Finch

· Yifei Sun

A simple neural net framework and visualizer that uses genetic networks to train.

Credit #

Idea from Boltzmann Brain by @pehringer.

Installation & Usage #

With Go installed, you can install the tool by running the following command:

1go get -u ysun.co/finch

To use as library, import the package:

1import "ysun.co/finch"

If you have a working Nix installation:

1nix run github.com/stepbrobd/finch

Source code: https://github.com/stepbrobd/finch

Running Finch #

Demo

Gates:

-input=2 -output=1 -hidden=2

2 input neurons, 1 output neuron, 1 hidden layer with 2 neurons

-population=128 -mutation=0.025

128 individules in the population, with 2.5% mutation rate

-example=./data/gates/input_data.csv -expected=./data/gates/{or,nor,xor}_label_data.csv

dataset paths

1finch -input=2 -output=1 -hidden=2 -population=128 -mutation=0.025 -example=./data/gates/input_data.csv -expected=./data/gates/{or,nor,xor}_label_data.csv

Remember to change which operation you want to run: OR, NOR, XOR.

Math:

-input=20 -output=19 -hidden=6,6,6,6

20 input neurons, 19 output neurons, 4 hidden layers with 6 neurons each

-population=32768 -mutation=0.01

32768 individules in the population, with 1% mutation rate

-example=./data/math/add_input_data.csv -expected=./data/math/add_output_data.csv

dataset paths

1finch -input=20 -output=19 -hidden=6,6,6,6 -population=32768 -mutation=0.01 -example=./data/math/add_input_data.csv -expected=./data/math/add_output_data.csv

MNIST:

-input=784 -output=10 -hidden=16,16

784 input neurons (28x28 greyscale images), 10 output neurons (numbers 1-10), 2 hidden layers with 16 neurons each

-population=4096 -mutation=0.1

4096 individules in the population, with 10% mutation rate

-example=./data/mnist/mnist_pixel_data_{32,64,128,256,512,1024,2048,4096,8192}.csv -expected=./data/mnist/mnist_label_data_{32,64,128,256,512,1024,2048,4096,8192}.csv

dataset paths

1finch -input=784 -output=10 -hidden=16,16 -population=4096 -mutation=0.1 -example=./data/mnist/mnist_pixel_data_{32,64,128,256,512,1024,2048,4096,8192}.csv -expected=./data/mnist/mnist_label_data_{32,64,128,256,512,1024,2048,4096,8192}.csv

Remember to change the size of MNIST dataset: 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192.


1# Use Glow to view this page in terminal
2$ glow https://static.ysun.co/txts/<path>.md