handles that case as well, as in case the model is robust, preds will contain both (line 113) -
preds, aleat_log_std = preds.TThere was a problem hiding this comment.
Hi, I understand the problem you are pointing out, have added a fix.
aviary/predict.py
Outdated
| # denorm the mean and aleatoroc uncertainties separately | ||
| mean, log_std = np.split(preds, 2, axis=1) | ||
| preds = normalizer.denorm(mean) | ||
| ale_std = np.exp(log_std) * normalizer.std |
There was a problem hiding this comment.
we need to put this back to the log space here based on the logic below.
There was a problem hiding this comment.
I would think it would be less code to just add the normalizer into the logic below rather than having to make a new logic block.
|
Hi, I agree with both of your suggestions. I made the changes. |
|
thanks for the fix, can you confirm that this actually works as intended? |
|
I have been using this branch for my research and it gives predictions as expected, I will try one more thing as a sanity check today and post here when done. |
|
Hi, Sorry for the big delay! I got caught up in exams and travelling. Meanwhile I noticed that this code doesn't handle classification predictions. I have added code for that. I am performing the sanity check I was talking about, will raise another PR. The check to detect a normalizer was also incorrect, i.e., this one - if "normalizer_dict" in checkpoint: |
Fix for issue #90. I have only added a few lines of code in
predict.pyso that Roost models can be used for prediction later.