How to use a CNN inside a for loop?

I have a net with 3 Convolutional Layers where I try to optimize the arrangement of used filters by trying out every possible combination between 1 and 5 filters per layer. I use a simple 3 layered for - loop. It takes 125 iterations to complete, while recording all validation accuracies inside a 5x5x5 tensor. This works with a few exceptions. Sometimes the training stops at the first iteration, as if my loss-threshold-criteria has been reached. After that it continous with the next iteration. Is this a known bug and is there a possiblity to overcome this hickup?

edit 08.06.2017: okay I solved it by using while loops and repeating the failed training attempts by setting a condition of minimum number of training epochs. Its not elegant but it works :)