What is the difference between validation percentage and batch size?

Question

I'm doing transfer learning using Inception on Tensorflow. The code that I used for training is https://raw.githubusercontent.com/tensorflow/hub/master/examples/image_retraining/retrain.py

If you take a look at the Argument Parser section at the bottom of the code, you will find these parameters :

• testing_percentage
• validation_percentage
• test_batch_size
• validation_batch_size

So far, I understand that testing and validation percentage is the number of images that we want to train at 1 time. But I don't really understand the use of test batch size and validation batch size. What is the difference between percentage and batch size?

Answer 1

The percentages refer to the number of samples to use (out of full dataset) as the validation and test datasets. So if you pass a dataset that consists of 100,000 samples to the model and set the validation and testing percentages to be 10% each, your model will train on 80,000 samples, validate them on 10,000 and save additional 10,000 samples for the final test.

The batch sizes refer to the number of samples in each batch during the test and validation evaluations. Your model probably can't process 10,000 samples in a single run (due to memory limitations) so during evaluation it breaks the dataset into batches, which are processed sequentially and the result is the mean of all batches.

When you are training, the batch size is an important hyper-parameter which has an affect on the properties and final results of the training process. During test/validation it has no affect and only needs to be small enough for your model to be able to run it (evaluation with different batch sizes will produce the same results).