Lorenz Schmidt

> This will give you the estimate of runtime speed and memory consumption ratios at the time I checked last the evaluation uses "thin" matrices, but in my case all...

probably my expectations are too high and I need to reparametrize the model, but still want to double-check that I'm using householder correctly

```bash $ python3 benchmark_one.py --repeats 5 --method hh_ours --b 1 --d 512 --r 512 70.72763475589454 1.2755393754559918e-06 $ python3 benchmark_one.py --repeats 5 --method hh_pt --b 1 --d 512 --r 512 318.1826988235116...

this also has the effect that when doubling the same data (e.g. appending audio file to itself), the band power changes

I ran into the same issue when using tarpaulin for the whole workspace in [rust-ml/linfa](github.com/rust-ml/linfa), this runs fine ``` cargo +nightly tarpaulin --all-features -v --run-types Tests Doctests ``` but passing...

for `explained_variance_diag` this relates to https://github.com/rust-ml/linfa-linalg/pull/11 when decreasing to single-precision floats we can't extract as many eigenvalues without losing coherence

the same occurs for `test_whitening_small`, it calculate covariance of 2x2 https://github.com/rust-ml/linfa-linalg/blob/main/src/lobpcg/svd.rs#L200

I propose first fixing first https://github.com/rust-ml/linfa-linalg/pull/11 by adding the 1/5 rules to `TruncatedSvd` and `TruncatedEig` and then test again

perhaps this can be emulated by setting the maximal number of iterations to 1 (or another value if you want to get informed every `` iterations) and using `fit_with`. The...

I think that having debug information in the estimated model is more useful than sparkling prints through the code, but as you said sometimes this is not feasible