goAdaline

GoAdaline

GoAdaline is a simple Go implementation of an Adaline neuron using the Delta Rule.

This project is helpful to demonstrate how Neural Networks work to those starting in Artificial Intelligence.

If you want to use a more advanced implementation using gonum you can checkout the gonum branch of the repository.

Getting Started

To try it out you just need to have Go installed. I used Go 1.9 so I can't assure that will work with lower versions but it should work fine.

To install it just run:

go get https://github.com/Plorenzo/goAdaline

Prerequisites

The dataset has to be in csv format and the last column of it must be the expected output.

Also you need to have the data set in a folder divided into 3 files:

• train.csv
• validate.csv
• test.csv

For best results train.csv should have about 70% of the rows and 15% the other files.

You can use this sample data to test the neuron. It has been normalized and randomized before splited into files.

Sample data

• Train data: https://pastebin.com/H3YgFF0a
• Validate data: https://pastebin.com/aeK6krxD
• Test data: https://pastebin.com/mt5P8AZS

You can find out more info about the dataset here http://sci2s.ugr.es/keel/dataset.php?cod=44

Flags

Required flags for the program to run:

-path       "/path/to/folder"
-cycles     100                     //nº of training cycles
-lr         0.1                     //Learning rate
-out        "path/to/destination"

Example:

./goAdaline -path=/Users/plorenzo/dev/data/ -cycles=100 -lr=0.1

Output

The neuron will output the squared error obtained in the test data and the weights used.

Test error:
0.017166281479163065
Weights:
[0.6283026308947595 0.5098346273068519 0.2182228283924059 -0.2505429228543444 0.11335767250598956 0.03879773117815263 0.09743364005207034 0.5027576658656207 -0.011776841084895283]

Also a csv file with the train and validate error for each cycle, the neuron output for the test data and the weights used. So you can make cool charts like this:

TODO's

Features to add, bugs to fix, improvements to make:

• Add ability to pass file paths from terminal
• Also cycles and learning rate
• Refactor readCSV() to convert to float64 whiles reads SO question
• Add optional flag for path to error output csv
• Print final weights
• Make readCSV() calls concurrent with gorutines