Number Recognition AI

A convolutional neural network that can read handwritten numerals and accurately predict what they are.

Please do not expect this to be perfect, it has undergone minimal updates/patches and will occasionally falsely predict the number given.

---

Usage

CLI Arguments

usage: python -m make_predicton [image file path]

Some example images have been provided in the repository.

---

How to use logs

You will need TensorBoard installed for this step.

If you don't have it yet, you can install it by running pip install tensorboard in your terminal.

• Once you have TensorBoard installed, run your prediction, and then open up a terminal window and run tensorboard --logdir=[log path]
• The prediction that you just ran will have produced log files which will be in the directory you create for them. In model.py the callback is configured for use of a directory ./logs but you can change that to fit your own needs by simply editing the TensorBoard instance. Use this in place of [log path]
• Open up a browser and go to localhost:6006 to view your logs.