Я выполняю задачу классификации, используя cnn во время обучения моей модели. Могу ли я обнаружить, что точность проверки увеличивается и уменьшается, а точность обучения увеличивается до максимума, а также потери при обучении и потери при проверке сильно отличаются? Может кто-нибудь сказать мне, почему точность проверки колеблется, как это? Являются ли эти колебания нормальными для самой тренировки?
Epoch 11/50
267/267 [==============================] - ETA: 0s - loss: 0.3756 - accuracy: 0.9615WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.3756 - accuracy: 0.9615 - val_loss: 0.8951 - val_accuracy: 0.7697
Epoch 12/50
267/267 [==============================] - ETA: 0s - loss: 0.7451 - accuracy: 0.9456WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 224ms/step - loss: 0.7451 - accuracy: 0.9456 - val_loss: 2.0827 - val_accuracy: 0.5076
Epoch 13/50
267/267 [==============================] - ETA: 0s - loss: 0.4142 - accuracy: 0.9617WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.4142 - accuracy: 0.9617 - val_loss: 0.9402 - val_accuracy: 0.7638
Epoch 14/50
267/267 [==============================] - ETA: 0s - loss: 0.4705 - accuracy: 0.9547WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.4705 - accuracy: 0.9547 - val_loss: 0.6366 - val_accuracy: 0.8744
Epoch 15/50
267/267 [==============================] - ETA: 0s - loss: 0.4692 - accuracy: 0.9621WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.4692 - accuracy: 0.9621 - val_loss: 0.9018 - val_accuracy: 0.7907
Epoch 16/50
267/267 [==============================] - ETA: 0s - loss: 0.3778 - accuracy: 0.9620WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.3778 - accuracy: 0.9620 - val_loss: 0.6718 - val_accuracy: 0.8565
Epoch 17/50
267/267 [==============================] - ETA: 0s - loss: 0.3352 - accuracy: 0.9653WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 59s 223ms/step - loss: 0.3352 - accuracy: 0.9653 - val_loss: 0.8769 - val_accuracy: 0.7693
Epoch 18/50
267/267 [==============================] - ETA: 0s - loss: 0.3756 - accuracy: 0.9644WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 59s 223ms/step - loss: 0.3756 - accuracy: 0.9644 - val_loss: 0.5441 - val_accuracy: 0.9067
Epoch 19/50
267/267 [==============================] - ETA: 0s - loss: 0.3297 - accuracy: 0.9675WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 60s 223ms/step - loss: 0.3297 - accuracy: 0.9675 - val_loss: 1.8063 - val_accuracy: 0.5415
Epoch 20/50
267/267 [==============================] - ETA: 0s - loss: 0.3349 - accuracy: 0.9681WARNING:tensorflow:Can save best model only with val_acc available, skipping.
267/267 [==============================] - 59s 222ms/step - loss: 0.3349 - accuracy: 0.9681 - val_loss: 1.3130 - val_accuracy: 0.6279