[METRICS] Various improvements on metrics #466
Conversation
|
The cast function is now called inside Moreover I used the cast function in There are tests for tensors inputs in metrics and in .map |
| @@ -89,6 +90,38 @@ | |||
| INCOMPLETE_SUFFIX = ".incomplete" | |||
|
|
|||
|
|
|||
| @contextmanager | |||
| def temp_seed(seed: int, set_pytorch=False, set_tensorflow=False): | |||
There was a problem hiding this comment.
Here there is an error with the tensorflow seed setter that I didn't have time to debug before my time off, maybe you can take a look? You can see the error simply by running the colab at https://colab.research.google.com/drive/1I_B1mcX0cOzOskr0rJN8u8xh_0CIken-?usp=sharing and using set_tensorflow=True in its call of temp_seed.
There was a problem hiding this comment.
I fixed tensorflow's rng for eager mode.
Not sure how to do it when eager mode is off. I tried a few things but it didn't work
|
I think we can merge |
* disallow positional arguments in metrics methods * allow to use numpy/pytorch/tf/pandas objects in metrics * clean up GLUE dataset and metric doc * better checks to avoid positional arguments * style and quality * temp seed for everyone * fixes * more control * move cast to python in encode_example + add encode_batch * add metrics tests * cast to python objects in map * test cast to python objects in map * style * quality * remove kwargs in add and add_batch * fix download issue * add local + aws signature test on metrics * fix temp_seed for TF + add tests * better test Co-authored-by: Quentin Lhoest <lhoest.q@gmail.com>
predictionsvsreferencesmistakes