infryq

Yesterday I fixed the state dataset for J, met with her in the afternoon to struggle through trying to get MATLAB to read the files in the correct format for the neural network time series tool to recognize them without much luck, then hacked on the problem until dinner with partial success. We did the crossword after lunch, I got in some quality SG-1 time before my meeting with J, and after dinner helped P evaluate the somewhat backwards way GitHub Classroom has you use pull requests as a feedback mechanism. A pretty relaxing day.

Yesterday was user study day. I’m pretty sure F didn’t sleep; she and M were up working until 4, at which point the bitbucket repository had a hissy cow and switched into read-only mode. Some frantic emails to me and bitbucket support later, it was all resolved, but yeah no fun at all. I went in on the 10:15 to help set up. We went through our set of use cases and struck a balance between debugging low-hanging fruit and jettisoning cases that resisted fixing. Out of 80 use cases we wound up with 9 that at least one of us could get to prove. Late lunch (pasta from the CUC, with cheese! Did well), and we were working right up until participants arrived at 3. We ran one time through with very little training, and one time through with more instruction and some extra hints printed by the reasoning system, but the only person (other than us) able to get anything to prove was the one with prior contact both with prolog and with LIA. We will probably include the expert performance statistics in the paper along with a brief acknowledgement that the NLG for eliciting information doesn’t work for novices yet. I stuck around until 6:30 hacking on the paper with F, late snack, shuttle home. Was disinterested in dinner so just sat with P chatting while he prepped and ate. Third session of classes for him; he’s at a low spot but can tell it’s concave up. Crossword, then gave him a brief intro to unit testing in python, then sleep.