Training a Neural Network

Is it possible to use string data to train a NN? I am trying to use 'labels' obtained from a previous network that labels an image. I want to use this data to predict the image. When I am using th is data it is stating it has to be a numerical/logical value.

[Olof Engström]

Neural networks can only take numbers as input, so you have to convert your text features to numerical features. If the 'labels' are categorical you could for example use one-hot encoding. There are also more advanced text embeddings where semantically similar texts are encoded into similar (numerical) vector representations (for example word2vec).

Answer 1

[Seth Furman]

To add to Olof's answer, the categorical type allows you to efficiently represent a finite set of discrete categories that you might otherwise represent with strings.

https://www.mathworks.com/help/matlab/ref/categorical.html

Take a look at the following example from the documentation demonstrating how to make predictions using categorical features.

https://www.mathworks.com/help/deeplearning/ref/seriesnetwork.classify.html#mw_3a7e0bb6-3f26-4000-af5e-279fb637f6eb