bugfixes

src/__main__.py

@@ -4,6 +4,7 @@
 import torch
 import pickle
 import numpy as np
+import os
 
 from .documenter import Documenter
 from .train import DiscriminatorTraining
@@ -45,17 +46,17 @@ def main():
         print(f"  Val points: {len(data.val_true)} truth, {len(data.val_fake)} generated")
 
         print("  Building model")
-        model_dir = doc.get_file(f"model_{data.suffix}.pth")
+        model_dir = doc.get_file(f"model_{data.suffix}")
         os.makedirs(model_dir, exist_ok=True)
         training = DiscriminatorTraining(params, device, data, model_dir)
 
-        if args.load_model:
-            print("  Loading model")
-            training.load(args.model_name)
-        else:
+        if not args.load_model:
             print("  Running training")
             training.train()
 
+        print(f"  Loading model {args.model_name}")
+        training.load(args.model_name)
+
         if args.load_model and args.load_weights:
             print("  Loading weights")
             with open(doc.get_file(f"weights_{data.suffix}.pkl"), "rb") as f:

src/plots.py

@@ -88,7 +88,8 @@ def plot_losses(self, file: str):
                 pdf,
                 "learning rate",
                 (self.losses["lr"], ),
-                (None, )
+                (None, ),
+                "log"
             )
 
 
@@ -97,7 +98,8 @@ def plot_single_loss(
         pdf: PdfPages,
         ylabel: str,
         curves: tuple[np.ndarray],
-        labels: tuple[str]
+        labels: tuple[str],
+        yscale: str = "linear"
     ):
         """
         Makes single loss plot.
@@ -107,6 +109,7 @@ def plot_single_loss(
             ylabel: Y axis label
             curves: List of numpy arrays with the loss curves to be plotted
             labels: Labels of the loss curves
+            yscale: Y axis scale, "linear" or "log"
         """
         fig, ax = plt.subplots(figsize=(4,3.5))
         for i, (curve, label) in enumerate(zip(curves, labels)):
@@ -122,6 +125,7 @@ def plot_single_loss(
             verticalalignment = "top",
             transform = ax.transAxes
         )
+        ax.set_yscale(yscale)
         if any(label is not None for label in labels):
             ax.legend(loc="center right", frameon=False)
         plt.savefig(pdf, format="pdf", bbox_inches="tight")
@@ -309,6 +313,7 @@ def plot_single_weight_hist(
         ax.set_xscale(xscale)
         ax.set_yscale(yscale)
         ax.set_xlim(bins[0], bins[-1])
+        ax.legend(frameon=False)
         plt.savefig(pdf, format="pdf", bbox_inches="tight")
         plt.close()
 

src/train.py

@@ -184,14 +184,15 @@ def train(self):
             train_loss = torch.stack(epoch_losses).mean()
             self.losses["train_loss"].append(train_loss.item())
             self.losses["val_loss"].append(val_loss.item())
-            self.losses["lr"].append(self.optimizer.param_groups[0]["lr"])
+            epoch_lr = self.optimizer.param_groups[0]["lr"]
+            self.losses["lr"].append(epoch_lr)
             if self.bayesian:
                 self.losses["train_bce_loss"].append(torch.stack(epoch_bce_losses).mean().item())
                 self.losses["train_kl_loss"].append(torch.stack(epoch_kl_losses).mean().item())
                 self.losses["val_bce_loss"].append(val_bce_loss.item())
                 self.losses["val_kl_loss"].append(val_kl_loss.item())
             print(f"    Epoch {epoch:3d}: train loss {train_loss:.6f}, " +
-                  f"val loss {val_loss:.6f}")
+                  f"val loss {val_loss:.6f}, LR {epoch_lr:.3e}", flush=True)
 
             if val_loss < best_val_loss:
                 best_val_loss = val_loss