0

I'm just getting started with NN in tensorflow/keras, and need some basic guidelines for what to do.

I'm going to use pretrained model, and set of random images - totally random, few (~20) of them. Now, I want to evaluate how model is performing.

I would like to generate some plots or score that would show, lets say low values -> indicating that model is not trained to recognize this kind of objects, and high values when predictions are good.

What should I keep track of? Overall model prediction certainty? It's not done via model.validate - rigt? Validation is made with validation dataset with correct labels, right?

I know its a broad question, but any answer will help!

Nickname11
  • 101
  • 2
  • 2
    Does this answer your question? [What should I do when my neural network doesn't learn?](https://stats.stackexchange.com/questions/352036/what-should-i-do-when-my-neural-network-doesnt-learn) – mhdadk May 25 '21 at 22:50
  • It's a little hard to tell with the information you provided. What task are you aiming to solve? The metric to be used will differ depending on that answer alone. – Sean May 26 '21 at 06:57
  • @Seankala I'm aiming to do transfer learning. So I wanted to check, how. some pretrained model is performing with a dataset. To check whether it was trained (hence can recognise) to recognize similar objects, or whether new dataset is completely abstract to it. – Nickname11 May 26 '21 at 08:42

0 Answers0