A simple convolutional neural net to recognize digits from the Optical Recognition of Handwritten Digits dataset. Users draw digits on an interface and query the model to see what the neural net thinks the digit is.
arch.py
computes the model and saves it to disk. The current model contains a Conv2D layer with 100 filters, a MaxPooling2D filter with size 2, followed by a flatten layer, and a Dense layer with 1000 units.a.py
uses the flask package to query the neural net based on user input. To start the server, runexport FLASK_APP=a.py
, followed byflask run
. The server should start onlocalhost:5000
.
keras, numpy, and flask.