neural_lib/

lib.rs

#![no_main]

//! # Neural Lib
//! In Progress !!! -> Some fixes in load  
//! This is a small library to make group of neurons.
//! Every group is create with own parameters and the functions required to make predictions and training

/// A Neuron is defined by a weight and an impact. Every value passed in is change by the property of the neuron
#[derive(Clone,Copy)]
struct Neuron{
    weight: f64,
    bias:f64,
    output:f64
}
impl Default for Neuron{
    fn default() -> Self {
        return Neuron{weight:0.5,bias:0.5,output:0.0}
    }
}
impl Neuron{
    fn new(weight:f64,bias:f64)->Self{
        return Neuron{weight:weight,bias:bias,output:0.0}
    }
}

/// Main Struct to make a Layer of neural network
/// It contains few Neuron struct and the activation function for them
pub struct Layer{
    neurons:Vec<Neuron>,
    act:Activate
}
impl Layer{
    /// Add layer, need to a vec of Neurons parameters(tuple f64,f46) and type of function
    pub fn new(fnact:&Activate,param:&Vec<(f64,f64)>)->Self{ 
        let mut network = Layer{
            neurons: Vec::with_capacity(param.len() as usize),
            act:fnact.clone()
        };
        for i in param{
            network.neurons.push(Neuron::new(i.0,i.1));
        }
        return network
    }
    /// Add layer, need to define number of neurons and type of function
    pub fn default(x:u32,fnact:&Activate) -> Self {
        let mut layer = Layer{
            neurons: Vec::with_capacity(x as usize),
            act:fnact.clone()
        };
        for _i in 0..x{
            layer.neurons.push(Neuron::new(0.5,0.5));
        }
        return layer
    }
    /// Calculate the Weighted Sum 
    fn sum_pond(&mut self,x:&Vec<f64>)->Vec<f64>{
        let mut prop:Vec<f64> = vec![];
        for (id,n) in self.neurons.iter().enumerate(){
            prop.push(x[id] * n.weight + n.bias); 
        }
        return prop
    }
}

#[derive(Clone)]
/// Enum to choose your Activation function, by default => Sigmoid function
pub enum Activate{  // Choose your function to activate neurons
    Sig,
    Rel,
    Tan
}
impl Default for Activate{  // Default Status => Sigmoid function
    fn default()->Self{
        return Activate::Sig
    }
}

fn activate(net:&mut Layer,x:&Vec<f64>)->Vec<f64>{
    let mut sum = net.sum_pond(x);
    for (id,r) in sum.iter_mut().enumerate(){
        match net.act{
            Activate::Sig=>{sigmoid(r)},
            Activate::Rel=>{relu(r)},
            Activate::Tan=>{tanh(r)}
        }
        net.neurons[id].output = *r;
    }
    return sum
}

fn sigmoid(x:&mut f64){
    *x = 1.0/(1.0+(-*x).exp());
} 

fn relu(x:&mut f64){
    if *x < 0.0{
        *x = 0.0;
    }
}

fn tanh(x:&mut f64){
    *x = ((2.0 * *x).exp()-1.0) / ((2.0 * *x).exp()+1.0);
}

/// Define a Neural Network
pub struct Network{
    pub inputs:Vec<f64>,
    layers:Vec<Layer>,
    pub output:Layer
}
impl Network{
    pub fn new(nb_lay:usize,lays_params:Vec<Vec<(f64,f64)>>,lays_fn:Vec<Activate>)->Network{
        let mut lays:Vec<Layer> = Vec::new();
        for l in 0..nb_lay{
            lays.push(Layer::new(&lays_fn[l], &lays_params[l]));
        }
        let mut output = vec![];
        output.push((0.5,0.5));
        return Network { inputs:vec![],layers: lays,output:Layer::new(&Activate::Sig,&output)};
    }
    pub fn default(nb_lay:usize,nb_n:u32,lays_fn:Vec<Activate>)->Network{
        let mut lays:Vec<Layer> = Vec::new();
        for l in 0..nb_lay{
            lays.push(Layer::default(nb_n,&lays_fn[l]));
        }
        return Network { inputs:vec![],layers: lays,output:Layer::default(1, &Activate::Sig)};
    }

    /// Launching a prediction with values -> This function can be call manually or by the training session
    pub fn prediction(&mut self)->Vec<f64>{   
        let mut res = self.inputs.clone();// Call a result on an entry
        // Propagation on hidden layers
        for layer in self.layers.iter_mut(){
            res = activate(layer,&res);
        }
        // Propagation on output layer
        res = activate(&mut self.output, &res);
        return res
    } 

    /// Initiate a training session
    /// You can specifie the inputs and the number of repetitions from the nb variable
    pub fn train(&mut self, inputs:&Vec<Vec<f64>>,targets:&Vec<f64>,learning_rate:f64,nb:usize){
        for _ in 0..nb{  
            for (input,target) in inputs.iter().zip(targets.iter()){
                self.inputs = input.to_owned();
                self.train_single(*target, learning_rate);
            }
        }
    }

    fn train_single(&mut self, target:f64, learning_rate:f64){
        let output = self.prediction();
        let error = target - output[0];

        for neuron in self.output.neurons.iter_mut(){
            let neuron_error = neuron.weight * error * neuron.output * (1.0 - neuron.output);
            neuron.weight += learning_rate * neuron_error * neuron.output;
            neuron.bias += learning_rate * neuron_error;
        }

        for layer in self.layers.iter_mut().rev(){
            for neuron in layer.neurons.iter_mut(){
                let neuron_error= neuron.weight * error * neuron.output* (1.0-neuron.output);
                neuron.weight += learning_rate * neuron_error * neuron.output;
                neuron.bias += learning_rate*neuron_error;
            }
        }
    }
    /// Generate a file which contains the parameters of your neurons and layers
    pub fn output(&self,n_file:&str)->Result<(),Error>{
        match fs::OpenOptions::new().create_new(true).write(true).open(n_file){
            Ok(mut f)=>{
                f.write_all("Schematic Neural Network\n".as_bytes());
                for (id,layer) in self.layers.iter().enumerate(){
                    f.write_all(format!("HiddenLayer :{}\n",id).as_bytes());
                    match layer.act{
                        Activate::Sig=>{f.write_all("Sigmoid fn\n".as_bytes());}
                        Activate::Rel=>{f.write_all("ReLu fn\n".as_bytes());}
                        Activate::Tan=>{f.write_all("Tangente fn\n".as_bytes());}
                    }
                    for n in layer.neurons.iter(){
                        f.write_all(format!("[{}:{}] ",n.weight,n.bias).as_bytes());
                    }
                    f.write_all("\n".as_bytes());
                }
                f.write_all(format!("OutputLayer\n").as_bytes());
                for n in self.output.neurons.iter(){
                    f.write_all(format!("[{}:{}] ",n.weight,n.bias).as_bytes());
                }
                f.write_all("\n".as_bytes());
                return Ok(())
            },
            Err(err)=>{
                return Err(err)
            }
        }
        
    }
}

enum Class{
    Binary,
    MultiClass
}

use std::{fs, io::{Error, Read, Write}};