[telemelody] Too many errors of template2melody when generating training data

[gandolfxu]

Does it conforms to expectation?

[jzq2000]

Yes, it conforms to expectation. The processing filters out non 4/4, duplicate and blank midis, so it is meeting the expected standards.

[gandolfxu]

train: 21332/108590 (19.64%) midi files successfully processed valid: 93/1109 (8.39%) midi files successfully processed test: 74/1114 (6.64%) midi files successfully processed

I find that the number of midi data for training is 21332, which is different from that in the paper. In the paper, this number is 45129. @ @jzq2000

[jzq2000]

Hi, 45129 is the total number of midi samples in LMD-matched dataset. Maybe you can check whether there are duplicate/other midis under your data_dir.