readme.md

midi notes trainer & generator in pytorch

How to run

1. make dataset directory and put training midi files in it
2. run train.py. Every 100 epochs, weight file will be generated.
3. change const.py setting (gen_load and optim_load variable) according to generated weight files
4. run gen.py. out.mid will be generated.

network

linear -> relu -> linear -> relu -> linear -> relu -> softmax

Each note will be serialized as 2 * 128 tensor

Input: last 16 notes

Output: probability distribution of note melody and note length