#triton #column #matrix #learning #modes #accuracy #mode

yanked triton_hydra

A branch of the triton project build with CUDA backend for matrix math

1 unstable release

0.0.1 Nov 16, 2023

#11 in #triton

MIT license

54KB
957 lines

hydra 🐍

A clone of the triton crate, with cuda bindings for GPU runtime of matrix multiplication and backprop

Installation

Use the package manager cargo to add triton to your rust project.

cargo add triton_hydra

or add the dependency directly in your cargo.toml file

[dependencies]
triton_hydra = "{version}"
serde = { version = "1.0", features = ["derive"] }
serde_json = "1.0"

Usage

Triton acts as a typical neural network implementation, but allows for a more dynamic way of solving problems you may not know how to solve. Acting as a 'brute force' approach to the world of deep learning, after n epochs in the training process triton will evaluate the specific error of each neuron and column, deciding whether to add a neuron to a column, add a new column entirely, remove a neuron or remove a column.

Triton will train and grow a desirable neural network until a specific accuracy is matched, returning the finished model

use triton_grow::network::{network::Network, activations, modes::Mode};

fn main() {
    let mut inputs = vec![vec![0.0,0.0],vec![1.0,0.0],vec![0.0,1.0],vec![1.0,1.0]];
    let mut outputs = vec![vec![0.0],vec![1.0],vec![1.0],vec![0.0]];
    let mut new_net: Network = Network::new(vec![2,3,1], activations::SIGMOID, 0.1);
    
    new_net = new_net.train_to_loss(inputs, outputs, 0.001, 100000, Mode::Avg, 0.001, 3, 10);
    println!("1 and 0: {:?}", new_net.feed_forward(&vec![1.0,0.0])[0].round());
    println!("0 and 1: {:?}", new_net.feed_forward(&vec![0.0,1.0])[0].round());
    println!("1 and 1: {:?}", new_net.feed_forward(&vec![1.0,1.0])[0].round());
    println!("0 and 0: {:?}", new_net.feed_forward(&vec![0.0,0.0])[0].round());
    println!("Net network made: {:?}", new_net.layers);

}

Proven Results

Upon testing Triton's self growth method against a traditional preconfigured network model. Three neural networks were all tasked with learning a simple XOR predictor with the following inputs and outputs:

Inputs

[ 1.0 , 0.0 ]
[ 0.0 , 1.0 ]
[ 0.0 , 0.0 ]
[ 1.0 , 1.0 ]

Outputs

[ 1.0 ]
[ 1.0 ]
[ 0.0 ]
[ 0.0 ]

Testing

Model Name Layers {input -[hidden] - output} Epochs Needed to Get 0.001 Avg Loss
Minimum 2 - { 3 } - 1 7,880,000
Well Fit 2 - { 3 - 4 - 3 } - 1 2,790,000
Triton 2 - { self growing } - 1 150,000

Triton was 98.09% more efficient than the minimum fit model, and 94.62% more than even the well fit model.

TODO

Currently, triton is in a very beta stage, the following features are still in development:

  • Mutating a neural network (1/2)
    • Adding a new layer with n neurons into any point of an existent network
    • Removing a layer from an existent network
  • Back propegation only affecting a single column (allows for a newly added layer to 'catch up')
  • Analysis mode during back propegation allowing for all individual errors to be recorded
  • Updated training function
    • Input desired success rate
    • Dynamic error analysis to allow for choosing if the network should grow or shrink
    • Acceptable threshold of +/- in the errors to allow for a less punishing learning process especially when a new neuron layer has been added
  • Model serialization (serde)

License

MIT

Dependencies

~5MB
~99K SLoC