Upload New File

visualisation.py

+import matplotlib.pyplot as plt
+
+def plot_learning_curves(avg_train_losses, avg_val_losses, parameters, is_save = False, save_path = None) :
+    
+    # Plot the learning curves 
+    fig = plt.figure(figsize=(12,6))
+    plt.plot(range(1,len(avg_train_losses)+1), avg_train_losses, label='Training Loss', color='black')
+    plt.plot([i for i in range(0, len(avg_train_losses), parameters['eval_frequency'])], avg_val_losses,label = 'Validation Loss', color='red')
+
+    # Plot at which epoch early stopping started
+    min_index = (len(avg_val_losses) - 6) * parameters['eval_frequency']
+    plt.axvline(min_index, linestyle = '--', label = 'Beginning of Early Stopping', color = 'green')
+
+    # Customizing the plot
+    plt.title('Learning curves')
+    plt.xlabel('Epochs')
+    plt.ylabel('Losses')
+    plt.legend(loc='upper right')
+    plt.grid(True)
+    plt.show()
+    
+    if is_save and (save_path != None): 
+        fig.savefig(save_path) #'/media/data/MSc_students_accounts/maxime/figures/Learning_Curves_Unet_baseline.png'
+
+    
+
+def plot_model_features(model, training_loader, device):
+    fig = plt.tight_layout()
+    activation = {}
+    def get_activation(name):
+        def hook(model, input, output):
+            activation[name] = output.detach()
+        return hook
+    vis_labels = ['inc', 'down1', 'down2', 'down3', 'down4', 'up1', 'up2', 'up3', 'up4'] #'layer4', 'layer5', 'layer6']
+
+    for l in vis_labels:
+        getattr(model, l).register_forward_hook(get_activation(l))
+
+    one_batch = next(iter(training_loader))
+
+    data = one_batch[0][0, ...]#['mri_slice']['data']
+    data = data.to(device = device)#, dtype = dtype)
+    output = model(data)
+
+    for idx, l in enumerate(vis_labels):
+        act = activation[l].squeeze()
+
+        # only showing the first 16 channels
+        ncols, nrows = 8, 2
+
+        fig, axarr = plt.subplots(nrows, ncols, figsize=(15,5))
+        fig.suptitle(l)
+
+        count = 0
+        for i in range(nrows):
+            for j in range(ncols):
+                axarr[i, j].imshow(act[count].cpu())
+                axarr[i, j].axis('off')
+                count += 1