modify app

app.py

@@ -161,8 +161,10 @@ with gr.Blocks() as demo:
                 title="ITO Loss Curve",
                 x_title="Step",
                 y_title="Loss",
-                height=400,
+                height=300,
                 width=600,
+                x_series_axis=gr.LinePlot.XAxisType.INTEGER,
+                y_lim=[None, None],
             )
             ito_log = gr.Textbox(label="ITO Log", lines=10)
 
@@ -187,6 +189,11 @@ with gr.Blocks() as demo:
             final_log = log
             loss_df = loss_data
 
+        # Update y_lim based on the actual min and max values
+        y_min = loss_df['loss'].min()
+        y_max = loss_df['loss'].max()
+        ito_loss_plot.update(y_lim=[y_min, y_max])
+
         return final_audio, final_params, final_log, loss_df
 
     ito_button.click(

inference.py

@@ -109,7 +109,7 @@ class MasteringStyleTransfer:
             if step == 0:
                 initial_params = current_params
             top_5_diff = self.get_top_n_diff_string(initial_params, current_params, top_n=5)
-            log_entry = f"Step {step + 1}, Loss: {total_loss.item():.4f}\n{top_5_diff}\n"
+            log_entry = f"Step {step + 1}\n   Loss: {total_loss.item():.4f}\n{top_5_diff}\n"
 
             if divergence_counter >= 10:
                 print(f"Optimization stopped early due to divergence at step {step}")
@@ -166,52 +166,6 @@ class MasteringStyleTransfer:
 
         return output_audio, predicted_params, self.args.sample_rate
 
-    def print_param_difference(self, initial_params, ito_params):
-        all_diffs = []
-
-        print("\nAll parameter differences:")
-        for fx_name in initial_params.keys():
-            print(f"\n{fx_name.upper()}:")
-            if isinstance(initial_params[fx_name], dict):
-                for param_name in initial_params[fx_name].keys():
-                    initial_value = initial_params[fx_name][param_name]
-                    ito_value = ito_params[fx_name][param_name]
-                    
-                    # Calculate normalized difference
-                    param_range = self.mastering_converter.fx_processors[fx_name].param_ranges[param_name]
-                    normalized_diff = abs((ito_value - initial_value) / (param_range[1] - param_range[0]))
-                    
-                    all_diffs.append((fx_name, param_name, initial_value, ito_value, normalized_diff))
-                    
-                    print(f"  {param_name}:")
-                    print(f"    Initial: {initial_value.item():.4f}")
-                    print(f"    ITO:     {ito_value.item():.4f}")
-                    print(f"    Normalized Diff: {normalized_diff.item():.4f}")
-            else:
-                initial_value = initial_params[fx_name]
-                ito_value = ito_params[fx_name]
-                
-                # For 'imager', assume range is 0 to 1
-                normalized_diff = abs(ito_value - initial_value)
-                
-                all_diffs.append((fx_name, 'width', initial_value, ito_value, normalized_diff))
-                
-                print(f"  width:")
-                print(f"    Initial: {initial_value.item():.4f}")
-                print(f"    ITO:     {ito_value.item():.4f}")
-                print(f"    Normalized Diff: {normalized_diff.item():.4f}")
-
-        # Sort differences by normalized difference and get top 10
-        top_diffs = sorted(all_diffs, key=lambda x: x[4], reverse=True)[:10]
-
-        print("\nTop 10 parameter differences (sorted by normalized difference):")
-        for fx_name, param_name, initial_value, ito_value, normalized_diff in top_diffs:
-            print(f"{fx_name.upper()} - {param_name}:")
-            print(f"  Initial: {initial_value.item():.4f}")
-            print(f"  ITO:     {ito_value.item():.4f}")
-            print(f"  Normalized Diff: {normalized_diff.item():.4f}")
-            print()
-
     def print_predicted_params(self, predicted_params):
         if predicted_params is None:
             print("No predicted parameters available.")
@@ -273,9 +227,9 @@ class MasteringStyleTransfer:
 
         top_diffs = sorted(all_diffs, key=lambda x: x[4], reverse=True)[:top_n]
         
-        output = [f"Top {top_n} parameter differences (initial / ITO / normalized diff):"]
+        output = [f"   Top {top_n} parameter differences (initial / ITO / normalized diff):"]
         for fx_name, param_name, initial_value, ito_value, normalized_diff in top_diffs:
-            output.append(f"{fx_name.upper()} - {param_name}: {initial_value:.3f} / {ito_value:.3f} / {normalized_diff:.3f}")
+            output.append(f"      {fx_name.upper()} - {param_name}: {initial_value:.3f} / {ito_value:.3f} / {normalized_diff:.3f}")
         
         return "\n".join(output)