NN_from_scratch_MNIST

A NN implementation based on the MNIST dataset for Handwritten number recognition including a GUI build-arround.

Usage:

• Unpack Training and testing data into a folde /data
• Run start.py and use further commands:

You can work with this model from the command line. The following command are available:

• "create 'number of neurons in hidden'" : create a new model
• "load pre" : loads pre trained weights for the model
• "train 'number of epochs'" : train the model based on the MNSIT dataset
• "evaluate" : Run the model on the MNIST test data
• "save" : Save your weights as txt files.
• "load own": loads previously saved weights for the model
• "draw" : opens a canvas to draw and submit numbers to the model
• "quit" : close the program