In some cases, we need to select a percentage of the input for the output. For instance, in CNNs, if we want to use every single neuron, the complexity of the network increases, leading to an overfitting problem.
To address this, we may use dropout layer techniques to randomly delete some neurons.

Similarly, in other programs, we may need to choose some number of inputs. We can define an option, such as --rowperrandom=0.2, to determine the percentage of values that we want to keep or delete.
For example, if we set this option to 0.2, it will keep 80% of the original values and delete the remaining 20%.