1

Following the iris classification problem in Tensorflow tutorials, I am trying to build a DNN classifier model for recognizing characters written in my handwriting. Came across doubts regarding the test data.

I assume that in normal machine learning test data is given as unlabelled. Why in tensorflow, test data should also be labelled? Is there any method to input unlabelled test data using DNN classifier or any other classifier in tensorflow?

silpa
  • 11
  • 1

1 Answers1

0

The purpose of the test set is to assess certain metrics of your classifier model (AUC, precision-recall) which your model has never trained on. The main purpose is to ensure you are not over fitting to your training set.

You will still need labels on your test set to run these metrics.

When you are running inference, these labels are not necessary.

Ouwen Huang
  • 1,037
  • 2
  • 9
  • 25