Add start_epoch variable to base trainer train()

[liamchalcroft]

As far as I can tell, there is no way to resume training currently without the trainer beginning from epoch 1. This should avoid overwrites and allow to continue correctly in schedulers (not tested)

[clementchadebec]

Hi @liamchalcroft,

Sorry for the late reply. I indeed think this is a useful feature that I will integrate in the near future. Nonetheless, I was thinking of a method called resume_training_from_folder that will take as input the path to a folder containing the checkpoints of the model, the optimizer and the scheduler. It will then reload their state_dict and launch resume the training as you propose.