run benchmarks

[tachella]

This PR adds the run_benchmark function to this repo such that users can quickly evaluate their model on a given benchmark as

from deepinv_bench import run_benchmark
my_solver = lambda y, physics: ...  # your solver here
results = run_benchmark(my_solver, "benchmark_name")

@tomMoral is there an easy way to do this with benchopt? see todo in the PR :)

[tachella]

many thanks for the proposed solution!

I'm getting an error when running locally, I was wondering if you get the same?

[tachella]

why do we need this?

[tomMoral]

This is to make it possible to run the benchmark with the CLI. We need the model to be valid.
Also this model is useful for testing the objective maybe worth putting in deepinv? (Like sklearn.dummy which gives you random perf)

[tomMoral]

This is to make it possible to run the benchmark with the CLI. We need the model to be valid.
Also this model is useful for testing the objective maybe worth putting in deepinv? (Like sklearn.dummy which gives you random perf)

[tachella]

I see - you can do this model as dinv.models.ArtifactRemoval (see here) and a dummy (identity denoiser). I think we could later on add the option to pass None to the denoiser