Two for one special

Week before last was a very disappointing week – hence no blog. I. Just. Could. Not. Make. It. Work. Research is hard and requires a lot of discipline; we all know this. But it also requires an adjustment to being wrong, being disappointed.

This week was really exciting though. I finally moved beyond Tensorflow boiler plate data massaging, beyond bag of word models, and onto RNNs! I went from having three windows, 15+ tabs each, to 1 window (and less anxiety).

SUCCESS: I trained my RNN on movie review data. I found an AMAZING Medium tutorial online, hacked my way through it, and somehow made my own thing (and used an LSTM to be extra fancy). It’s amazing the confidence boost that comes with getting something to work. My big question, going into next week, is: how do you know which architecture is right? It all seems to be a little of prior knowledge mixed with guesswork.

I’m going into this coming week feeling refreshed. I’m reading up on RNNs, LSTMs, and I’m going to add attention to my model! My goal is to FINALLY transition onto the Transformer framework.