The Catch-Up.

It’s been quite a while since I last blogged! So many good and bad things have happened.

The good: over the past few weeks, I’ve been experimenting with my RNN. Tuning hyperparameters, using different sources of data , and overall trying to get my model to do really well. AND I DID! 92% accuracy! Now I know others have gotten better accuracy but this was such an amazing feeling.

The bad: interwoven with the good has been days of frustration. Tensorflow is not well documented IMO. The amount of time I spend trying to find good docs, asking my mentor for help, and resorting to online tutorials has been astounding! Also, tensorflow and tensorflow-gpu – why??? THEN I ran into memory issues with AWS – had to increase volume size. Even after that – memory issues. I used chrome:tracing and Tensorflow timeline to try to figure out what was going on. My mentor and I were both confounded – a lot of operations going on. When I changed my dataset (25k examples) to a much larger 100K set – everything broke. I’ve spent most of my time trying to fix these issues + chat with AWS.

But it’s all been incredibly rewarding. This week, I’m looking to fix any issues left and THEN! swap my RNN for a Google Transformer! I’m really excited to use this architecture and overall pleased with how far I’ve come! I actually feel like I know what I’m doing 🙂