Message Boards Message Boards

0 Replies
1 Total Likes
View groups...
Share this post:

Classify[]'s extracted features

Posted 10 years ago

Hi, When using Classify with NeuralNetwork option, Mathematica automatically determines the number of extracted features. This number seems to be correlated with the number of examples provided. Can we actually see what makes up the input layer?

For example, in Classify's documentation, there's an example where we classify works by William Shakespeare, Oscar Wilde and Victor Hugo. The inputs are complete works: Othello, Hamlet... and each has ~30k words. Running

Options[Classify[<|"William Shakespeare" -> {Othello, Hamlet}, 
  "Oscar Wilde" -> {TheImportanceOfBeingEarnest,  ThePictureofDorianGray}, 
 "Victor Hugo" -> {LesMiserables, NotreDamedeParis}|>,  Method -> "NeuralNetwork"]]

reveals that Mathematica actually tokenizes and constructs a TF-IDF vector out of that large body of text! But the number of nodes is {6, 5, 3} which means there are 6 input nodes, 5 hidden nodes and 3 output nodes. So what actually makes up/goes into the 6 input nodes?

POSTED BY: duy nguyen
Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
or Discard

Group Abstract Group Abstract