To replicate this environment:

1. regenerate the source_sep_environment.yml (conda env export > source_sep_environment, with the source_sep being the active environment)
2. install it on the other machine (or virtual environment) with conda. 
3. I should keep my asteroid fork up to date, but for now just copy the asteroid directory on my user. (~/Documents/asteroid). It is preferrable to install it as dev (pip install -e .). This will be easier than installing a fresh asteroid since I have changed some things.
3. Clone clone the leakage_removal repo on my github account. 
4. Copy lr_musdb (short for leakage removal musdb) from my account, or clone it from https://github.com/Alia-morsi/lr_musdb.  This has a modified verion of the musdb dataloader to rely on the folder structure of my custom dataset. 
5. run python setup.py install from within lr_musdb, to install the modified musdb module. 

To train an existing model:


To create a new model:

Then, at this point all you should do is to make sure your conf paths etc are correct and just start running :))

To create a runnable model from a checkpoint, copy the run_checkpoints.py to the folder where the model outputs is stored, update the checkpoint name, and run it. You will get a file called runnable_ckpt.pth

To run the demucs environment for eval:
1. create a copy of the source_sep conda env
2. Activate it, and update the env with the environment-cuda or cpu depending on your machine. Be sure to remove the name: key from the env fileso that it actually updates the environment you are in rather than create a new one.
3. clone the demucs repo
4. checkout branch v2
5. pip install . -e
Then, the eval scripts for demucs or tasnet can be run normally.