How to set up ESIONN correctly?

Question

Neural networks

Hello, I actually implemented ESOOINN and ran into a problem that it learns differently every time it starts, that is, the classification results differ significantly from time to time. Of course, I understand that a lot depends on the initial initialization of the network nodes, but this happens in any network, probably)
Well, the questions are:
1) what should be done to get at least slightly similar results.
2) What should be paid attention to in the data in order to select the parameters c1 and c2 when removing noise.
I tested on Fisher's irises and it would be great if someone shared their classification quality and network configuration parameters.
Thanks