Training ChromBPNet from a checkpoint/previous-saved model
arodel21 opened this issue · 4 comments
Hello again!
Is there a way of training ChromBPNet from a previous saved model or a checkpoint?
Thanks in advance.
I am not saving all checkpoints to be mindful of disk space. But here is how I think you can do this -
Create a fork of the current repo and set the following argument - save_best_only
- to False in this line here -
chrombpnet/chrombpnet/training/train.py
Line 32 in 8e7fd8a
You can find more information about this argument here - https://keras.io/api/callbacks/model_checkpoint/
And what is the name of the file the checkpoint is saved with? Is it chrombpnet.h5 or chrombpnet_no_bias.h5?
Thank you!
You will have to look at the format but it will be chrombpnet _epoch(etc id).h5 You will need to isolate the chrombpnet_no_bias.h5 for each of the checkpoints yourself, I can point you to the function to do this.
May I ask why you will need this, perhaps I can be of better of help?
I could not find the file h5 you mention. I just need the checkpoints because of the limitations the cluster I am using to train the model gives me.