midi_degradation_toolkit double_pianoroll_to

double_pianoroll_to_df is very slow.

Open apmcleod opened this issue 6 years ago • 2 comments

Eval on the test set for 1 model I trained ran nearly 30 mins.

Oct 18 '19 01:10 apmcleod

Getting the data loaders to return the clean and degraded dfs should cut this time in 1/3.

But this has proven to be difficult:

TypeError: default_collate: batch must contain tensors, numpy arrays, numbers, dicts or lists; found <class 'pandas.core.frame.DataFrame'>

Nov 01 '19 04:11 apmcleod

This isn't priority for ACME v1.0, kicking to Next milestone (but may want to kick even further if it's a bit tricky).

Nov 13 '19 17:11 JamesOwers