Let's take a look at how our network is doing, from the following screenshot. When you inspect these graphs, keep a close eye on the scale on the y-axis. While the swings look dramatic, they aren't that big:
The first thing to notice here is that at epoch 1 the network is doing a pretty good job. After that, it rapidly begins to overfit. Overall though, I think our results are pretty good. At epoch 1, we're correctly predicting the sentiment about 86% of the time on the validation set.
While this case study covers many of the topics that we've discussed so far in the chapter, let's look at one more where we can compare using pre-trained word vectors for our embedding layer with word vectors we learn ourselves.