Apply backend.result_type to bincount, substract, matmul, multiply, mean and max
#18534
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
|
All tests passed except the failure in Codecov. |
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## master #18534 +/- ##
==========================================
+ Coverage 77.58% 77.67% +0.09%
==========================================
Files 334 334
Lines 32211 32302 +91
Branches 6286 6297 +11
==========================================
+ Hits 24990 25092 +102
+ Misses 5636 5631 -5
+ Partials 1585 1579 -6
Flags with carried forward coverage won't be shown. Click here to find out more.
☔ View full report in Codecov by Sentry. |
fchollet
reviewed
Oct 2, 2023
| dtype = getattr(x, "dtype", None) | ||
| if hasattr(dtype, "name") and "float" in dtype.name: | ||
| return cast(outputs, dtype) | ||
| compute_dtype = dtypes.result_type(x.dtype, "float32") |
There was a problem hiding this comment.
Shouldn't this be result_type(x.dtype, config.floatx()) rather than hardcoding float32?
There was a problem hiding this comment.
There is a note:
# `jnp.mean` does not handle low precision (e.g., float16) overflow
# correctly, so we compute with float32 and cast back to the original type.it should come from this PR: keras-team/keras-core#410
I have added a test to verify the overflow behavior:
# test overflow
x = np.array([65504, 65504, 65504], dtype="float16")
self.assertAllClose(knp.mean(x), np.mean(x))np.mean(x) |
jnp.mean(x) |
jnp.mean(x, dtype="float32") |
tfnp.mean(x) |
tfnp.mean(x, dtype="float32") |
torch.mean(x) |
torch.mean(x, dtype=torch.float32) |
|---|---|---|---|---|---|---|
| 65504 | inf | 65504 | inf | 65504 | inf | 65504 |
As a result, we should use float32 for jax, tensorflow and torch to compute mean, even if backend.floatx() == "float16"
keras/backend/numpy/numpy.py
Outdated
| return np.mean(x, axis=axis, keepdims=keepdims) | ||
| x = convert_to_tensor(x) | ||
| ori_dtype = standardize_dtype(x.dtype) | ||
| compute_dtype = dtypes.result_type(x.dtype, "float32") |
| return tfnp.mean(x, axis=axis, keepdims=keepdims) | ||
| x = convert_to_tensor(x) | ||
| ori_dtype = standardize_dtype(x.dtype) | ||
| compute_dtype = dtypes.result_type(x.dtype, "float32") |
| x = cast(x, "float32") if x.dtype in TORCH_INT_TYPES else x | ||
| return torch.mean(x, axis=axis, keepdims=keepdims) | ||
| ori_dtype = standardize_dtype(x.dtype) | ||
| compute_dtype = dtypes.result_type(x.dtype, "float32") |
fchollet
approved these changes
Oct 3, 2023
google-ml-butler
bot
added
kokoro:force-run
ready to pull
google-ml-butler
bot
removed
awaiting review
ready to pull
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.
Followed by #18482
This PR has applied
result_typeto the following ops:bincountsubstractmatmulmultiplymeanmaxThe corresponding unit tests have also been added.