Open urinieto opened 7 years ago
wangsix
commented
7 years ago Hey @urinieto, without deep investigation into my own codes, the first point seems fine. About the second one, I agree with you I should comply to the option of pre-defined boundaries. I will refine the interface to have that option.
urinieto
commented
7 years ago Sounds good. The current version in MSAF is stable, but looking forward to your potential improvements. Thanks!
urinieto
commented
7 years ago Hey Cheng-i, did you have a chance to look into this? Thanks!
The new Variable Markov Oracle algorithm (included in this PR) didn't pass the unit tests because of two reasons:
The results of the main vmo call did not follow the expected format:
len(my_bounds) == len(my_labels) + 1. To fix this, I simply removed the last label (since it seemed that it was always repeated). @wangsix, could you confirm this is correct?When calling VMO to obtain the labels only (i.e., when we have used another algorithm to compute the boundaries), the previously computed boundaries were ignored. This was an MSAF problem, which I fixed by making sure we use the old boundaries in the interface. @wangsiz, do you think VMO could be improved by including previously computed boundaries to obtain the labels only? If not, the current labels synchronization to the previously computed boundaries should suffice.