An Artificial Neural Network written using Keras / Tensorflow for prediction of handwritten digits. The Neural Network is based on and has been trained and evaluated using the MNIST Data Set of Handwritten Digits.
This repository contains a Flask Python Application which is intended to serve as a Back-End Server API used in conjunction with an Angular Client Application which can be found HERE!
The Flask App provides one end-point /uploads in which a function is called to make a prediction of the image passed in the request object and return that prediction to the client. The application employs Pillow and Numpy in order to manipulate the image converting it to grayscale, resizing and transforming to an array of pixels in prepartion for passing it to the Model for prediction.
The MNIST data set of handwritten digits is a subset of a larger set available from NIST. MNIST contains a training set of 60,000 samples and test set of 10,000 samples. The digits have been size-normailised and centered in a fixed-size image. This application employs the MNIST data set in order to train, evaluate and predict using a Keras Neural Network Model.
Example of MNIST Digits
A more detailed tutorial on Neural Networks can be found in another one of my repositories that is located HERE! The repository contains a Jupyter Notebook in which an artifical neural network is used to train a model for predicting species of Iris Flower. The same logic is applied in this application as it also deals with categorical data.
As this is a Python based application, Python is a must and can be downloaded from the Anaconda open distribution which includes all of the . If you are on OSX you can simply install Python using Homebrew.
brew install python3
And install additional dependencies such as SciPy and Pillow as needed using the pip package manager.
pip3 install scipy
pip3 install Pillow
In order to use the Model file minst-neural.h5 that is included in this repository you may also need to install h5py for dealing with the file format of the Model.
pip3 install h5py
Follow the instructions at the following URLs for installing these dependcies.
- Clone this repository
git clone https://github.com/damiannolan/mnist-neural-network.git
- Start the server application
python3 app.py
This repository already includes a h5 Model file - mnist-neural.h5. However should you want to rebuild the model using the associated Keras script please run python3 keras-mnist.py. For the purpose of speed and effiency the model is loaded using load_model in the app.py Flask script. Building a new Model everytime the server is started is unnecessary and will slow down the application starting.