Research Journal: Oct 23 - Oct 29

The longest I've ever had to wait for a computer

October 23

I managed to fix the code and fit a full Mixed-Effect Bayesian Network. It ended up just being some discrepancies between my own tidyverse-centered code and their raw R code. After checking the predictive performance of the MEBN against traditional methods, I found that it does perform *a tiny bit* better, but not too significantly. I should have expected this, given that the learned structure doesn’t link anything new to the treatment-outcome relationship.

This project is now stuck in a peculiar place. I think that the methodology will still work, but I need a proper real-world dataset to demonstrate this. I can write up everything else, but it’s up to me to look for a candidate that will work. That’s just the reality of these statistical methodology papers, there needs to be a demonstrable real-world benefit first.

October 24

There’s only about 6 simulations left to do, so I can finally start thinking about the different visualizations I need. There’s at least one table and one figure that I can think of, but I feel like I can extract more value from them. I think I need to demonstrate what happens to the individuals as different thresholds for the discontinuation/futility/graduation probabilities. This will be my most immediate task for this paper.

I think that in preparation for the conference, I may need to start thinking of data to simulate if I am not able to find an appropriate dataset before then. Not an ideal scenario, but I need to have something for the presentation.

October 25

Today I imagined some figures to create for the Platform paper. I thought about some of the operating characteristics that would be useful to demonstrate in the paper based on my first set of simulations. I’ll need to compile this data together to craft the visualizations first, and that’s a lot of data to go through.

October 26

Today I tried to get all of the data loaded in the 400 GB worth of simulations into my computer, but it failed all day. Thankfully some memory management functions in R help out with this, but that was certainly frustrating to lose a day.

October 27

Today I finally ironed out the issues with the memory management for loading in that massive amount of Stan models into my computer. This will take a day or two before I can start actually loading in data from these models. I started to create the graphs for the allocation probability and the actual treatment allocation to examine my design’s behavior. It seemed promising, but there’s definitely weaknesses that should be addressed in the manuscript.

October 28

Today I found out the Google internship opened up! I spent a lot of the day thinking about the cover letter and making sure my application was in the best shape I could get it to. I organized my research notes a bit more thanks to Tiago Forte’s Build a Second Brain, so that I’m not losing my notes so much.

October 29

Today was a much needed break day from writing. We went to Irvine to get some good food.