This is great. Found myself on 5/30/22. Looking forward to catching up with photos of people I hiked with. Also love the pics from ‘79 - total time capsule! Short-shorts never went out of style in the hiking community.
I'm an AT hiker but I hate the short-shorts. I prefer the length to be just above the knee. If you've ever walked through a tall patch of poison ivy before you noticed what it was, you'll understand.
I’m in the shortest-shorts camp too. Less about the UL-value of the shorts, more about the mobility to lift my leg as high as I can (specifically to reduce friction going uphill). When you take literally millions of steps on a thru-hike, even the smallest things start to add up.
I’ll bite. I worked at a very large Fortune 500 company for a few years. Through some twist of fate, I ended up on an overfunded ML organization that was under no pressure to produce anything of business value. The executive in charge of this operation let this fly for about a year before they realized they weren’t getting anything valuable out of it. They put pressure on the VP of the org to start producing business value or else.
Fast forward another month or so and there are six hastily thrown together initiatives in which our organization is contracted to other organizations to “help them out”. The team dynamics were atrocious - our team was met with skepticism as to why we were helping out (was the other team not doing their job well enough?), and people on our team were often clueless on the business subject matter that we were supposed to be consulting on.
I was assigned to one of these initiatives, and upon digging into the business problem, our team realized that the feature importance was completely contained within a single categorical feature in the model. All other features in the model were comparable to uniformly randomly generated features. In the selfish interest of being able to claim that they “solved their problem with ML”, it was clear that either the model developer or the team at large had obscured this fact.
In an effort to save face, the VP kept us on the project, despite the poor relations and lack of useful features to improve the model. We didn’t improve the model and eventually, the VP ended up being “pushed out”. Last I checked, this VP is currently the CEO of a startup that recently raised a $100M seed round.
I imagine that’s just about the starkest within-US move you could make with regard to bike friendliness. I once made the mistake of visiting Houston without a rental car, and it was bleak. Welcome to NYC!
This is such a good idea, I wish I’d thought of it. I have trouble maintaining consistency in journaling, but this makes it a heck of a lot easier. I just signed up - good luck!
This is great. I remember finding another really good resource on the Bernoulli bandit that was interactive. Putting feelers out there to see if anyone knows what I’m talking about off the top of their heads.
I’m in a bit of a privileged position, but I took a fair bit of time off to pursue personal goals. I got extremely burnt out after 6 years of full-time work at big corp (simultaneously being in part-time grad school for the final 3 years).
I found it took me 3-4 months of absolutely nothing to not feel burnt out. It took much longer than that to get to a point where I was able to pick up my computer and have fun programming again. I’m now working on a webapp that I intend to turn into a business. I have been pouring myself into it in a way that I haven’t done since I was a new hire.
Hang in there. Maybe ask your boss if you can take an unpaid sabbatical. You will ultimately be much more productive if you get a break, and the time off will give you a chance to clarify to yourself what it is that you want.
> I found it took me 3-4 months of absolutely nothing to not feel burnt out.
That seems a universal experience. And to be more clear, absolutely nothing to think about, because even a sabbatical wouldn't save you if you think about post-sabbatical matters.
I took a year of non-employment in 2022 after 7 1/2 years of working in the same big corp. My experience shows that it takes me 1 to 2 years until I start to feel burnt out and need something new to work on, and fortunately the company was able to provide such refresher on time (not necessarily at my request) but wasn't able to do this time. So I quit, and next 3--4 months were absolutely non-productive time for me. (I have a ton of side projects at any moment and I realized my burn-out from not being able to work on any of them.) But that followed by a very productive period without any frustration [1], and I was eventually energized enough at the end of 2022.
Sadly not everyone can't afford this, I know, but if you can, consider doing so too. A year should be enough to clear your mind and start anew again, assuming you don't have any other significant problem affecting your life.
Entirely correct, it's really unfortunate that this advice is not actionable for everyone. I was able to pull it off partly because I had enough saving and I lived alone.
I’m unemployed in NYC right now - so, zero consistent income. I’m one of those people with the very worst marketplace health plans. Here’s what I get for $640/mo (with a $120 tax credit, oh boy!):
- Visits to my general practitioner (but not so fast! no diagnostic tests or ANYTHING else - those are 100% out of pocket until my $9,000 deductible is met)
- Pregnancy care (it’s currently a biological impossibility for me to give birth)
Everything else is entirely out of pocket until my $9,000 deductible is met. My coverage is basically a $6,000 guarantee that I won’t be completely bankrupted if I get run over by a garbage truck and live to tell the tale. Otherwise, it’s completely useless. Something’s gotta give.
This is what infuriates me the most about healthcare in the US. I give someone my card, they enter the numbers with no indication of whether or not everything will be covered by insurance (they assume you just know).
Then, a month later, I receive a bill indicating a routine procedure that I assumed out of ignorance was covered by insurance (since they took my card and entered the numbers without saying anything to the contrary) for $500 (or, god forbid, more).
If I knew a salad at a restaurant were $200, I probably wouldn’t order it. There is no basic transparency in medical billing, and that needs to change.
It’s actually even worse. If you (or the Dr’s) could reliably tell how it was even going to be coded (aka categorized/identified) in the system in advance, it would already be a huge step up.
Then you’d only have a handful of different prices you might have to pay.
Specifically for A/B or A/B/N testing, you can use a beta-bernoulli bandits, which give you confidence about which experience is best and will converge to an optimal experience faster than your standard hypothesis test. Challenges are that you have to frequently recompute which experience is best and thus, dynamically reallocate your traffic. They also only works on a single metric, so if your overall evaluation criterion isn’t just something like “clickthrough rate”, this type of testing becomes more difficult (if anyone else knows how multiple competing metrics are optimized with bandits, feel free to chime in).
Beta-Bernoulli multi-armed bandits (BB-MAB) are definitely a good way to get started on a Bayesian version of A/B testing where you've the additional benefit that your population is dynamically allocated to the most performant option (actually this dynamic allocation makes it similar to interim analysis [1] rather than vanilla A/B testing).
There are some caveats though - and I mention these from the experience of running such solutions on a large scale in production. First, BB-MAB can't adapt to context by design. They only look at click/no-click behavior across the population. So, if your population has two distinct segments - youth and elderly - who behave very differently wrt purchases, the BB-MAB won't pick a different winning advt. per group; its blind to these groups.
The solution is to use something like a contextual MAB - which assimilates user features (or whatever you might throw at it) into the MAB. There are simple ways to adapt simple MABs to the contextual setup [2] (in my experience, these can also be effective) but, of course, the literature in this area is wide and deep.
A second caveat is that if the ratio of the size of the pool of advts. to the number of impressions is high, the BB-MAB won't converge or converge to a good optima; the search space is simply too large relative to the data. In cases like this it becomes important to begin with the right Beta priors, instead of the standard recipe of starting with a Beta that looks like a uniform distribution.
Can you point to a book or other resources that support this claim? Interested in learning more. Always looking for another reason to hate cars as a New Yorker
I'm reading "The Geography of Nowhere" right now, it covers much of this in a reasonably entertaining fashion, although its not chocka-block with references.
It does make some predictions that haven't dated well though (like hitting peak oil in 2023)
