Linear Regressor

linear_regressor.py

class LinearRegressor(Object):
  
    def gradient_descent(self, x, y, alpha, max_iterations):
        num_rows = x.shape[1] + 1
        w = np.array([random.random() for item in range(num_rows)]).reshape(num_rows,1)
        
        best_weights = []
        lowest_cost = None

        iter = 1
        for k in range(0, max_iterations):
            gradient = self.compute_gradient(x, y, w)
            cost = self.compute_costval(x,y,w)

            w = w - alpha*gradient

            if lowest_cost is None or cost < lowest_cost:
                best_weights = w
                lowest_cost = cost

        return best_weights
  
    def compute_gradient(self,x,y,w):
        P = len(y)
        grad = 0
        for p in range(P):
            x_p = list(x[p])
            x_p.insert(0,1)
            x_p = np.array(x_p)
            x_p.shape = (len(x_p),1)
            y_p = y[p]
            temp = np.dot(x_p*x_p.T,w) - x_p*y_p
            grad+=2*temp
        return grad

    def compute_costval(self,x,y,w):
        P = len(y)
        cost = 0
    
        # run over all data points and weights and compute total error
        for p in range(P):
            x_p = x[p]
            y_p = y[p]
            cost += (w[0] + np.dot(x_p,w[1:]) - y_p)**2
        return cost