I need to build a neural network accepting data from a particular .csv file where most columns’ type is object
, i.e. they are neither numerical nor string. My question is: can a neural network accept such an input? If not, how should I overcome such adversity?, and, if so, where could I learn more about it? I am working with Python.
Thanks in advance.
Advertisement
Answer
Simple answer is no.
Two key components in neural networks are matrix multiplications and activation functions, both are operate for numbers, integer (rarely) or floating point. In case of strings, you need to encode them into vectors of numbers first before you can use them in neural networks. Same for objects.
Take pictures as example, the convolutional neural network will treat pictures as pixels in 2D and then present them as matrix of numbers. So same in your case, you need to present your “object” (for whatever it is) as tensors (i.e. n-dimensional arrays of floats) first.