Mean Squared error in Python

Question

I'm trying to made function that will calculate mean squared error from y (true values) and y_pred (predicted ones) not using sklearn or other implementations.

I'll try next:

def mserror(y, y_pred):
    i=0
    for i in range (len(y)):
        i+=1
        mse = ((y - y_pred) ** 2).mean(y)   
        return mse

Can you please correct me what I m doing wrong with the calculation and who it can be fixed?

Answer 1

You are modifying the index for no reason. A for loop increments it anyways. Also, you are not using the index, for example, you are not using any y[i] - y_pred[i] , hence you don't need the loop at all.

Use the arrays

mse = np.mean((y - y_pred)**2)

Answer 2

I would say :

def get_mse(y, y_pred):
d1 = y - y_pred
mse = (1/N)*d1.dot(d1) # N is int(len(y))
return mse

it would only work if y and y_pred are numpy arrays, but you would want them to be numpy arrays as long as you decide not to use other libraries so you can do math operations on it.

numpy dot() function is the dot product of 2 numpy arrays (you can also write np.dot(d1, d1) )

Answer 3

firstly, you are using the i repeatedly and increments it but in range it is automatically iterative to next number. So don't use i again. The other thing that you are taking the mean of y but instead of taking mean of this, take the mean of ((y - y_pred) ** 2). I hope, you got the point.

Answer 4

Here's how to implement MSE in python:

def mse_metric(actual, predicted):
    sum_error = 0.0
    # loop over all values
    for i in range(len(actual)):
        # the error is the sum of (actual - prediction)^2
        prediction_error =  actual[i] - predicted[i]
        sum_error += (prediction_error ** 2)
    # now normalize
    mean_error = sum_error / float(len(actual))
    return (mean_error)