Training: remove trivial loss function

training.py

@@ -21,7 +21,6 @@
 parser.add_argument('--learning_rate', type=float, default=2e-5)
 parser.add_argument('--save_time', type=int, default=60 * 60)
 parser.add_argument('--baby_parameters', action='store_const', const=True, default=False)
-parser.add_argument('--use_dumb_loss_function', action='store_const', const=True, default=False)
 parser.add_argument('--no_random_degradation', action='store_const', const=True, default=False)
 args = parser.parse_args()
 
@@ -318,8 +317,7 @@ def eval_model(model, eval_datasets):
             for target_sample, dataloader in eval_datasets:
                 for history, fragments in iter(dataloader):
                     pred = model(target_sample, history, fragments, middle_repeats=middle_repeats)
-                    lf = trivial_loss_function if args.use_dumb_loss_function else loss_function
-                    loss: Tensor = lf(pred.float(), fragments.float()).to(dtype=precision)
+                    loss: Tensor = loss_function(pred.float(), fragments.float()).to(dtype=precision)
                     if loss.isnan():
                         raise LossNaNException()
                     total_loss += loss.item()
@@ -342,11 +340,6 @@ def loss_function_freq_significance(width, device):
     return loss_function_freq_significance_cache[1]
 
 
-def trivial_loss_function(pred, truth):
-    """Very stupid, but 100% bug-free loss function"""
-    return torch.mean((truth - pred) ** 2)
-
-
 def loss_function(pred, truth):
     """Custom loss function, for comparing two spectrograms. Not the best one, but it should work."""
     # TODO: this can be infinitely improved:
@@ -417,8 +410,7 @@ def train_on_bite(model: EchoMorph, optimizer: torch.optim.Optimizer, train_spec
     for history, fragments in iter(dataloader):
         optimizer.zero_grad()
         pred = model(target_sample, history, fragments)
-        lf = trivial_loss_function if args.use_dumb_loss_function else loss_function
-        loss: Tensor = lf(pred.float(), fragments.float()).to(dtype=precision)
+        loss: Tensor = loss_function(pred.float(), fragments.float()).to(dtype=precision)
         if loss.isnan():
             raise LossNaNException()
         loss.backward()