Hi. I’m trying to conclude this course. I already defined a NN which achieves 99.9% of accuracy on 4 epochs. Below is the history of training procedure.
Epoch 1/25
2/2 [==============================] - 1s 307ms/step - loss: 0.8358 - accuracy: 0.5000
Epoch 2/25
2/2 [==============================] - 1s 296ms/step - loss: 0.9179 - accuracy: 0.8000
Epoch 3/25
2/2 [==============================] - 1s 299ms/step - loss: 0.4438 - accuracy: 0.7500
Epoch 4/25
2/2 [==============================] - ETA: 0s - loss: 0.2016 - accuracy: 1.0000
Reached 99.9% accuracy so cancelling training!
2/2 [==============================] - 0s 288ms/step - loss: 0.2016 - accuracy: 1.0000
However, after I submit the assignment, the grade of the model is set to 0/50 due to
Failed test case: model was originally trained for more than 20 epochs.
Expected:
20 epochs at most,
but got:
25.
I tried different architectures and several hyper-parameters, and the problem persists: The number of epochs taken in my jupyter is always less than 25 and, for grading, it tooks longer.
Any thoughts?