Mean square error

Module 4. Linear Models Linear regression
#1

Hi!

there is something I do not get. This might be trivial because no one asked before :blush:

why do not we use the standard deviation instead of the mean squared error ?

#2

The standard deviation of which quantities?

#3

the standard deviation of error between the prediction and the expection.

# mean ( sqrt( (true_values - predictions)**2))

def goodness_fit_measure(true_values, predictions):

   *    error = np.ravel(true_values) - np.ravel(predictions)
   *    
   *    indicator = np.sum(np.sqrt(error**2))/len(np.ravel(true_values))
   *    
   *    return indicator
#4

The reason is that the linear regression model is minimizing this mean squared error. So it is interesting to look at this metric.

We could have used any other metric. Computing the std. dev. will give you another aspect of the error.