Just me messing around with TensorFlow 2 to generate goddess names based on the Wikipedia List of goddesses. Thanks Wikipedia!
Based on How to Train Your Own Neural Network by Beth Skwarecki. What a lovely article. Thanks Beth!
You will need to set up a virtualenv.
I used python3 and did this:
python3 -m venv --system-site-packages envsource env/bin/activate
then
which python
…should point at the python interpreter in the env dir.
Once you've got your venv up and running:
python -m pip install --upgrade pippython -m pip install setuptools --upgradepython -m pip install --upgrade tensorflow
Check the TensorFlow install:
python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
It should output... some stuff about tensorflow 😂
More info on installing TensorFlow can be found in the TensorFlow install docs.
Now install textgenrnn:
python -m pip install textgenrnn
Now you should be good to go.
python generate.py
It should output some debugging info from TensorFlow & textgenrnn, and then a list of new goddess names!
Have a look at the source for more clues on how to have fun with the output. The main fun-dial is the temperature setting: around 0.1 output is quite boring, while values over 1 get increasingly funky.
Don't forget to deactivate the venv when done playing around.
First, train it on your new wordlist. Reference your wordlist in the train.py script.
Run the training:
python train.py
In that script, the num_epochs variable effects the quality of the modelling—the higher the better, but it will take more time. Try a value of around 5 first.
That will output a .hdf5 file. Reference that .hdf5 file in generate.py, then run it:
python generate.py