2

Previous developer apply neural network and give me result of loss, MSE and MAE. How do I compare these result with my models (Linear Regression)? I can calculate MSE and MAE, but what is loss?

J.Doe
  • 21
  • 2
  • _Loss_ is a general term used to refer to different types of functions for evaluating a model. While MSE might be a very likely candidate, it could also be something like log-likelihood or MAE, as you mention yourself. So it is difficult to say what exactly it is without some additional information. Do you have access to the original model code, or even the saved model? Or is it just the loss value? – user1953384 Aug 06 '20 at 07:29

1 Answers1

0

Loss is the "cost function" that is being optimized. The metric is assigned (i.e. you can set the loss to MSE, MAE, acc=accuracy, etc.).

You can take a look at the .compile line of your code to see what it is set as.

https://www.tensorflow.org/api_docs/python/tf/keras/Model#compile

https://www.youtube.com/watch?v=0twSSFZN9Mc

user2529589
  • 330
  • 4
  • 16
  • The loss value can be MSE and MAE, but this previous developer left me a database with 3 values of (loss, MSE, MAE). So he was wrong, am i right? – J.Doe Aug 06 '20 at 03:31
  • If all you have is the database, I would say he was wrong not to leave comments or a data dictionary for you to identify what that variable is. If you have access to the code (or developer), it wouldn't be too hard to figure out what metric the loss function was optimized on. If not, (if I were you), I would just ignore the column all together and use MSE or MAE. Not to make the problem any more complicated, but I would actually be more concerned with how the dataset was validated (Cross validation? stratification?) . Your model would need to use the same splits to compare apples to apples. – user2529589 Aug 06 '20 at 08:11