a gentle murbling

[kaberett]

Paper: currently at the point where every time I think I can write a sentence, I end up disappearing into the rabbit-hole of double-checking references and re-wrangling data for an hour or two. This is A Good Thing, in that it means that I've got to grips with the job in question sufficient to see the small manageable chunks, rather than burying my head in the sand about it. Of course I've just run into a bit of a dead end, at least until I can get my supervisors' input (probably in the new Gregorian year), but happily there are many other paths for me to branch off down.

Relatedly: PhD2048 is dangerous. I'd successfully avoided all previous iterations of the game, which unfortunately meant I wasn't innoculated when this nonsense started doing the rounds.

Foot: bruise came up briefly! Swelling was temporarily visible! Now at the point where I can walk on it around the house a little provided I'm careful and don't push it beyond, ooh, the bedroom-bathroom-kitchen-sofa route before I have a rest. In turn this means I'm wearing the boot a little less (it has so much velcro), which means I'm crossing my legs by default, which means I'm putting weight through my foot awkwardly, which... means I'm wearing the boot more again! But so it goes.

Thyroid: I'm feeling hypothyroid-ish worse again but my numbers are now pushing toward hyperthyroid! Still no autoimmune markers (and all my vitamins et cet are fine), not clear to me that the GP is actually aware that you plausibly want to end up "hyperthyroid" by-the-numbers when treating hypothyroidism, but regardless I'm intending to go back in the new year and ask for (i) free T3 levels as well as free T4 and TSH, (ii) trying adding in straight-up T3 in case there's a conversion problem ongoing, and (iii) a referral to an endocrinologist because I'm really bored of this.

Books: I kinda sorta ended up, on Wednesday, going into Foyles and then Fopp with awesome ex-housemate-C, who was briefly around; in consequence I acquired a present for Adam, an unambiguously-for-grown-ups book by Shaun Tan (I slipped and fell, it was by the checkouts, I flipped through briefly and saw the giant snails and succumbed), some Actual Pink Floyd Of My Very Own (for the listening to on our backs in the dark at 1am, sorry neighbours), and another couple of Carrie Fisher's memoirs (they were two for a fiver, It Had To Be Done). And, er, then going into Waterstones unsupervised, where books were Buy One Get One Half Price, which meant that I acquired a copy of Why I'm No Longer Talking To White People About Race (having had the library ebook automagically evaporate from my device when I was 2/3rds through, due to disorganisation) and Mortal Engines (because I loved the film and wanted to know more about some of the worldbuilding and characterisation that were sketched for the cinema but were clearly explored in a great deal more detail in the source text).

Comments

[sciatrix]

Oh my god, I had never seen that 2048 version. I'm both laughing and crying, I think.

[kaberett]

R I G H T

RELATIONSHIPS RUIN EVERYTHING

[sebastienne]

How dare u

A relationship turned my Viva into a PhD

Just don't think about the ethics of that k

[sciatrix]

aADKFHKL DYING

[sciatrix]

LOL my partner looked over my shoulder and went "Ah, yes..." very dryly when I explained the relationship thing. (I did get very into 2048 way back when it got big first, so they were initially confused as to why I was finding this variant hard.)

Apparently the solution is to freeze in learned helplessness the moment a relationship happens to you mid-PhD.

[kaberett]

LOL YEP.

I sooooooomtimes manage to cynically exploit them more effectively than that, but yeah.

[jedusor]

why have you done this to me. (5792 first try, which is a lot better than I remember doing with the original 2048)

[kaberett]

because I have finally got to the point where I am Resting On My Laurels and can mostly put it down rather than playing compulsively, so I need to ameliorate my angsty feelings about Wasted Time by spreading the pain <3

[chiasmata]

❤️

[kaberett]

<3

[rmc28]

Hurrah for good things.

[vass]

Relatedly: PhD2048 is dangerous. I'd successfully avoided all previous iterations of the game, which unfortunately meant I wasn't innoculated when this nonsense started doing the rounds.

Oh no. It's even worse than the original.

Ooh, I've read that Shaun Tan. It's a good one. And he's from NSW, but I've lived in parts of Victoria that look like that.

[vass]

Having just won the game, my findings are that the right relationship is if not necessary to acquiring a PhD, it is at least highly beneficial. In order to maximise the benefits from NRE and minimise the productivity risks of a breakup, one should meet one's new lover while literally in the middle of the viva. In fact, I'm afraid I may have just earned my PhD by fucking the examination panel.

[cesy]

That 2048 is very funny and also dangerous.

[fyreharper]

Boooooooks :D

[simont]

Oddly enough, I've been wasting time on 2048 too recently. I had had the idea that if you cut the grid size down to 3×3 then the game's state space becomes small enough to iterate over in only a few minutes, which means it's possible to (a) get a minimax analyser program to tell you what the chances are of a perfect player making a tile with a given value; (b) get the same program to write out a 10G data file from which it can read off the optimal move given a position; (c) hook that up to a thrown-together clone of the original game and watch it go; (d) boggle at how different its strategy is from any I've ever seen before; (e) puzzle for hours about what it could possibly have seen as a good reason to make that move (for about half the moves it makes); (f) try to relearn the game myself using its strategy rather than my previous one; (g) wonder why it's three weeks later and I seem to have the makings of a lengthy witter about game theory and no progress on any actually sensible project...

The 3×3 version also takes a lot less time than the full-size version to have a quick "just one game" of, which is not the benefit it sounds like, because now there are many more occasions when one is not put off starting a game by the knowledge that it would take too long!

[me_and]

That sounds fascinating! I would be very interested if you can characterise the strategy of the perfect moves in any particular way.

I remember back when the game first came around reading an essay from someone who'd written something statistics-based (IIRC) for the full game that wasn't perfect but was generally very good at the game, and it mostly played like the "good" human strategies I've seen advocated, albeit with a bit more of a tenancy to switch which corner it was working things into.

[simont]

OK, here's a brief summary of what I know about 3×3 2048.

My analyser reports that a perfect player can make 128 with 100% probability; 256 with probability 99.6%; 512 with probability 77.9%; 1024 with probability 1.4%; 2048 with probability zero. (Which makes "3×3 2048" a bit of a misnomer, but what can you do.)

The 100% for making 128 is equivalent to saying that even if the random tile drops at the start and after each move were instead chosen by a perfectly-playing opponent, it would still be possible to make 128 in spite of the opponent's best efforts. And the 0% for 2048 is equivalent to saying that even if you got to choose your own random drops to suit yourself perfectly, you still couldn't do it.

In my own far-from-perfect play, I find 256 to be a moderate challenge (I'd guess I hit it somewhere between half and two thirds of the time) and 512 to be something I've managed about twice ever. I have no hope at all of ever making a 1024 under my own steam.

Interestingly, those probabilities don't all seem to go together in the same strategy. If I tell the analyser to maximise the probability of making 1024, then it seems to shift into a high-risk high-reward mode in which it often gets quite close to 1024 (I've never actually seen a 1024 tile but I have seen e.g. nicely placed 512+256+128), but also it often falls on its face before even hitting the supposedly-easy 256. But if I tell it to maximise the probability of making 256 instead, then it gets 256 basically all the time (as it should) – but under that condition, it considers itself to have won as soon as it's done so, and doesn't have any idea what to do next! I have to suppose that if you tiered the strategies so that after reaching 256 it started asking the "make 512" file for its moves, then it would make 512 with rather less success, on the basis that it would reach 256 in a way that left the board less useful for followup play.

Watching the actual play: my main tactical observation that the auto-player is very reluctant to merge tiles. Once it has two tiles next to each other that it clearly should merge together, it will delay actually doing so for as long as possible, and instead move at right angles to set up the next few things, using the space taken up by the unmerged tiles to keep the grid nicely full so that only the row it actually wants to affect will move.

On the strategic level: the very early play (say, pre-128) looks utterly random, constantly giving a human viewer the feeling of "what on earth are you doing?". My theory is that that's just a side effect of it being inhumanly foresighted at micro-optimisation: I think the early play just isn't very critical, so making almost any old mess is just fine because the system is confident it can clean it up easily later on.

As larger numbers show up, the play starts to settle down enough that you can actually see some kind of overall strategic concept, which turns out to be essentially corner-based. My own strategy (in both 4×4 and 3×3), and that of everyone else I've discussed the game with, has always been to keep the high-value tiles in increasing order along the bottom row, and strive to (a) always keep the bottom row completely full so that you can move everything else left and right without breaking its ordering, and (b) never have to press Up, which makes a horrible mess. But the 3×3 computer player instead adopts a strategy in which its highest-value tile is in one corner, with the next two highest on the two adjacent edge cells, and building up the fourth highest tile in the centre. Once the centre matches one of the two edges, you can merge that edge out into the centre tile, and then merge that down into the other edge.

I can see the advantages of that system: in particular it lets you use all four move directions (instead of that "if you move up you've basically lost" business), which must surely give more flexibility. But I haven't had much luck teaching myself to play in that style yet!

[ps, kaberett, sorry to have spammed game theory all over your largely unrelated commment thread! I hope at least some of it was interesting :-) ]

[kaberett]

[I am always happy to host this kind of conversation in any of my spaces, it's not actually largely unrelated, and you made Adam very happy :-p -- but more later, when I'm not trying to rush Cambswards...]

[kaberett]

also: ooh, sounds like the script more or less agrees with my approach to playing :D

[me_and]

Interesting, thank you for writing all that up!

The maximising probability issue is interesting, and makes perfect sense. The 4x4 2048 bot I'd seen I think just worked on the basis of picking strategy to maximise score, but that obviously works a lot better when you're satisfied with "probably good" rather than "perfect".

[simont]

I picked my particular optimisation goal because my original aim was to decide what the reasonable winning criterion for a 3×3 board should be: I wanted to know what value of tile would be a not-hopelessly-unfair thing to demand of the player before printing "you win". That was really why I wrote the analyser in the first place, so I could ask it to find me a tile value which is within reach of a perfect player more or less all the time (and hence, if a mortal fails to get it, it's almost certainly by lack of skill rather than just bad random numbers).

If I'd set out in the first place to try to write an auto-player, then yes, I might well have considered some other more continuous criterion, such as maximising the expected score or perhaps the expected highest tile. But the auto-player was an afterthought, so it currently works on the value function I already had :-)

[untonuggan]

*waves* good luck with the thyroid. books are dangerous and wonderful.

[shreena]

I have hypothyroid and I am really surprised that the GP is handling it. Mine refused to and all if ny care is from the hospital endocrine clinic.

[highlyeccentric]

That 2048 is DANGEROUS.

[kaberett]

RIGHT.

[katzenfabrik]

Yay, books! Yay, being that far in the paper-writing process! That does definitely sound like a positive state of progress. Hope your supervisors have useful input when you get to ask them about it.

Re: thyroid, would you possibly have a link on hand about this: you plausibly want to end up "hyperthyroid" by-the-numbers when treating hypothyroidism? I am in the market for a new endocrinologist, once my terrible previous one has finally sent me his report on my last blood test, and it would be really helpful to have something to brandish at the new one when I find them. Only if it's easy, though! Otherwise I can google for it myself. Sorry you are still struggling with this. It sounds like you have a good plan, though. *internet hugs*

[kaberett]

I was thinking of

Levothyroxine (T4): The main treatment for hypothyroidism. Treatment is usually started at either 25mcg or 50mcg per day, depending on the severity of the condition. Testing is carried out at various intervals and dosages increased over the next few months until the test results show within normal range. According to the BMA's booklet, "Understanding Thyroid Disorders", many people do not feel well unless their levels are at the bottom of the TSH range or below and at the top of the FT4 range or a little above.

Dr Toft states in Pulse Magazine, "The appropriate dose of levothyroxine is that which restores euthyroidism and serum TSH to the lower part of the reference range - 0.2-0.5mU/l.

In this case, free thyroxine is likely to be in the upper part of its reference range or even slightly elevated – 18-22pmol/l. Most patients will feel well in that circumstance.

But some need a higher dose of levothyroxine to suppress serum TSH and then the serum-free T4 concentration will be elevated at around 24-28pmol/l.

This 'exogenous subclinical hyperthyroidism' is not dangerous as long as serum T3 is unequivocally normal – that is, serum total around T3 1.7nmol/l (reference range 1.0-2.2nmol/l)."

source

<3

[fyreharper]

Unrelated to post: ran across this on the book of faces and brave little singing robot <3

[kaberett]

YES ISN'T IT GOOD AND BRAVE <333