Day 7: started pretty much at the last possible time, and yeah it wasn’t nonstop work, but ig it was like a bunch of hours, maybe 3h15 and there’s 1h30 left to the day, so shucks.

Day 6: worked good for like 3 hours, then procrastinated for 5, then slept til 11:45 pm. Statistics is just unbearable. He pushed the deadline to tomorrow so one more day of suffering

day 5: did like 6 hours so far and it's 3:30 pm 😎

logistic regression, multiclass logistic regression, and their gradient descent derivations and probabilistic interpretation (maximizing likelihood wrt w <=> min cross entropy for logistic, maximizing log likelihood wrt w <=> min cross entropy for multiclass regression, I think?), one-hot encoding, softmax, logistic function as special case of softmax for binary classification, u need only to fit C-1 coeffs for multiclass with C : #classes, Perceptron, multilayer perceptron, regularization (L2 is Ridge, L1 is Lasso), LASSO with higher reg (lambda > 0) leads to sparse weights bc a bunch of em go to 0, whereas for Ridge the L2 norm of the weights goes down so larger weights goes down but not to 0 and more regularization fights overfitting and more nonlinear bases increases overfitting (that's model complexity). LASSO represents Laplace prior, whereas Ridge represents Gaussian prior i think? And then there's regularization that does both Lasso and Ridge to strike a balance? LASSO uses soft or hard thresholding, Ridge has a closed form solution, I think LASSO doesn't, and isn't differentiable bc of the sharp corners of the L1 norm? then there's bias-variance with noise, stuff, that i gotta review. got 10/10 on the quiz woo

day 4: nope

@Bayesian I strongly reccomend a task list/checklist if.you don't use one currently.

@RobertCousineau I technically have todoist but i dont use it, good idea hmm

@Bayesian I've found that a small (fits in my jacket pocket), real paper notebook is by far the most likely for me to use. Apps are annoying. Paper checklists just do something for me.

@RobertCousineau omg I like the vibe, I got a small notebook, I will try this.

day 3, stayed awake in a haze til noon, then woke up at 10pm, so got no work done, rip

day 2 done, got an exam at 8:30am and ive been going through it a bunch. minimax algorithm and coarse correlated equilibria and correlated equilibria and potential functions and top trading cycle algorithm and price of stability and price of anarchy and condorcet winner and best response dynamics and multi item vcg mechanism and pricing for second price auction and lemke howson algorithm and all that, last of which i still dont understand at all

Work like 'highly concentrated work and actually get stuff done' or more like 'being somewhere or attending a meeting'?

@Lion it's, I put on a timer for an hour and won't do other activities during that time other than say solving problems for an assignment or going through a textbook's chapter. so i might space out for 30 seconds and it still counts as work, but if i go read a book or take a break it doesn't count. Last week I also counted watching a recorded lecture but that's lower effort and maybe i shouldn't count it, hmmm

What are the circumstances?

@RanaG wdym?

@Bayesian 4 hours seems like a weirdly low number? Is your time taken to something else?

@Bayesian I guess my question is what are the constraints in the way towards 5

@RanaG 4 hours is higher than i've successfully done in the last months, I have a hard time getting myself to work on school at all. addiction / adhd / lack of motivation / anxiety / etc. I increased from 4 to 5 because uh steps up the ladder to self-actualization..