A Pickup Line Generator with Tensorflow – Towards Data Science – Medium

I began training on the Shakespeare data set and it initially went well for a few epochs (20–40) after which the accuracy stalled at 50% and the output was good but not great. After making sure the model output was reasonable I then trained on my own data set. At first I trained for 10 epochs to get a sense of where the model would land given the small data set. Again, the results were good, but not great. As an experiment, I trained for 200 some odd epochs until the accuracy did get very high, but the model began outputting lines it had seen before.