0

Based on below data I'm attempting to use gradeint descent to predict what tags will be associated with a new user. Note the numbers are for illustravite purposes only, in actuality these numbers correspond to words.

username tags  title    group

user1    1     Senior   group1 
user2    2     Senior   group2 
user3    3,4,5 Junior   group2
user4    2,8   Dev      group1

So if a new user is added : "user5" and I know the the title and group of this user can I use gradient descent in 'r' to predict what tags the user may require ?

blue-sky
  • 51,962
  • 152
  • 427
  • 752
  • 3
    Gradient descent is a numerical optimization method. Unless you specify a statistical model and a loss function that you want to optimize, it is not meaningful to ask for "predictions using gradient descent". Check the R package `glmnet` for example; I think it uses gradient descent to fit a regularized linear model. – sieste Apr 27 '13 at 11:48

1 Answers1

4

Gradient descent is a method of learning a regression model (for example), not a way of making predictions. That is, given a set of training data, you use gradient descent together with a loss function to determine the coefficients of your regression model.

In your particular case, it doesn't seem to me that a regression model would be the most appropriate one though. I assume that the numbers for the tags are IDs and have no actual meaning as numbers. That is, the fact that tags 1 and 2 and next to each other and 1 and 3 are one apart doesn't mean anything.

Instead, I would look at a classification model such as a decision tree. If you're just getting started with machine learning, I recommend that you have a look at Weka, which has many different machine learning algorithms and a relatively easy to use user interface. You can use it to quickly explore how different algorithms and models perform on your data.

Lars Kotthoff
  • 107,425
  • 16
  • 204
  • 204