Removal of metrics deepcopy before computing the metrics

[gregjauvion]

What does this PR do?

This PR removes the deep copy of the metrics: Dict[str, Metric] object in the function _compute_and_log_metrics.

This function does the following:

• Copy of the metrics (using copy.deepcopy)
• Computation of the values of the metrics on the copy

It is called 3 times in composer.trainer.Trainer:

• At every batch in the training loop to compute the metrics on this batch, and at the end of the epoch
• At the end of the evaluation loop to compute the metrics on the evaluation dataset

I don't see the need for copying the metrics before calling metric.compute(). In particular, the metrics are reset with metric.reset() at the start of the training on each batch, and at the start of the evaluation loop, making this copy useless. Is there a specific reason I am missing?

What issue(s) does this change relate to?

Related to issue #3153 describing that copying the metrics results in a memory leak in a specific use-case I'm working on.

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[gregjauvion]

Hi @mvpatel2000 should I assign this MR to a maintainer of the repo? Please tell me who I should assign, thank you.

[mvpatel2000]

Hi @mvpatel2000 should I assign this MR to a maintainer of the repo? Please tell me who I should assign, thank you.

@gregjauvion feel free to request from me. I can triage if necessary

[mvpatel2000]

@gregjauvion thanks for the PR, I think this is correct. I will run a series of manual tests though to verify identical behavior

[gregjauvion]

Thanks a lot for your reactivity! Please tell me if I can help in any way.

[mvpatel2000]

Test failures seem related to a flakey test introduced in previous PR, patching elsewhere

[mvpatel2000]

LGTM!