Game Ideas Generation Using Neural Networks
- Pyhton 3.8
- Numpy 1.2
- TensorFlow 2.8
- Keras 2.8
Before you can generate ideas for games you must train the neural network:
python MLGamedevIdeas.py -train
This will train the neural network with all the descriptions in the file 'final_data_new.json.gz' (about 80000) and 20 epochs. Depending on your hardware this may take quite a while. To make the training process less time consuming you can use fewer game descriptions:
python MLGamedevIdeas.py -games 5000 -train
You can also use less epochs:
python MLGamedevIdeas.py -games 5000 -epochs 10 -train
For good results I recommend values of 'loss' below 2. Consult the parameters to better adjust the training:
python MLGamedevIdeas.py -h
Once the neural network has been trained, a file 'weights.hf5' will have been generated.
Note that you will have to retrain the neural network if you change the parameters.
Now you can start generating crazy ideas:
python MLGamedevIdeas.py
Code released under MIT License.