0

I have to implement a loss function using mean absolute error without calling the built-in function. Is the code below correct? Because my loss value goes from 28.xx to 0.00028 quickly.

Meanwhile, other loss function like RMSE has a more standard loss curve

loss = tf.reduce_sum(tf.abs(y_pred - Y) / nFeatures)
Dude1234
  • 51
  • 2
  • 10

1 Answers1

2

You can implement your own lost function base on MAE formula: img

import tensorflow as tf
MAE = tf.reduce_mean(tf.abs(y_true - y_pred))

Also you can check customized loss function in this answer

or

import numpy as np
MAE = np.average(np.abs(y_true - y_pred), weights=sample_weight, axis=0)

or

from tensorflow.python.ops import math_ops
MAE = math_ops.abs(y_true - y_pred)
Mario
  • 1,631
  • 2
  • 21
  • 51