Currently, we run gridsearch in parallel with the help of joblib, which pickles the data.
However, the linear_op cannot be pickled right, as we have some internal objects whose all the data is not rightly loaded back. Due to this, the wavelet operator in gridsearch always works with the same default linear_op.
This issue cropped up as we wanted to test with 3D gridsearching.
I will update the codes to initialize the linear_op on the fly. However, we also need to ensure the testing is better.
Finally, this is just fixing the effect and not the actual cause of the issue, that is the linear_op is not being pickled right.
Currently, we run gridsearch in parallel with the help of joblib, which pickles the data.
However, the linear_op cannot be pickled right, as we have some internal objects whose all the data is not rightly loaded back. Due to this, the wavelet operator in gridsearch always works with the same default linear_op.
This issue cropped up as we wanted to test with 3D gridsearching.
I will update the codes to initialize the linear_op on the fly. However, we also need to ensure the testing is better.
Finally, this is just fixing the effect and not the actual cause of the issue, that is the linear_op is not being pickled right.