Training and model generation code for pokedem-plus
api.py implements an entry point for the API that exposes a single trained model via /predict?features=[x1,x2,x3,...] method
python api.py --input model
Where model is produced by training scripts described below
train.py implements multiple dense layers with tanh as an activation
function with a softmax on top. Training set should be in SVM format.
python train.py --train features.svm
-
Install following ubuntu packets (or use their counterparts in other systems/build from sources):
- libblas-dev
- liblapack-dev
- gfortran
- python-numpy
- python-scipy
- python3-numpy
- python3-scipy
-
Install python dependencies:
pip install -r requirements.txt -
Install Tensorflow for your system using the official guide