Skip to content

Latest commit

 

History

History
26 lines (16 loc) · 1.13 KB

usage.md

File metadata and controls

26 lines (16 loc) · 1.13 KB

Using the program

If you're missing permissions, do: chmod +x bin/*

When this project is finished, it will have two programs: train and classify. This is a WIP.

The train will allow you to train a neural network, and then dump all weights of all Synapses that connect Neurons into a file.

You might use it like this:

./bin/train data/training/and_gate.json data/networks/and_gate.json

The first arg is where to load the labeled training data. The second is where the resulting network will be saved.

When you run the program, the network will have a total_error, although the value printed will be the average for the entire epoch (all training documents). That average should approach zero until it reaches the total_target_error specified in the training file. If it doesn't approach zero, try amending the settings at the top of the training data file.

Next, the classify program:

./bin/classify data/networks/and_gate.json data/new_data/and_gate.json

The first arg is the network that was saved during training. The second is the data that is used with a trained classifier. This might be test data or unseen data.