Neural Network Training

The training can be done with either Pytorch or Keras, but Pytorch is recommended because of the unreliable API of Tensorflow (the foundation of Keras)

Build and Requirements

Make sure you install all requirements:

pip install -r requirements.txt

Additionally for verification (we don't trust Tensorflow/Keras/Pytorch blindly) the Neural Network Extension package must be installed. The source code is in the python folder (from the repository root). There are also precompiled python wheels on Github available.

To train the network simply run in a terminal:

python train_torch.py
python quantize.py

or use the Jupyter notebook to train the network.

Training

Results

For quantisation fixed point quantisation has been used.

Network	Accuracy
Float	0.9832
Fake Quant	0.9832
Quant:	0.9349