The 3 most common loss functions for Machine Learning Regression

 (.env) boris@boris-All-Series:~/HUBER$ cat HuberLoss.py

import matplotlib.pyplot as plt

import numpy as np

# MAE loss function

def mae_loss(y_pred, y_true):

    abs_error = np.abs(y_pred - y_true)

    sum_abs_error = np.sum(abs_error)

    loss = sum_abs_error / y_true.size

    return loss

# Plotting

x_vals = np.arange(-100, 100, 0.01)

y_vals = np.abs(x_vals)

plt.plot(x_vals, y_vals, "red")

plt.grid(True, which="major")

plt.draw()

# MSE loss function

def mse_loss(y_pred, y_true):

    squared_error = (y_pred - y_true) ** 2

    sum_squared_error = np.sum(squared_error)

    loss = sum_squared_error / y_true.size

    return loss

# Plotting

x_vals = np.arange(-20, 20, 0.01)

y_vals = np.square(x_vals)

plt.plot(x_vals, y_vals, "blue")

plt.grid(True, which="major")

plt.draw()

# Huber loss function

def huber_loss(y_pred, y, delta=1.0):

    huber_mse = 0.5*(y-y_pred)**2

    huber_mae = delta * (np.abs(y - y_pred) - 0.5 * delta)

    return np.where(np.abs(y - y_pred) <= delta, huber_mse, huber_mae)

# Plotting

x_vals = np.arange(-65, 65, 0.01)

delta = 1.5

huber_mse = 0.5*np.square(x_vals)

huber_mae = delta * (np.abs(x_vals) - 0.5 * delta)

y_vals = np.where(np.abs(x_vals) <= delta, huber_mse, huber_mae)

plt.plot(x_vals, y_vals, "green")

plt.grid(True, which="major")

plt.show()