WIP: Generic trainer

• Adds a generic trainer class and trainer script
• Moved the model specific stuff to config files; example config added
• uses .to(device) throughout

bob/learn/pytorch/scripts/train_generic.py

@@ -99,13 +99,23 @@ def main(user_input=None):
   logger.debug("Use GPU = {}".format(use_gpu))
   logger.debug("Perform cross validation = {}".format(do_crossvalidation))
 
+  # use new interface
+
+  device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+
+
   # process on the arguments / options
   torch.manual_seed(seed)
   if use_gpu:
     torch.cuda.manual_seed_all(seed)
   if torch.cuda.is_available() and not use_gpu:
+    device="cpu"
     logger.warn("You have a CUDA device, so you should probably run with --use-gpu")
 
+  logger.debug("Device used for training = {}".format(device))
+
+
+  # Which device to use is figured out at this point, no need to use `use-gpu` flag anymore
   # get data
   if hasattr(configuration, 'dataset'):
     
@@ -131,7 +141,7 @@ def main(user_input=None):
   
   # train the network
   if hasattr(configuration, 'network'):
-    trainer = GenericTrainer(configuration.network, configuration.optimizer,configuration.compute_loss,learning_rate=learning_rate, batch_size=batch_size, use_gpu=use_gpu, verbosity_level=verbosity_level,tf_logdir=output_dir+'/tf_logs',do_crossvalidation=do_crossvalidation)
+    trainer = GenericTrainer(configuration.network, configuration.optimizer,configuration.compute_loss,learning_rate=learning_rate, batch_size=batch_size, device=device, verbosity_level=verbosity_level,tf_logdir=output_dir+'/tf_logs',do_crossvalidation=do_crossvalidation)
     trainer.train(dataloader, n_epochs=epochs, output_dir=output_dir, model=model)
   else:
     logger.error("Please provide a network in your configuration file !")