Make your own text generator
For this exercise you’ll need to be running our Star Trek Episode Title Generator either in Colab or on your own computer.
Notebook: Star Trek Titles charRNN (Github) (open in Colab)
- Start by opening the example code for inventing Star Trek episode titles. (try the advanced version which is less verbose)
- Step through the code, and make sure that you can get it to generate some interesting text.
- Try changing the diversity/temperature to produce more “normal” and more “weird” results.
- Now it gets interesting. Think of some data that YOU would like to generate which can be represented as text. This could be actual text, or something else (images, or sound) represented as text characters. Almost everything in computing can be represented as text, so there’s a lot of scope for experimentation! Change the LSTM text generator to learn from your dataset and to generate new creative data.
- Think about how your creative data generator could be applied in a work of art. Could it be integrated into a website, or some kind of art installation? Come up with a artistic concept that relies on genuine data generation. What questions would this work ask of technology, and of us?