docs (model): corrected default for tf_epochs

aucmedi/data_processing/data_generator.py

@@ -300,7 +300,9 @@ def _get_batches_of_transformed_samples(self, index_array):
 
         # Stack images and optional metadata together into a batch
         input_stack = np.stack(batch_stack[0], axis=0)
-        if self.metadata is not None:
+        if self.metadata is not None and self.labels is None:
+            input_stack = [[input_stack, self.metadata[index_array]]]
+        elif self.metadata is not None and self.labels is not None:
             input_stack = [input_stack, self.metadata[index_array]]
         batch = (input_stack, )
         # Stack classifications together into a batch if available
@@ -414,4 +416,4 @@ def __set_index_array__(self):
 
     """ Internal function at the end of an epoch. """
     def on_epoch_end(self):
-        self.__set_index_array__()
+        self.__set_index_array__()

aucmedi/neural_network/model.py

@@ -178,7 +178,7 @@ def __init__(self, n_labels, channels, input_shape=None, architecture=None,
             However, be aware of unexpected adverse effects (experimental)!
 
         Attributes:
-            tf_epochs (int, default=5):             Transfer Learning configuration: Number of epochs with frozen layers except classification head.
+            tf_epochs (int, default=10):             Transfer Learning configuration: Number of epochs with frozen layers except classification head.
             tf_lr_start (float, default=1e-4):      Transfer Learning configuration: Starting learning rate for frozen layer fitting.
             tf_lr_end (float, default=1e-5):        Transfer Learning configuration: Starting learning rate after layer unfreezing.
             meta_input (tuple of int):              Meta variable: Input shape of architecture which can be passed to a DataGenerator. For example: (224, 224).

tests/test_datagenerator.py

@@ -175,9 +175,9 @@ def test_RUN_Metadata_noLabel(self):
         for i in range(0, 10):
             batch = data_gen[i]
             self.assertTrue(len(batch), 1)
-            self.assertTrue(len(batch[0]) == 2)
-            self.assertTrue(np.array_equal(batch[0][0].shape, (5, 224, 224, 3)))
-            self.assertTrue(np.array_equal(batch[0][1].shape, (5, 10)))
+            self.assertTrue(len(batch[0][0]) == 2)
+            self.assertTrue(np.array_equal(batch[0][0][0].shape, (5, 224, 224, 3)))
+            self.assertTrue(np.array_equal(batch[0][0][1].shape, (5, 10)))
 
     # Usage: Metadata for training
     def test_RUN_Metadata_withLabel(self):
@@ -245,4 +245,4 @@ def test_utils_iter(self):
             else: 
                 self.assertTrue(np.array_equal(batch[0].shape, (1,224,224,3)))
             counter += 1
-        self.assertTrue(counter == 4)
+        self.assertTrue(counter == 4)

tests/test_neuralnetwork_metadata.py

@@ -0,0 +1,136 @@
+#==============================================================================#
+#  Author:       Dominik Müller                                                #
+#  Copyright:    2024 IT-Infrastructure for Translational Medical Research,    #
+#                University of Augsburg                                        #
+#                                                                              #
+#  This program is free software: you can redistribute it and/or modify        #
+#  it under the terms of the GNU General Public License as published by        #
+#  the Free Software Foundation, either version 3 of the License, or           #
+#  (at your option) any later version.                                         #
+#                                                                              #
+#  This program is distributed in the hope that it will be useful,             #
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of              #
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
+#  GNU General Public License for more details.                                #
+#                                                                              #
+#  You should have received a copy of the GNU General Public License           #
+#  along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
+#==============================================================================#
+#-----------------------------------------------------#
+#                   Library imports                   #
+#-----------------------------------------------------#
+#External libraries
+import unittest
+import tempfile
+import os
+from PIL import Image
+import numpy as np
+#Internal libraries
+from aucmedi import *
+
+unittest.TestLoader.sortTestMethodsUsing = None
+
+#-----------------------------------------------------#
+#              Unittest: NeuralNetwork               #
+#-----------------------------------------------------#
+class NeuralNetworkTEST(unittest.TestCase):
+    # Create random imaging and classification data
+    @classmethod
+    def setUpClass(self):
+        np.random.seed(1234)
+        # Initialize temporary directory
+        self.tmp_data = tempfile.TemporaryDirectory(prefix="tmp.aucmedi.",
+                                                    suffix=".data")
+
+        # Create RGB data
+        self.sampleList_rgb = []
+        for i in range(0, 10):
+            img_rgb = np.random.rand(32, 32, 3) * 255
+            imgRGB_pillow = Image.fromarray(img_rgb.astype(np.uint8))
+            index = "image.sample_" + str(i) + ".RGB.png"
+            path_sampleRGB = os.path.join(self.tmp_data.name, index)
+            imgRGB_pillow.save(path_sampleRGB)
+            self.sampleList_rgb.append(index)
+
+        # Create classification labels
+        self.labels_ohe = np.zeros((10, 4), dtype=np.uint8)
+        for i in range(0, 10):
+            class_index = np.random.randint(0, 4)
+            self.labels_ohe[i][class_index] = 1
+
+        # Create metadata labels
+        self.labels_metadata = np.zeros((10, 5), dtype=np.uint8)
+        for i in range(0, 10):
+            class_index = np.random.randint(0, 5)
+            self.labels_metadata[i][class_index] = 1
+
+        # Create RGB Data Generator
+        self.datagen = DataGenerator(self.sampleList_rgb,
+                                     self.tmp_data.name,
+                                     metadata=self.labels_metadata,
+                                     labels=self.labels_ohe,
+                                     resize=(32, 32),
+                                     shuffle=True,
+                                     grayscale=False, batch_size=3)
+
+    #-------------------------------------------------#
+    #                  Model Training                 #
+    #-------------------------------------------------#
+    def test_training_pure(self):
+        model = NeuralNetwork(n_labels=4, channels=3, batch_queue_size=1, meta_variables=5)
+        hist = model.train(training_generator=self.datagen,
+                           epochs=3)
+        self.assertTrue("loss" in hist)
+
+    def test_training_iterations(self):
+        model = NeuralNetwork(n_labels=4, channels=3, batch_queue_size=1, meta_variables=5)
+        hist = model.train(training_generator=self.datagen,
+                           epochs=5, iterations=10)
+        self.assertTrue("loss" in hist)
+        self.assertTrue(len(hist["loss"]) == 5)
+
+        hist = model.train(training_generator=self.datagen,
+                           epochs=3, iterations=2)
+        self.assertTrue("loss" in hist)
+        self.assertTrue(len(hist["loss"]) == 3)
+
+    def test_training_validation(self):
+        model = NeuralNetwork(n_labels=4, channels=3, batch_queue_size=1, meta_variables=5)
+        hist = model.train(training_generator=self.datagen,
+                           validation_generator=self.datagen,
+                           epochs=4)
+        self.assertTrue("loss" in hist and "val_loss" in hist)
+
+    def test_training_transferlearning(self):
+        model = NeuralNetwork(n_labels=4, channels=3, batch_queue_size=1, meta_variables=5)
+        model.tf_epochs = 2
+        hist = model.train(training_generator=self.datagen,
+                           validation_generator=self.datagen,
+                           epochs=3, transfer_learning=True)
+        self.assertTrue("tl_loss" in hist and "tl_val_loss" in hist)
+        self.assertTrue("ft_loss" in hist and "ft_val_loss" in hist)
+
+    #-------------------------------------------------#
+    #                 Model Inference                 #
+    #-------------------------------------------------#
+    def test_predict(self):             
+        labels_temp = self.datagen.labels
+        model = NeuralNetwork(n_labels=4, channels=3, batch_queue_size=1, meta_variables=5)
+        hist = model.train(training_generator=self.datagen,
+                           epochs=3)
+
+        self.datagen.labels=None
+        for i in range(0, 3):    
+            batch = self.datagen[i]
+            self.assertTrue(len(batch), 2)
+        preds = model.predict(self.datagen)
+
+        self.assertTrue(preds.shape == (10, 4))
+        for i in range(0, 10):
+            self.assertTrue(np.sum(preds[i]) >= 0.99 and np.sum(preds[i]) <= 1.01)
+        self.datagen.labels=labels_temp
+        for i in range(0, 3):    
+            batch = self.datagen[i]
+            self.assertTrue(len(batch), 2)
+            self.assertTrue(np.array_equal(batch[0][0].shape, (3, 32, 32, 3)))
+        self.datagen.labels=labels_temp