How to do K-fold for active learning and 3B?

[dhaaraa.darji]

HW3 update for k-fold is for active learning too, right?

For active learning we are anyway picking less training say 

5% and then adding 5% till it reaches to 50%.

I don't understand doing k-fold for active learning Question.

Same for 3B, randomly picking c% for training.

[cs15...@gmail.com]

You are asked to do "both":

- cross validation split into K=10 folds.

- for each fold, train separately on randomly selected c% (c  = 5,10,15...) out of the K-1 training folds and test over the whole designated test fold.  Then average per c values across folds.

- for active learning, also follow the folds: for each testing fold, train over actively-selected c% (c  = 5,10,15...) out of the K-1 folds, test over the testing fold. Average per c value, etc. Then compare with randomly-c% performance from above

--virgil

On Mar 15, 2013, at 12:51 PM:

I'm a bit confused about the update you made to homework 3. First of all question 3 now has two different sets of instructions on what to do. First the newer instructions say to do k fold validation where for each run you use k-1 of the folds as the training set, yet a little farther down it still has the old instructions of using c% of the datapoints as training. It does not seem possible to do both of those at once as some of the c's requested like 30% cannot be gotten by using k-1 folds as that would require the folds to not be the same size (1 fold has to be 70% and the other k-1 folds have to add up to only 30%). So first of all what exactly am I supposed to be doing here?

Normally I would just assume that I was supposed to follow the newer set of instructions but using them leads to problems with what problem 4 is asking. Problem 4 says to compare the results of active learning with "the performance of the Adaboost algorithm on the c% randomly" and as I already set the first set of instructions doesn't mesh with certain values of c. So if I'm supposed to be using the newer set of instructions and making my training sets out of k-1 folds what am I supposed to be doing here as I won't be able to compute the numbers I'm supposed to be comparing the results to.