7.2 Meeting Videos

7.2.1 Cohort 1

Meeting chat log

00:53:20    Justin Dollman: just fyi, it’s grid_latin_hypercube() that will use a latin hypercube
00:53:30    Justin Dollman: grid_regular does evenly spaced stuff
00:53:49    Jiwan Heo:  doesn't grid_regular call hypercube somewhere?
00:54:14    Justin Dollman: i don’t think so…
00:54:42    Justin Dollman: but if i’m wrong i’d like to know 😅
00:55:03    Jiwan Heo:  yea I remember incorrectly answering a slack question, and Max corrected me one time lol

Meeting chat log

00:16:18    Bob Dylan:  The smaller the penality, the less regularization. The drop-off at the larger penalties means the model becomes too inflexible.
00:17:09    Bob Dylan:  This is Bob Dylan in the chat.
00:17:16    Jiwan Heo:  lol
00:22:28    Bob Dylan:  It's not that the second one has a smaller standard error, it's that its AUC is statistically indistinguishable from the "best" one and its penalty is higher (i.e., simpler model)
00:32:02    Bob Dylan:  I *think* it's just changing the data type.
00:32:09    Bob Dylan:  It's not sparse rather than dense.
00:33:13    Bob Dylan:  It's efficient and sparse versus inefficient and sparse.
00:33:51    Bob Dylan:  It's what dgCMatrix is doing
00:43:16    Bob Dylan:  I see them!