Skip to content

Generates goddess names using Python 3, TensorFlow & textgenrnn.

Notifications You must be signed in to change notification settings

jpw/ai-goddess-namer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ai-goddess-namer

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!

Hello python, my old friend

You will need to set up a virtualenv.

I used python3 and did this:

  • python3 -m venv --system-site-packages env
  • source env/bin/activate

then

which python

…should point at the python interpreter in the env dir.

Prereqs

Once you've got your venv up and running:

  • python -m pip install --upgrade pip
  • python -m pip install setuptools --upgrade
  • python -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.

Running

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.

Generating other names based on a different wordlist

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

Releases

No releases published

Packages

No packages published

Languages