gwinndr / musictransformer-pytorch Goto Github PK

Need to select the best epoch based on the validation set and use that best epoch to get the final accuracy and loss.

Using the test set to select the best epoch, biases the model towards the test set. The test set is supposed to represent "in the wild" data and evaluate how well the final model performs on yet-to-be-seen data.

Hi! Is there a way to access the weights to get a pre-trained version of the network? I have searched the repo but couldn't find anything, and I cannot train such a network on my small laptop.

In train.py line 137
for epoch in range(start_epoch, args.epochs):
When I want to continue to train based on a model weight, there will be an error. Maybe it should be start_epoch+args.epochs ?

I also woud like to ask if you can provide a trained weight, which can be very convenient.

Thank you

Hey Damon,

I think I finally did it and I was able to make a fully working Google Colab that actually plays well. I used my TMIDI processors and I also streamlined the colab/implementation.

https://github.com/asigalov61/SuperPiano/blob/master/%5BTMIDI%5D_Super_Piano_3.ipynb

The only thing it does not have is the control_changes/program_changes/sustains (I still need to implement it in my processors) but it still plays pretty well IMHO on my dataset. Not sure about MAESTRO but you are welcome to try it.

Let me know if it is useful.

Thanks.

Alex.

Hello, I'm a Japanese college student.
This is my first time to use MusicTransformer-Pytorch.
I refered to READ ME and Google Colab version, and I tried to train this model.

When I wrote
python3 train.py -output_dir rpr --rpr -batch_size=4 -epochs=150 -max_sequence=2048

,the terminal outputs
TypeError: forward() got an unexpected keyword argument 'is_causal'

Please tell me how to train this model.

Hey Damon,

I just wanted to say hi and thank you for your fantastic work on Music Transformer PyTorch repo/code.

It works great and I was able to reproduce the results.

I have created a nice Google Colab with your repo/code, so I wanted to invite you to check it out. And if you like, feel free to add it to your repo so that people can try it easily and quickly.

https://github.com/asigalov61/SuperPiano/blob/master/Super_Piano_3.ipynb

Sincerely,

Alex

Hi!
Small bug to correct, line 92 of the e_piano.py file:

tgt[raw_len] = TOKEN_END throws an error because raw_len is out of bounds. This should be raw_len-1. This is not a problem with the maestro dataset, which probably contains more than max_seq=2048 events for all performances, but this can show up in custom datasets.

When I tried to generate midi longer and I change the parameter target_seq_length[in argument_funcs.py] to 4096(default 1024), the issues can be reproduced:
RuntimeError: The size of tensor a (2049) must match the size of tensor b (2048) at non-singleton dimension 0

I want to know it is the Transformer's limit or just the MusicTransformer settings cause the problem?

Hello, I would like to know the current project MusicTransformer-Pytorch and Optimus-VIRTUOSO in music generation which effect is good, or your sample data set where to find it, I want to test or train your project to see the effect, I am a beginner, the current know not much, would like to ask the two big brothers guidance, thanks.