We observed a lot of errors, that occurred very frequently. This PR is meant to identify the causes and resolve the issues.

[x] On some occasions numbers close to zero caused our more efficient calculations for the interpolation of uncertainties to return zero instead of results close to zero, where the a little more involved calculations were correct. So we removed the apparently more efficient but sometimes wrong calculations.
[x] test_dft_deconv() needed another increase of the threshold to test its results are correct
[x] For some tests of interp1d_unc() the difference of the orders of magnitude of the input signal and interpolation nodes must not be too large, to stay precise with the calculations results, so we adapted the strategy values_uncertainties_kind() to ensure, the values can be requested to stay in a smaller bounded region than in general cases.
[x] Sometimes for pretty similar reasons interpolation nodes being close to be equal are considered equal and cause problems in the scipy internals, as interp1d() requests the values to be different, which required another adaption of the strategy.
[x] The random draw of a node step for the make_equidistant() calls caused out of memory errors, when they became too small, so we do not draw them randomly anymore, but compute them from the amount of interpolation nodes to spread them more or less evenly in the randomly drawn interpolation interval.
[x] Apparently interp1d() does not guarantee anymore, that interpolated values strictly stay within the original bounds, so we had to include a check in test_linear_in_interp1d_unc() for the values at least being close to the original bounds if not within.
[x] We reduced the effort to compute the test case for test_wrong_input_lengths_call_make_equidistant() just because it goes.
[x] We introduced the signal and filter uncertainties into the threshold in test_firuncfilter_mc_uncertainty_comparison.py
[x] Finally we discovered a bug in one of the subroutines of LSIIR which in some cases let to an undesired increase of the delay parameter tau compared to the original implementation. This is fixed now in 6286c754d7aad93ca903270b74b9cef15c494207.

Codecov Report

Merging #287 (6663bcd) into main (c264800) will increase coverage by 16.77%. The diff coverage is 100.00%.

:exclamation: Current head 6663bcd differs from pull request most recent head 3e37e33. Consider uploading reports for the commit 3e37e33 to get more accurate results

@@             Coverage Diff             @@
##             main     #287       +/-   ##
===========================================
+ Coverage   60.42%   77.20%   +16.77%     
===========================================
  Files          29       29               
  Lines        2239     2233        -6     
  Branches      366      363        -3     
===========================================
+ Hits         1353     1724      +371     
+ Misses        783      382      -401     
- Partials      103      127       +24

Impacted Files	Coverage Δ
src/PyDynamic/model_estimation/fit_filter.py	`90.34% <100.00%> (+0.68%)`	:arrow_up:
src/PyDynamic/uncertainty/interpolate.py	`90.12% <100.00%> (+1.61%)`	:arrow_up:
src/PyDynamic/misc/filterstuff.py	`63.93% <0.00%> (+1.63%)`	:arrow_up:
src/PyDynamic/uncertainty/propagate_convolution.py	`100.00% <0.00%> (+4.00%)`	:arrow_up:
src/PyDynamic/uncertainty/propagate_filter.py	`78.46% <0.00%> (+14.35%)`	:arrow_up:
src/PyDynamic/uncertainty/propagate_DFT.py	`65.57% <0.00%> (+18.30%)`	:arrow_up:
src/PyDynamic/misc/SecondOrderSystem.py	`100.00% <0.00%> (+45.45%)`	:arrow_up:
src/PyDynamic/uncertainty/propagate_MonteCarlo.py	`68.62% <0.00%> (+45.88%)`	:arrow_up:
... and 2 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

PTB-M4D / PyDynamic

Fix all those errors in the test suite that occur frequently in the pipeline #287

Codecov Report