Fast, lightweight neural net for ML/AI. Named for the word for "doe" in Gaulish. A doe is fast and lightweight, like this application.
This software does not yet have many features, but it works and can create and train neural networks and perform classification predictions.
Example usage (after checking out the code and running make):
bin/elantia create my_neural_net.ai --inputs 3 --outputs 2 --layers 2 --neurons 50 --function SELU
This will create a neural net with 3 input nodes, 2 output nodes, and 2 middle layers of 50 neurons each using the SELU function.
The neural net will be saved to a file named my_neural_net.ai for later reuse.
To train the net, use a command like the following:
bin/elantia train my_neural_net training_set.txt --iter 10000
The --iter parameter specifies how many training iterations.
The training data file should be in a format similar to:
0: 1.0 0.7 0.0
1: 0.0 0.7 1.0
0: 0.3 0.2 0.2
0: 0.6 0.4 0.3
0: 0.8 0.7 0.5
1: 0.1 0.2 0.5
1: 0.2 0.5 0.6
0: 0.9 0.1 0.3
1: 0.7 0.9 0.8
1: 0.6 0.8 1.0
The above data will train the neural net to identify rising number sequences as output 1 and falling sequences as output 0.
To make a prediction, you can then run:
bin/elantia predict my_neural_net 0.2 0.5 0.9
This should give an output of 1 with a confidence percentage.
Note: for both predictions and training, input values should never exceed the range of -0.5 to 1.0.
You can see the elantia ML engine in action using the color identification test, located at tests/color_recognition.
When run, this test will create a neural net, train it to seven saturated colors (red, orange, yellow, green, blue, violet, magenta),
and then have it categorize several randomly generated colors. The neural net will be saved as color_test.ai for reuse.
This file contains all that's necessary to identify your own categories among the seven options, for example:
bin/elantia predict color_test.ai 0.2 0.5 0.3
...which will most likely output 3, meaning green (the output values are a zero-based index). You can also train it on the provided color training data, or your own data, substituting the .txt filename below if necessary:
bin/elantia train color_test.ai tests/color_training_data.txt --iter 1000000