A3Q1 Classification of new data set
Author Message
g3pariso Offline
Student


Reputation: 0
Post: #1
A3Q1 Classification of new data set
In the assignment Q1 it says:

You can evaluate a trained model by conditioning on the training set ratings for each user, and predict the ratings for each of that user's test items.

But with the new data posted yesterday, there is no longer any overlap between users in the training and test set. So to compute the estimated rating of a test user i on a particular movie j, given the training set, do we also condition on that test user's ratings for all movies other than j, or do we make decisions without taking into account any information about the test user i?
2015-03-24 12:42
Find all posts by this user Quote this message in a reply
g1kan Offline
Student


Reputation: 0
Post: #2
RE: A3Q1 Classification of new data set
I think you would condition the cluster k that i belongs to (Z_{i} = k) on all movies j, then predict the ratings that a user belonging to cluster k would give the test movies.

Or, probably more likely, condition the mixture model on all movies and users in training set, then predict what each user would rate each test movie based on the cluster k than user i belongs to.
2015-03-25 05:36
Find all posts by this user Quote this message in a reply
g3pariso Offline
Student


Reputation: 0
Post: #3
RE: A3Q1 Classification of new data set
The problem is we don't know a user i's cluster k beforehand so we need to integrate out the latent variable. If we aren't given any training ratings by that user, we can at best assume that the probability that user i belongs to a cluster k is the prior probability of a cluster. Conditioning on a user's training ratings will make the user belonging to some clusters more probable.

I am not sure conditioning on other user's ratings, after learning parameters, will affect the probability that a particular user i rates something. I think the data is independent given the cluster. When I work through the math of summing over all cluster assignments, the contribution from other users cancel out in the conditional.
2015-03-25 11:15
Find all posts by this user Quote this message in a reply
Post Reply 
 
Thread Rating:
  • 0 Votes - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5


Forum Jump:


User(s) browsing this thread: 1 Guest(s)