Training data from Semeion Handwritten Digit Data Set: https://archive.ics.uci.edu/ml/datasets/semeion+handwritten+digit
Training is completed by running train.py. Maximum success rate is found on a tuning set and the layer weights are saved into files named hiddenweights.txt and outputweights.txt
python train.py <parameter_file.json>
-d prints out statistics on loss and tuning set success rate on each epoch
-s prevents overwriting the saved hiddenweights.txt and outputweights.txt files
Prediction is completed by running predict.py
To run with an input image file:
python train.py <parameter_file> <hiddenweights_file> <outputweights_file> <inputfile>
To run with a hand drawn input using the mouse and an on-screen GUI:
python train.py <parameter_file> <hiddenweights_file> <outputweights_file> -d
Press or during drawing to run the prediction Press to reset the drawing
Load python 3.7 using: module load python/3.7.0
Install openCV using: pip3 install --user opencv-python
Add user packages using: module load python/0_user/python-site
Run training and prediction using the same command line arguments as above except use python3 instead of python