How many files to be uploaded/ epoch variation in training model

Course Q&A TensorFlow Developer Professional Convolutional Neural Networks in TensorFlow
1

history.pkl file? :frowning you mean there are two files to be uploaded for grading?

one more question my model is getting trained, what I can see the training percentage first increased and then reduced and again increased after 6 epoch, can this happen? or there is an issue??

@nilosreesengupta created :slight_smile:

2

Hello @Deepti_Prasad ,

  1. The history.pkl file stores the training and validation metrics for each epoch. It is used to track and monitor the model’s performance over the period of time.
    These are the metrics that help in tracking as per our assignment are :
  • Training accuracy
  • Validation accuracy
  • Training loss
  • Validation loss

The code to download the history file is already pre-coded at the below of the assignment. So you must have seen it got auto-downloaded.

  1. It’s not unusual to have fluctuating accuracy. It depends on several factors:
    Dataset complexity, Model architecture, optimization algorithm, Insufficient data,Overfitting, Underfitting , Convergence.

You have passed the C2W1 assignment, so you need not worry about it. But later on you will get assignments where you will need to upload .pkl file again and see such fluctuations.

With regards,
Nilosree Sengupta

3

actually I understood the epoch variation in the 2nd week of this tf1 course. why there is a variation in model training. data augmentation explains it more. so that helped understanding this doubt more. thanks again.

1 Like
4

Hello @Deepti_Prasad ,

You’re welcome!

With regards,
Nilosree Sengupta

1 Like