remove njit from simple functions

[jmoralez]

Contributes to #753 by removing the njit decorator from functions that can be fast with just numpy. Also removes the season_length argument from _repeat_val_seas, since it can be inferred from the size of the season_vals.

Below are some benchmarks for the non-trivial functions, some increase its runtime but they're all in microseconds, so the impact wouldn't be noticeable (even with many series).

_intervals

x = np.random.randint(low=0, high=100, size=10_000).astype(np.float32)
%timeit _intervals(x)

main	this PR
43 µs	38 µs

_chunk_sums

x = np.random.rand(10_000)
%timeit _chunk_sums(x, 5)
%timeit _chunk_sums(x, 10)
%timeit _chunk_sums(x, 20)

chunk_size	main	this PR
5	10 µs	35 µs
10	7 µs	23 µs
20	6 µs	13 µs

_seasonal_window_average

x = np.random.rand(10_000)
h = 100
fitted = False
season_length = 52
%timeit _seasonal_window_average(x, h, fitted, season_length, 10)
%timeit _seasonal_window_average(x, h, fitted, season_length, 20)
%timeit _seasonal_window_average(x, h, fitted, season_length, 50)

window_size	main	this PR
10	6 µs	20 µs
20	8 µs	20 µs
50	15 µs	19 µs

_repeat_val_seas

seas_vals = np.random.rand(24)
%timeit _repeat_val_seas(seas_vals, 500)

main	this PR
3 µs	5 µs

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[AzulGarza]

awesome @jmoralez. LGTM :)