This program implements, with some minor deviations, the algorithm described in
Artificial intelligence: a Modern Approach third edition by Stuart Russel and
Peter Norvig (ISBN 978-0-13-604259-4). The program is implemented entirely in
mlp.go, and is annotated directly with comments describing how it
relates to the Back-Prop-Learning algorithm described in Chapter 18 (Figure
18.24).
This code is not "production ready". It has not been written with performance in mind, and it does not have the ability to save a model out to disk for later execution. It is purely a learning tool.
$ go run mlp.go --help
usage: nn [-h|--help] -i|--input-file <file> [-e|--epochs <integer>]
Simple neural network example.
Arguments:
-h --help Print help information
-i --input-file Input file to read NN spec from.
-e --epochs Number of epochs to train the neural network for.. Default:
1000
Example:
$ go run mlp.go -e 50000 -i and.json
training network...
50000 / 50000 [-------------------------------------------------------------------------------------------------------] 100.00% 317140 p/s
Running with input=[0 0], result=[0.013674204588149027]
Running with input=[1 0], result=[0.9894154921863585]
Running with input=[0 1], result=[0.995382389326916]
Running with input=[1 1], result=[0.99519137557522]
For a representative example, see and.json.
Inputs are provided in JSON format, and describe the learning rate, network structure, training examples, and inputs in a single file. The number of epochs to run for is provided by a CLI argument.