optim.lua

local optim={}
local matrix = require "matrix"
local cmath = require "cmath"
local perceptron = require "perceptron"

function optim.gradient_descent(X,target,weights,learning_rate)
	local m = X.size[1]
	local n = X.size[2]
	local updated_weights = matrix.new(weights.size[1],weights.size[2])
	local X_t = X:transpose()
	local scaling_factor = learning_rate/m
	local disparity = (X*weights) - target
	local descent = scaling_factor * (X_t * disparity)
	updated_weights = weights - descent
	return updated_weights
end
function optim.cost_mse(X,target,weights)
	local disparity = (X * weights) - target
	local sum_dis = (cmath.sum(cmath.pow(disparity,2)))/X.size[1]
	return sum_dis
end

function optim.cost_LOGISTIC(X, target, weights)
	local hypothesis = X * target
	local hypothesis_com = 1 - hypothesis
	local term_1 = y:hadamardProduct(hypothesis)
	local term_2 = (1 - y)
	term_2 = term_2:hadamardProduct(hypothesis_com)
	local sum =cmath.sum((term_1 + term_2))
	sum = sum/X.size[1]
	return sum
end

math.randomseed( os.time() )
local input = 10 * matrix.random(4,3)
local neuron = perceptron:new(3,0,cmath.sigmoid)
neuron.weights = matrix.random(3,1)
local y = neuron:activate(input)