Is there any way to find a jamb in a neural network?

Question

Neural networks

I read and try to write a fairly simple perceptron... At first it worked, but then I changed the functions to sigmoids, changed the algorithm a little and the learning ability came to naught, you see, I screwed up somewhere. How can you find the reason for "not learning"?

Answer 1

slightly changed the algorithm

You must always change to a mathematical model in order to know what will happen. After - to build a network.
Now - also: Restore the mathematical model. Calibrate. Then - transfer the logic to the code.

Answer 2

Debug, look at the output of each function, understand what and how to happen and go in order.
The key to a successful programmer is to be able to debug.