why im getting 1.000 val_acc and below 95% acc
If the validation data comes from a distribution that model is good at classifying, then, validation accuracy will be higher than training accuracy. Consider changing the model architecture to improve training performance.
Please see this note from the assignment:
To pass this assignment your model should achieve a training accuracy of at least 95% and a validation accuracy of at least 80%
the solution to increase the accuracy:
- Don’t use
MaxPool2D. It’s deprecated in favor ofMaxPooling2D. (tf.keras.layers.MaxPooling2D) - Keep batch size 32. (batch_size=32,)
- Increase number of conv filters as you go deeper in the network. (add the 4th and 5th conv layers) → (check them at c1 w4 ungraded lab 3 from github)