Does Epoch matter in Exam?

In the exam, When I trained 10 epochs and submitted, I got 1/5 marks. And when I put 20 epochs, I got 2/5 marks. If I had put 100 epochs would I have got more marks? I am not sure if I changed any code between 10 and 20 epoch change. Hence this question. So if the model I submitted is trained more times(epochs), would that model give better results when someone else(in this tensorflow testing framework), tests it?

hi @JoeDL

the importance of epoch is dependent of your model architecture, your batch size, dataset. as you have posted your query in resources section I cannot state what was reason behind your marks variability. But if I assume the epoch importance comes with the number of times of your training your designed model based on your chosen parameters as well as other compile statement.

a much simple model might train with less number or epoch if you have significant batches from the dataset based on the features of your dataset but if you are having a complex model with multiple layer of architecture and has numerous features and larger batch, then running an epoch might yeild you better outcome.

kindly let us know which assignment you did, so your query is shifted to the correct course and week category, you can mention assignment name too.


it was house hold electric power consumption. a Time series question and the highest marks question(5)

Are you referring to tensorflow developer certificate?


please share a screenshot of your submission

Are you by chance asking this question while taking the exam?

1 Like

haha I wish. Actually I failed the exam. And this question I only got 2/5.

1 Like

what you mean?

Sorry to hear that. Time series problem is the hardest among the 5 questions.

Do see tensorflow developer specialization which talks about time series data from certification perspective.

Mentors don’t have access to the tensorflow developer exam grader. Here are my observations based on experience:

  1. The grader expects your model prediction to not have error beyond a certain threshold. The more errors / deviation your model makes, lower the score.
  2. Number of epochs don’t matter since the grader cares only about the saved model file containing model weights. For your model to score well, training just for a few epochs is not a good idea. Be sure to have access to a good GPU for the exam.
  3. Grader doesn’t want a submission to use lambda layers.
1 Like