A Dann model learns to predict pokemon combat winners.
Try it out!
Here are the values trained to the neural network (48 999 samples trained)
Inputs:
0: Pokemon 1 element 1 (Value mapped to each element)
1: Pokemon 1 element 2 (Value mapped to each element)
2: Pokemon 1 legendary (Boolean)
3: Pokemon 1 HP (Normalized)
4: Pokemon 1 Attack (Normalized)
5: Pokemon 1 Defense (Normalized)
6: Pokemon 1 Sp Attack (Normalized)
7: Pokemon 1 Sp Defense (Normalized)
8: Pokemon 1 speed (Normalized)
9: Pokemon 2 element 1 (Value mapped to each element)
10: Pokemon 2 element 2 (Value mapped to each element)
11: Pokemon 2 legendary (True or False)
12: Pokemon 2 HP (Normalized)
13: Pokemon 2 Attack (Normalized)
14: Pokemon 2 Defense (Normalized)
15: Pokemon 2 Sp Attack (Normalized)
16: Pokemon 2 Sp Defense (Normalized)
17: Pokemon 2 speed (Normalized)
Output:
0: Winner (Boolean)
The dataset can be found [here](https://www.kaggle.com/tuannguyenvananh/pokemon-dataset-with-team-combat)
Run the webapp on a local server
npm start
Parse dataset (csv to json)
npm run parse
Train the model, this command will create a new model if you haven't trained one yet. If you allready saved a model, this command will train the model with the most epochs (from file name).
npm run train [epochs]
Minify the latest model to a JS function in public/scripts/minifiedModel.js
npm run minify