Can you approve is is correct? I was sure that colums should be feature in most part of cases and rows — numbers of examples. Or is it out-of-cases examples maybe to multiply with W matrix wich rows are weights of the example? Can you explain more clearly this topic?

The convention for neural networks in the Deep Learning Specialization is that,

In the X matrix,

- a column represents a sample;
- a row represents a feature.

In the W matrix of the l-th layer:

- a column represents the number of features from (or neurons in) the previous, l-1-th, layer,
- a row represents a neuron of the l-th layer itself

This is the convention for neural networks in the Deep Learning Specialization. You can see a different convention in other places.

Raymond