Trial and Error

This past week I did a lot of writing for ISM. Not only did I draft a final product proposal, but I also wrote a research assessment over extended-connectivity fingerprints (ECFPs), slides for my original work presentation, and more code for my LD50 prediction model. However, it was a very short week for me, which was nice. Writing a proposal for my final product helped me to get an idea for how I would complete it, and the calendar provided a timeframe. However, I found that it was difficult to ascertain how long each task would take, since each step required successful completion of the previous one. I’m glad I took the time to research ECFPs because I’ve been attempting to use them for featurization in my model. I had seen some promising results on the dataset, but the benchmark results were not so good. I’m not sure if it was because I used too few trials or the size was too small, but I plan on pursuing that more. I met with my mentor and discussed my progress, and he suggested simply trying new approaches while also trying to be logical. This helped me understand what I’d have to do when writing a poster or paper after finishing the code: explain why I chose a certain approach and why it worked. I’m tempted to buy a GPU even though they’re expensive because Google Colab is slow, and if I use my laptop for benchmarking, then I can’t use it for other things. Nevertheless, I’m looking forward to being finished with my original work speech and trying more models and featurizers.