fix: Approx. entropy in eager mode

[abstractqqq]

1. Fixed approximate_entropy. It works in eager mode now and is a lot faster than tsfresh
2. Reverted .count() back to .len(), because Series do not have .count(). I guess the 1-level of indirection is negligible.
3. Improved perf for absolute_maximum by 30-50%, and potentially lowered the memory usage by a lot. The is done by noticing that the max of abs(x) is either abs(x.max()) or abs(x.min()). So we do not need to create an intermediate series x.abs(), which saves a lot of memory. The speed up also comes from the fact that we can use pl.max_horizontal to parallelize the computation for abs(x.max()) and abs(x.min()).
4. Fixed some test cases, but not all. Longest_strike's test cases are still failing.
5. Added Heiner's change on mean_second_derivative_central which makes it work for both Expr and Series.
6. Improved perf for cid_ce and root_mean_square by using .dot. The reason is again due to the small memory footprint of .dot.
7. Improved last/first_loc_of_min/max by noticing a quirk. The speed up is due to a performance regression on arg_max, and therefore the alternative method is now faster. I will keep this in mind and revert it once the issue with arg_max is solved.
8. Slightly improved perf for autoregressive_coefficients by generating the data matrix faster using Polars instead of NumPy vstack.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
functime-docs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Oct 2, 2023 7:02pm