feat(datasets): update simulated data setup

jolars commented 2 years ago

This PR is tangent to https://github.com/benchopt/benchmark_lasso/pull/90 and based on discussions with @mathurinm.

Changes:

add rho = 0 to the matrix (which should've been the case all along)
add a n ≈ p case (or p slightly larger than n rather)
update the dimensions somewhat (I'm not sure these choices are exactly the right ones since they're quite time-consuming)
add an n_signals parameter and make sure that the signal is sparse for the p > n cases

jolars commented 2 years ago

Maybe the 200 x 300 case should be skipped altogether, I'm not sure.

jolars commented 2 years ago

I've dropped the n ~~ p case for now and but upped the number of features to 10000 in the p > n case. I've decreased min(n, p) to 200 to combat the very long running times that these simulations otherwise take. Finally I've also harmonized the number of signals to be 20.

        "n_samples, n_features, n_signals": [
            (10_000, 200, 20),
            (200, 10_000, 20),
        ],

I'm merging this now because I think controlling the number of signals is a necessary change, but feel free to open an issue about this if you disagree with the other changes and we'll discuss them more.

benchopt / benchmark_lasso_path

feat(datasets): update simulated data setup #11