This is my entry for a Kaggle competition designed to detec pairs of duplicate questions asked on Quora. My goal was not to win the competition---or even to do particularly well---but just to learn some new things. I managed to install and use the XGBoost package for my classifier, and I learned a little bit about Word2Vec analysis. Most of the things I tried came from reading the Kaggle message boards about this competition.
I also learned that training a legit model for these types of competitions requries way more computing power than I have on hand. That's OK as long as I don't want to win. I can do baby versions of the real thing on my laptop, just as long as the dataset itself has a managable size. For this one, the training and testing data together totaled about 400MB, and that's more-or-less the upper limit of what I can handle locally. Processing this data, even with my simple model, took several hours of computation.