Remove Savitzky-Golay filter in suppression detection #102
Conversation
mattbit
changed the title
Start by checking the existing ones (currently failing). Make sure that the detection does not break. Then add similar tests for signals containing artifacts, i.e. what happens when you have an artifact in the middle of a suppression? Write a test which ensures the correct behaviour. |
Replace strict inequality by large inequality.
|
Done ! |
| with np.errstate(invalid='ignore'): | ||
| ies_mask = envelope < threshold | ||
| ies_mask = envelope <= threshold |
There was a problem hiding this comment.
Because otherwise something occurs when, during the suppression, there is a single sample with value threshold, placed at a time t < min_duration after the beginning of the suppressions. In this case, the interval before the problematic sample is not considered as a suppression. This is what made the tests fail on the first commit. An AssertionError was raised when running test_detect_suppressions(3). Because the sample 532 was in the situation I mentionned before.
| # The suppressions must not overlap with the artifact | ||
| sec = pd.Timedelta('1s') | ||
| assert detections.loc[0].end < 3.2 * sec | ||
| assert detections.loc[1].end > 4.8 * sec |
There was a problem hiding this comment.
Oh yes, sorry about that, I'll change it
|
Cannot reproduce, see #79 |
Allows the detect_suppressions function to work on recordings on which artifacts have been detected.
I'm not sure what tests you want me to add to the pull request, please let me know if I need to add stuff.