Closed GabrielKP closed 2 months ago
The bug is caused by downsampling in this line: https://github.com/GabrielKP/enc/blob/b18ec313d1a06b1366d4af6837a917e741b1857f/src/main.py#L42
The data is loaded in with 27683702
samples at 44100
hz -> 313.87
TRs which after trimming becomes 303.87
TRs which then in this line is rounded down to 303
, whereas the fMRI data has 304
TRs:
https://github.com/GabrielKP/enc/blob/b18ec313d1a06b1366d4af6837a917e741b1857f/src/features.py#L25
I cannot just replace the int
with round
, because the story naked
will have the same problem if you start rounding.
I have written a function which takes the y_data
and computes the n_samples appropriately.
This could distort the audio alignment slightly but the alternative is to mess with the padding subtraction.
There is a bug when loading
undertheinfluence
: a shape mismatch between theX
(303 tr's) and they
(304 tr's) data.