FFCV support

[lrzpellegrini]

Current status: barely working draft

How to enable the support for FFCV in Avalanche:

• Install FFCV using the official guide (involves some specific conda packages)
• Use the enable_ffcv function, as shown in the ffcv_enable.py script in examples.

Offers a very good speed when compared to the PyTorch loader especially on "big" benchmarks like TinyImagenet. Minor improvement with the CIFAR or MNIST datasets, where it's still good to offer the same performance for fewer workers.

A few benchmarks, such as CORe50 NICv2 are not properly supported as they are implemented as one leaf dataset per experience. Currently supported benchmarks are the ones for which one single dataset is split into multiple experiences (like the SplitMNIST, SplitCIFAR, ...). Better support for CORe will come in the next PRs.

[coveralls]

Pull Request Test Coverage Report for Build 5716867115

• 500 of 806 (62.03%) changed or added relevant lines in 14 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-0.2%) to 72.749%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
avalanche/training/templates/problem_type/supervised_problem.py	1	2	50.0%
avalanche/training/templates/base_sgd.py	5	7	71.43%
avalanche/training/plugins/replay.py	5	8	62.5%
tests/unit_tests_utils.py	72	75	96.0%
avalanche/benchmarks/utils/data_loader.py	25	32	78.13%
tests/test_transformations.py	155	163	95.09%
avalanche/benchmarks/utils/transforms.py	51	70	72.86%
avalanche/benchmarks/utils/dataset_traversal_utils.py	103	136	75.74%
tests/benchmarks/ffcv/test_ffcv_support.py	19	64	29.69%
avalanche/benchmarks/utils/ffcv_support/ffcv_components.py	58	243	23.87%

Files with Coverage Reduction	New Missed Lines	%
avalanche/benchmarks/utils/transforms.py	1	79.01%

Totals
Change from base Build 5693163363:	-0.2%
Covered Lines:	17379
Relevant Lines:	23889

---

💛 - Coveralls

[lrzpellegrini]

Updated PR description, added examples, added a unit test, and removed the remaining workarounds.

I think the PR is ready!

[AntonioCarta]

Hi Lorenzo, I did a more thorough review of the PR. It's mostly about missing docstrings.

Looking at the changes, I fell even more strongly that we should try to move the dataloader creation outside. In your PR you had to add custom code wherever we created dataloaders. This is quite difficult to track (in fact, we have some plugins that make their own dataloaders).

Apart from the minor changes requested in the comments I think the PR can be accepted as is

[AntonioCarta]

Let me know when I can merge this.

[lrzpellegrini]

It can be merged now!