So now that we've seen these, let's go back to the IMDB dataset that we used earlier
in this course. Here, I'll just use an embedding that I flattened before
it goes into the dense. My model will look like this, with a 171,533 parameters, and the performance
will be like this. It's nice accuracy,
but clear overfitting but it only takes about five
seconds per epoch to train. If I change this to use an LSTM, I'll now have only
30,129 parameters, but it will take
about 43 seconds per epoch. The accuracy is better, but there's still
some overfitting. If I try a GRU layer instead, with a GRU being
a different type of RNN, and I make it bidirectional, my network will have
a 169,997 parameters. My training time will fall
to 20 seconds per epoch, and my accuracy is again
very good on training, and not too bad on validation but again, showing
some overfitting. With a convolutional network, I'll have a 171,149
parameters and it only takes about six
seconds per epoch to get me close to 100 percent
accuracy on training, and about 83 percent
on validation, but again with overfitting.