Directly return peak position from neighbor average waveform

[maxnoe]

This is a small change to not compute the full array of average neighbor waveforms for each pixel and then in a separate numpy call calculate the peak position but do it directly in the same numba function.

This reduces memory usage and call overhead, runtime is divided in half from 800 to 400 µs for this test case:

import numpy as np
from ctapipe.instrument import TelescopeDescription, SubarrayDescription
import astropy.units as u
from ctapipe.image.extractor import NeighborPeakWindowSum
from timeit import repeat


tel = TelescopeDescription.from_name('LST', 'LSTCam')
geom = tel.camera.geometry
subarray = SubarrayDescription(
    'perf',
    tel_positions={1: [0, 0, 0] * u.m},
    tel_descriptions={1: tel},
)


extractor = NeighborPeakWindowSum(subarray)

waveforms = np.random.normal(size=(geom.n_pixels, 40)).astype('float32')
selected_gain_channels = np.zeros(geom.n_pixels, dtype=int)


# number_of_islands(geom, mask)
code = '''
extractor(waveforms, 1, selected_gain_channels)
'''

exec(code)

N = 200
times = np.array(repeat(code, number=N, repeat=10, globals=globals()))
print(f'{np.mean(times / N) * 1e6:.2f} ± {np.std(times / N) * 1e6:.2f} µs')

[codecov]

Codecov Report

Merging #1914 (c733f01) into master (b11ce6c) will increase coverage by 0.04%.
The diff coverage is 66.66%.

@@            Coverage Diff             @@
##           master    #1914      +/-   ##
==========================================
+ Coverage   92.05%   92.09%   +0.04%     
==========================================
  Files         189      189              
  Lines       14872    14872              
==========================================
+ Hits        13691    13697       +6     
+ Misses       1181     1175       -6     

Impacted Files	Coverage Δ
ctapipe/image/__init__.py	100.00% <ø> (ø)
ctapipe/image/extractor.py	83.39% <28.57%> (+0.25%)	⬆️
ctapipe/image/tests/test_extractor.py	100.00% <100.00%> (ø)
ctapipe/conftest.py	94.23% <0.00%> (ø)
ctapipe/io/tests/conftest.py	100.00% <0.00%> (ø)
ctapipe/io/tests/test_prod2.py	100.00% <0.00%> (ø)
ctapipe/tools/tests/test_tools.py	100.00% <0.00%> (ø)
ctapipe/io/tests/test_datawriter.py	100.00% <0.00%> (ø)
ctapipe/tools/tests/test_process.py	95.27% <0.00%> (ø)
ctapipe/io/tests/test_eventseeker.py	98.11% <0.00%> (ø)
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b11ce6c...c733f01. Read the comment docs.