Libraries to generate and manage a multilple layers Convolutional Neural Network. Classification problem can be solved such as recognition of image or pattern or even topology optimzation. The number of layers and neurons are let to the user during the creation of the network.
Use your Cargo package manager adding to its config file the latest verison available:
[dependencies]
cnnks = "xxx"You can generate and train the CNN via:
let mut m_l = cnnks::MultiLayerPercetron::new(7, 10, vec![10,10], 1.0, 0.50);
let init_weight = 0.50;
m_l.set_weight(init_weight);
println!("\n//------------------------------- NN -------------------------------------- \n");
println!(
"The output should be 8 however the output has this noise: \n {:?} \n",
m_l.run(Vec::from([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]))
.clone()
);
// Starting the training procedure
for i in 0..2000 {
let mut mse = 0.0;
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0]),
Vec::from([1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]))?; // 0 pattern
mse += m_l.back_propagation(
Vec::from([0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]),
Vec::from([0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]))?; // 1 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0]),
Vec::from([0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]))?; // 2 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]))?; // 3 pattern
mse += m_l.back_propagation(
Vec::from([0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]))?; // 4 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]))?; // 5 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0]))?; // 6 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]))?; // 7 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]))?; // 8 pattern
mse += m_l.back_propagation(
Vec::from([1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0]),
Vec::from([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]))?; // 9 pattern
if i % 100 == 0 {
println!("MSE = {}", mse);
}
}
// Run the NN after the training procedure
println!("\n------------------ After training ---------------------- \n");
let result_after_training = m_l.run(Vec::from([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]));
println!("{:?}\n ", result_after_training);The test unit give as input a numberpads which mean a number from 0 to 9. During the training procedure several numberpads segments pattern are exposed, telling the corrisponding given number. The NN are the beginning will garbage since the weight are set by the user but after the training it should return the number the numberpad meant. It is possible to execute the test via:
cargo test