priyap97 / number-recognition Goto Github PK

this file was designed to be only functions that can be called in different files for ease of use/modularization but right now it is a script. All code should be in a function

• pretty graphs for report
• how to decode neurons & provide concrete output
• get numbers about accuracy

• move functions to data_utils.py
• ensure that it still runs in the new location

• 784 input neurons
• 1000 neurons in ONE hidden layer
• 10 neurons in the output layer

each layer should be fully connected to the next, initialized with weights of zero
the goal is that this will be done so that we can all jump on learning

Add comments so that it's clear what the Nengo statements do - to make for easier collaboration/training and documentation in the report

• figure out conceptually how to implement PES using Nengo
• code not yet required but that's the next step

Use existing functions in data_utils.py

Remove extra comments, leave documentation

• Restructure neural network as needed
• Finish research
• Write code so that the network can be trained

Depends on driver being fixed for code to get written

Currently, the driver does not run - looks like some confusion between data_utils and number_recognition. This relies on data_utils being refactored, but training relies on this issue being closed so that we can start to test

• specific work distribution TBD