new difference figure

UNet/Auswertung_64.py

 import torch
 import numpy as np
-import UNet_V16 as UNet
+import torch.nn as nn
+import UNet_V14 as UNet
 import copy
 
-
 def rescale(output, normalization):
     output_rescale = output.reshape(output.shape[2],output.shape[3],output.shape[4])
     if normalization is not None:
@@ -15,22 +15,21 @@ def rescale(output, normalization):
         output_rescale += min_label
     return output_rescale
 
-def dataset_evaluation( normalization, model, dataset, threshold = 0.05):
+def dataset_evaluation( normalization, model, dataloader, threshold = 0.05):
     model.eval()
-    mean_error = np.empty(len(dataset))
-    max_error = np.empty(len(dataset))
-    correct_per = np.empty(len(dataset)) #percentage of voxel that are guessed corrected, according to threshold
-    mean_deviation_abs = np.empty(len(dataset)) #absolute mean deviation between prediction and label per RVE
-    mean_deviation_per = np.empty(len(dataset)) #percentile mean deviation between prediction and label per RVE
-
-    for index in range(len(dataset)):
-        input, output = dataset[index]
+    mean_error =  []
+    max_error =  []
+    correct_per = []#percentage of voxel that are guessed corrected, according to threshold
+    mean_deviation_abs = [] #absolute mean deviation between prediction and label per RVE
+    mean_deviation_per =  [] #percentile mean deviation between prediction and label per RVE
+
+    for batch in dataloader:
+        input, output = batch
         input = copy.copy(input)
         output = copy.copy(output)
         input = torch.unsqueeze(input,0)
         output = torch.unsqueeze(output,0)
-        xb = UNet.to_device(input, device)
-        prediction = model(xb)
+        prediction = model(input)
         input = input.detach().numpy()
         prediction = prediction.cpu().detach().numpy()
         output = output.detach().numpy()
@@ -38,12 +37,12 @@ def dataset_evaluation( normalization, model, dataset, threshold = 0.05):
         output = rescale(output, normalization)
         error = (abs(output - prediction)/output)
         right_predic = (error < threshold).sum()
-        mean_error[index] = error.mean()*100.
-        max_error[index] = error.max()*100.
-        mean_deviation_abs[index] = prediction.mean() - output.mean()
-        mean_deviation_per[index] = (prediction.mean() - output.mean())/output.mean()
+        mean_error.append(error.mean()*100.)
+        max_error.append(error.max()*100.)
+        mean_deviation_abs.append(prediction.mean() - output.mean())
+        mean_deviation_per.append((prediction.mean() - output.mean())/output.mean())
 
-        correct_per[index] = (right_predic/error.size)*100.
+        correct_per.append((right_predic/error.size)*100.)
     return [mean_error, max_error, correct_per, mean_deviation_abs,mean_deviation_per]
 
 def best_sample_id(result):
@@ -109,7 +108,7 @@ def export_vtk(error, grains, stress, label,path):
     
 if __name__ == '__main__':
     export_path = '/home/yk138599/Hiwi/damask3/UNet/output/result_14'
-    Training_data = torch.load(f'/home/yk138599/Hiwi/damask3/UNet/Trainingsdata/TD_norm_64_angles.pt.pt')
+    Training_data = torch.load(f'/home/yk138599/Hiwi/damask3/UNet/Trainingsdata/TD_norm_64_angles.pt')
     #Training_data = torch.load(f'/content/drive/MyDrive/Bachlorarbeit/Input/TD_norm_32_phase.pt')
 
     normalization = np.load(f'/home/yk138599/Hiwi/damask3/UNet/Trainingsdata/Norm_min_max_64_angles.npy', allow_pickle=True)
@@ -117,8 +116,10 @@ if __name__ == '__main__':
     model = UNet.UNet()
     device = UNet.get_default_device()
     model = UNet.to_device(model.double(), device)
-
+    dataloader = UNet.Create_Dataloader(f'/home/yk138599/Hiwi/damask3/UNet/Trainingsdata/TD_norm_64_angles.pt',batch_size=1)
+    dataloader = UNet.DeviceDataLoader(dataloader,device)
     model.load_state_dict(torch.load(f'/home/yk138599/Hiwi/damask3/UNet/output/V14/Unet_dict_V14.pth',map_location=torch.device('cuda')))
+
     result= dataset_evaluation( normalization = normalization, model = model, dataset = copy.copy(Training_data), threshold = 0.05)
     print(f'\t mean error over whole set: {result[0].mean():.4}%')
     print(f'\t max error average: {result[1].mean():.4}% and maximum {result[1].max():.4}%')