scesRate vs. mini-batch accuracy

[notalway...@gmail.com]

Dear Sir. 

Thank you again for this wonder code, you shared us. 

My question is that i am little bit confused with scesRate that is calculated for every iteration.

In some deep learning program, i have experienced, usually calculate mini-batch accuracy (accuracy of particular mini-batch at given iteration). (It does not require setting test data). 

The way i understood your code is that it uses test data (MNIST.I_test) during the training. Does this mean that it uses test data to calculate scesRate at each iteration?

If so, scesRate (calculated from tranining and test data) is different from mini-batch accuracy (calculated only from training data). Am I right??

Thank you 

[Hagay Garty]

You are right the scesRate is calculated on the test data only, after each training iteration. 

Iteration is not minibatch, its a higher value, (hundreds of mini-batches usually)

for example, after every 1000 training samples, the network is tested on 300 samples from a different set (not trained on) and accuracy is printed. 

the 1000 training samples can be trained in a 10 sample minibatch