More details on this week's goal

See link

The objective is:

• to train the transductive learner on labeled and carefully chosen unlabeled vectors
• the hypothesis is that if you chose the vectors systematically, the results will be better

There are two approaches that I identified for systematically selecting the vectors are - Option 1 , Option 2. I will contrast them with each other and with a non-systematic random approach.
An additional parameter in this exercise is the number of unlabeled vectors to be chosen. I've identified two options again - for starters, I'll keep it simple and make it (60 * number of labeled vectors of that category).

I'm basically playing around with the technique and not systematically exploring it. As a result, my results might be misleading. A systematic approach would:

1. Vary size of the dataset
   
2. Vary the categories (I'm going to perform this only for one of the 67 categories)