>>42386
>Public English education anywhere is bad in general with a few exceptions
India seems to do decently well at English education, as do most parts of the EU? Although that might just be because they're more connected with the anglosphere so they end up learning by osmosis or out of necessity.

>underestimate how powerful MTL were before
That's not the case: before deep-learning, most MTL was pretty poor in general, especially for languages whose grammar diverged from English even slightly. It's only around the 2010s that architectures such as LSTMs and RNNs showed promise in terms of language modeling, surpassing the SOTA of more bespoke statistical NLP systems at that time. Seq2Seq was one of the most prominent examples here.

These RNN/LSTM models were decent-ish for "English-like" languages yet they still lacked the semantic modeling capabilities that are needed for implicit-context-heavy languages like Japanese. It's only with the advent of transformer based models (e.g. GPT) that this became a "solved" problem (well almost, now the issue is getting the inference cost down.)

>If there was going to be a big change it would have happened by now.
Not quite, widespread access to LLMs is bottlenecked by compute cost at the moment. But you can already see that e.g. DeepL is order of magnitudes better than google translate at Japanese (I don't believe the precise architecture of DeepL was ever revealed, but given their timing it's probably some Attention-based thing. It also helps that being spun out of Linguee they have a quality dataset.)

>apps that can MTL what you type/tell them instantly already
See above; with the exception of DeepL (which itself is still probably more primitive "small language models" instead of the massive behemoth that is llama/gpt-4/etc.), LLMs have not yet been widely deployed for MTL, mainly due to cost.

>given how hard it would be even for a LLM to pick up constantly developing slang
That is part of the beauty with LLMs, they can be fine-tuned fairly easily. Coupled with plentiful access to data (which admittedly might be harder to come by now that everyone is starting to silo APIs), this is actually the "easy part".

> LLM to essentially do a slight better version of an official MTL still need to hire a quality checker with knowledge of the language
This is correct; however it is a matter of scale, it is far easier to QC an existing translation than to come up with your own, so more scripts can be translated with fewer human employees.

>>42388
>(well almost, now the issue is getting the inference cost down.)
Sounds like a job for SNNs
https://medium.com/@hiiask/spiking-neural-networks-in-neuromorphic-computing-fe671061470a
>SNNs are >100x more efficient than the best recurrent ANN and 1000x more efficient than LSTM.
On traditional hardware though, it'll be markedly slower.
https://news.ucsc.edu/2023/03/eshraghian-spikegpt.html

Post edited on 2nd Mar 2024, 2:19pm

>>42389
>Sounds like a job for SNNs
Not too sure of this. I actually have not heard of SNNs before this, but the model seems actually fairly similar to standard topologies, only difference is that you have an accumulator added before the activation function.

Any claims of efficiency needed to be verified rigorously with an apples to apples comparison, if you're comparing dedicated hardware against a CPU implementation then of course the dedicated hardware will win out. I'm not convinced this "neuromorphic" hardware is any better than just the standard systolic-array architecture for matmul accelerator.

>>42390
>the model seems actually fairly similar to standard topologies, only difference is that you have an accumulator added before the activation function.
>Any claims of efficiency needed to be verified rigorously with an apples to apples comparison, if you're comparing dedicated hardware against a CPU implementation then of course the dedicated hardware will win out
I'm not familiar enough with neural networks to make those comparisons. I am taking an undergraduate class about neuromorphic computing though, and to paraphrase, my professor has said that in every task that's been tested, neuromorphic algorithms have been more energy efficient while being at least as accurate. I asked him if that was dependent on what hardware is used; he answered that while specialized hardware is faster, neuromorphic algorithms are more energy efficient on any hardware.

These are two of the papers him and a graduate student have written
https://ieeexplore.ieee.org/document/8967864
https://ieeexplore.ieee.org/document/9340948

Post edited on 2nd Mar 2024, 3:14pm

>>42391
> I am taking an undergraduate class about neuromorphic computing
Something about teaching a highly specialized non-mainstream field like neuromorphic computing without introduction to more standard traditional architectures is setting off alarm bells. If you haven't learned traditional architectures you have no point of reference. (And unlike say economics where one could actually start off with something like Austrian economics as an introduction given that it's not like "mainstream" schools have had any better luck, traditional networks have by objective metrics succeeded in the real-world, whereas spiking neural networks have been nothing more than a curiosity in academia).

>in every task that's been tested, neuromorphic algorithms have been more energy efficient while being at least as accurate
Again sets off alarm bells. If this is the case _regardless of hardware_ then people would obviously have deployed them in the real world by now. Either one of those does not hold, or it only works for "toy models" and doesn't scale to real-world problems. There is not much standard literature on SNNs so I am not qualified to say either way, but it just doesn't pass the sniff test. Machine Learning is not one of those academia-gated groups: assuming you aren't compute bottlenecked, if there is a good idea then it will be borne out by real-world trials (see diffusion models and LLaMa, where a lot many improvements came from independent researchers).

I found [1] from Geoff Hinton who is definitely qualified to have an opinion(tm) on this field, he seems to say the hardware might have potential but we don't yet have the algorithms to train a SNN for problems that we care about. Which to me roughly makes sense, you can make some handwavy argument about non-linearity making it hard to derive gradients and thus meaning you can't do backpropagation.

[1] https://www.youtube.com/watch?v=2EDP4v-9TUA&t=2809

>>42392
I took an intro to AI class, but that was about stuff like A*; the most advanced topic covered was perceptrons. There's a deep learning and a machine learning class too, but I wasn't able take those for scheduling reasons. None of these 4 classes are a prerequisite for another. *shrug*

I find your skepticism overly pessimistic though. It sounds like you think there's chicanery going on. There will be more progress and I don't see why SNNs can't be applied to machine translation.

edit: Backpropagation is also possible
https://arxiv.org/abs/2009.08378

Post edited on 2nd Mar 2024, 5:11pm

>>42388
>they're more connected with the anglosphere so they end up learning by osmosis or out of necessity.
I can agree with this correction. Citizens in the EU are often raised bilingual to be able to communicate with close neighboring countries. Compare that to Latam where the majority of countries speak Spanish and people can have a well-paid job or political position without basic knowledge of English.
Free basic English classes in culturally and geographically isolated countries outside of the anglosphere are often seen as a mandatory subject in schools due to English being considered the global language rather than a strict necessity (I'm including Japan in this). It isn't taken as seriously as it should be. For reference, dedicated online English speaking Japanese learning communities are aware of the issues with the way Japanese is commonly taught but that's a general popular language pedagogy problem, in my opinion.

>That's not the case: before deep-learning...
Well, I'm sorry. I made that post under the dumb assumption that GoogleTL still uses SMT and DeepL being SMT (I have it bookmarked but never go beyond the /translate page so I never noticed). I have to agree because I did notice online translators slowly becoming better over the 2010s growing up and honestly, I'm not knowledgeable about technology as you can tell, I had to google some abbreviations to fully understand that post. I'm only skeptical of them being able to collect data to pick up slang unless they can get it from SNSs, but we'll see.

Post edited on 3rd Mar 2024, 4:24am

>>42394
>being able to collect data to pick up slang unless they can get it from SNSs
Also even if that slang never appears in the training set, a lot of it can be inferred from context. Again one of the strengths of LLMs is that they can handle "out of distribution" tokens with grace, in a manner similar to a human would where context is used to derive the most likely "meaning" for that token.

I was lurking an old thread about imageboard addiction and one anon pointed out something I hadn't considered about their appeal among the repetitive posts. Besides being able to express yourself anonymously and a low commitment to conversation, it works as a way to create and be creative with the words and images you use, ¨The thrill begins when you start typing and ends when you click SUBMIT¨(or reply). When I first tried to limit my time on them years ago, switching to lurking worked. I didn't have to check to see if I got any replies and eventually I did something else. I'm taking this to heart because I've been procastinating and there are other creative projects I want to do in my free time.

>>42370
>>42371
#18 is the anti-neet law. I don't think it's a good idea to take that book seriously unless you are forced to deal with certain people or for self reflection.

Imageboards made me realize that hierarchies are not required for culture and norms to exist, and it made me wonder if it was possible for an imageboard without a distinct culture to exist. I was also considering how giving an individual an identity leads to distinction, which leads to classification and then the eventual formation of a hierarchy. However, I am unsure if imageboards are truly free of hierarchy. There are "mods" and "admins." Then, there are also tripfags. Another instance of this, is where those do not follow the board culture or unspoken rules are deemed as outsiders. But, contrary to this, the outsider may become the "insider" by simply lurking, getting to understand the culture, and refraining from using "outsider" language, which is something that can be changed. A lot of hierarchies in real life are often formed based on traits of individuals that are pre-determined by uncontrollable life circumstances or genetics. For instance, your wealth, your looks, and your race. There is also the thing in real life where an individual's status in a group is determined by how long they have been in the group. As a result of it, their thoughts and opinions will have greater weight than those who have a lower status. But, on imageboards this does not occur because everyone is anonymous, excluding moderators, admins, and places where anon is not forced. You never know truly who you are talking to. You may be arguing with one anon in a thread and then agreeing with him in another. My thoughts is that all imageboards have an individual "common writing style" where users unconsciouly adhere to a writing style that all of the other anons on the board use. I mean anonymity can be broken by having an overly unique posting style. Again, though, this can be remedied by simply lurking and absorbing the culture. But, to go back to my other train of thought - the idea of an imageboard without a unique culture - without gate-keeping, without having unspoken rules, without having norms, will this draw in the wrong sort of individuals - not the types of people you see on imageboards but those you might see on more common social media websites -- the types of individuals who are predisposed to form cliques. Is the differentiation of the "outsider" and the "insider" inevitable in imageboards?

>>42408
>an imageboard without a distinct culture to exist
An imageboard without a culture would just be entirely full of shitposting, since there would be no real "rules" to enforce I think. You would need some standard to hold the posters to otherwise you will have people kicked out of everywhere else, not to mention that people would run bots and spam and other such things, which would make the site ungovernable. I suppose that would be an imageboard without any real culture, unless you consider a complete lack of standards and chaos a 'culture', but it would be near unusable.

>Imageboards made me realize that hierarchies are not required for culture and norms to exist
I agree with the points you have raised, although I wouldn't really put tripfags in the hiearchy as they should be anons anytime they aren't using their trip for the designated purpose, but I do think the amount of time someone has spent on an imageboard will always be a contributing factor to the "hiearchy" as someone who has been around for a while, but again it isn't really much of a factor as you said it would be impossible to know just how long someone has been there, and going around bragging about being an oldfag who has been around 20 years is seen as uncouth in most places. Very much unlike the real world, where just on virtue of being someone for a while gives you authority over others, although this is dependant on the situation, more often than not there will be some authority to go with seniority.

>Is the differentiation of the "outsider" and the "insider" inevitable in imageboards
Yes, I think it is an inevitability, but on the other hand it's far easier to converse on an imageboard than anywhere else. In real life or on site such as twitter or reddit, it would be quite difficult to have any sort of meaningful discussion with someone, as sites like those are by their nature against the spirit of a completely unbound conversation, as you have to either have a brand new account or allow someone to view all your past discussions, tainting the conversation. You could identify someone based on posting style or identify someone based on the content of their posts, but usually it's an anonymous discussion unless you are intentionally bringing attention to yourself. I don't think it's a bad thing to have an outsider and insider differentiation, but in real life, as you said, it's very much based on factors far outside your control, and in many cases people will decide they dislike your or not based purely on appearance, discounting all else. This is true whether it's in a professional or academix context, or even when trying to make friends.

I also think that in general ford drivers have begun to tend towards violent outbursts, although I'm not sure whether this has been a recent development or not, I have noticed more "normal people" becoming increasingly aggravated and quick tempered. I think it's a natural result or how hectic the world has become, as most people can't keep up so they either turn to drugs or they are eternally on edge and ready to snap, taking their anger out even when there was no need to do so. I had a recent interaction with someone, and what had been a simple, and what I had thought innocent, question resulted in them becoming aggravated with me over a misunderstanding on their part, with them swearing at me in anger and accusing me of attempting to trick them. This is someone I've had some interaction with before, although I was not remotely close to being their friend, and I suppose it just starkly reminded me how quick people are to snap when you are perceived as an outsider to them, and they seem to see little aversion for nasty behaviour towards others, even if such behaviour is unwarranted. Not to mention how many people act when behind the wheel of a car.

>>42409
Also, I was thinking a little more, and I might retract what I said about moderators and admins as part of my example. I've noticed that in imageboards specifically that admins, mods, and tripfags, not in the mode you mention, are particularly scorned. (Maybe "scorned" is not the word - but I definitely see major pushback between imageboard users and any authority; it may also be due to contrarianism.) As you said, it is uncouth to brag about being an oldfag, and I am almost thinking that this applies to those who try to call some sort of attention to their status, at all. I never really see the ass-kissing of mods/admins on imageboards that you see on forums and Reddit. There just seems to be something inherent in the design and innate culture of imageboards which prevents a solid hierarchy from forming. Although a "hierarchy" may exist, it is minimal, and it has little effect on imageboards.
>although I wouldn't really put tripfags in the hiearchy as they should be anons anytime they aren't using their trip for the designated purpose
I think that's a valid use for a trip. When I was referring to tripfags, I was referring more to the attentionwhore types who seem to want to have an "identity" on imageboards. But for instances where a trip may be needed, I wouldn't put them in the hierarchy, either.
>view all your past discussions, tainting the conversation.
Ah, yes, that is what I dislike precisely about non-anonymous forms of communication online. Having a post history allows users to infer more details about them than they directly share in a conversation. These details might be about aspects of themselves which they can not control, and some of the real-life dynamics start to leak into the internet. Not just a seniority bias seeps in, but many other pre-conceived biases others may have about you seep in, as well. This might be a controversial take, but I've noticed that the media tends to overwhelmingly paint imageboards as "misogynistic" and "racist." However, I was thinking that imageboards would almost be better for marginalized groups because everyone is able to be on an equal footing, provided you do not share details about yourself which would cause others to judge you based on pre-conceived notions of your characteristics, whether that be your looks, your wealth, or some other thing.
>I also think that in general ford drivers have begun to tend towards violent outbursts, although I'm not sure whether this has been a recent development or not, I have noticed more "normal people" becoming increasingly aggravated and quick tempered. I think it's a natural result or how hectic the world has become, as most people can't keep up so they either turn to drugs or they are eternally on edge and ready to snap, taking their anger out even when there was no need to do so.
I agree with you. I have noticed this, too. I'm not sure, but I've noticed that the world and the internet in general have skewed towards this way in 2016, but it may have been earlier than that. I guess people have become extremely polarized on all sides. All opinions are the wrong opinion to have because they will inevitably upset someone. It does not matter how benign or innocent what you say is; it can be twisted in a million different ways. But, to add to the second part of what you said, I also think it has to do with societal standards becoming increasingly unrealistic while it becomes more and more difficult to meet those standards due to the current state of the world. However, I don't think it's just causing aggression and violence, but I think it's causing people to look for strange ways of deluding themselves in order to be able to function. Everyone seems really obsessed with self-pathologization. If they look at themselves and see a characteristic that is "odd," then they immediately jump to suggesting that they are an "autist" or a "schizo" - not just in a joking manner - but it genuinely seems like there are a lot of people who genuinely think these things. I feel like people use these diagnoses to ease societal pressures on themselves and to be able to tell themselves that it is okay that they are not living up to the unrealistic standards that are in place. However, I also think that this is due to a breakdown in people's ability to develop an identity as a result of being unable to find a community in real life because everyone is so isolated; both due to social media usage and because no one really goes outside anymore. If they do, then they remain shut down. Otherwise, what is considered "normal" has become so narrow and confined, and I think that puts quite a bit of stress on people. Then, there is also the problem that there are just simply so many people on the planet, and jobs are limited, which creates a sense of fierce competitiveness to survive. This is also a struggle for those with atypical life circumstances who manage to find themselves "behind." But, even those without those circumstances people experience intense pressure.
Then, as a final small comment on how people act in cars, I'm honestly scared at how non-chalantly people take driving to be. When I first started driving, I had pretty severe anxiety about it, and I struggled with those implications - that I was driving a thousand pound vehicle that could kill anyone with ease. Then, there was the thought that one small mistake could send my insurance prices through the roof. It took me way longer to learn to drive than average for that reason. I effectively had to learn to dissociate from my emotions to be able to drive. Yet, others never seem to really think of these things.

> but I definitely see major pushback between imageboard users and any authority
It depends on the board I think. On 4chan this is definitely the case since probably many see the mods as only contributing to the problem rather than solving it. On TC there isn't any such disparity (in fact they blend in so seamlessly that I genuinely don't know if there are any mods besides Tohno himself). On the infamous orange-bar site, the mods are viewed fairly positively.

>I'm honestly scared at how non-chalantly people take driving to be
I can't drive for the same reason.

>>42418
>the infamous orange-bar site
Which site?

>>42422
hacker news

>>42418
>the mods are viewed fairly positively
There's only one guy doing it these days too.

>>42426
I was about to ask about Scott (sctb), but it seems you're right, he left around 2020 and now it's just Dan. That's fairly impressive. The overall moderation system isn't perfect (there's clear political and social bias in terms of which articles get flagged and automatically taken down), but in terms of direct human moderation Dan always seemed fairly hands-off.

>>42427
I pretty much agree.

>Dan always seemed fairly hands-off.
Even though he's being paid to do this, I cannot help but mention my appreciation for his rather even-handed and cool approach to moderation. Not many in his position would be able to do the same, as seen on other sites.

Whenever music sounds interesting enough, people say it sounds like it comes from a video game. That's probably because video games are the only time they can remember hearing music with a strong melody and musical development.

I've noticed a pattern where women who represent themselves with a character that has pink hair and cat ears, is probably a cunt. Mikeneko and Nyanners come to mind. You think there's something to this? Maybe it's a form of compensation or concealment. Pink hair and cat ears are fine on their own. Together it's obviously excessive and saccharine, so only somebody with a twisted mind would go for that look.

>>42548
I think this is a bias where people assume that if someone presents themselves in a certain way they must be a nice person. It happens in real life too, like how others assume people into anime must be meek and nerdy and are surprised when they form their own groups in their communities like everyone else, then they complain about how nerds are surprisingly rude when it's their fault for judging people by their appearance and believing that stereotypes apply to everyone.
Or maybe my mind is twisted because I've always liked customizing my characters, profiles and avis on the internet in a way that can be described as excessive and saccharine. The internet is the only place I can do that, I simply like the way it looks.
Honestly, knowing myself, I can safely say people who do that tend to think in extremes but at least I'm self aware and try not to be a cunt. Assuming those women are influencers, they probably do it to keep up an extremely positive feminine image of themselves and gain followers.

>>42549
>bias where people assume that if someone presents themselves in a certain way they must be a nice person
I wonder if that's part of the reason why a lot of people on social media tend to have anime girls as profile pictures.

>>42550
I don't like misrepresenting myself, so I always use PFPs like this. It's Japan related, but also kind of weird and off-putting without being overtly risque.

Post edited on 18th Apr 2024, 9:44pm

>>42550
The ones that use cropped images from h-manga as their avatars are 9/10 times going to be the worst.

>>42548
>>42549
>>42550
i think you all are overanalyzing things on some level, i think most people just choose an image or a character they like without putting much thought into it. i've noticed i tend to usually set my pfp to girls, but it's usually because i either like the character or the design a lot (assuming i even put that much thought into it, there are a couple of times i've just chosen what the last saved image was).
in the case of nyanners, specifically, it's odd, though, since she seems to not even like things like that in the first place.
>>42552
but...some of my favorite characters are from ero-doujin...

>>42521
I love listening to 70's prog rock and electronic music and you can tell many video game composers took inspiration from them.
Like nobuo uematsu from triumvirat and ELP, or ZUN and ZUNTATA from yellow magic orchestra

>>42556
Not to mention the relatively many Japanese prog rock groups that are heavily influenced by ELP. There was even a "girl band" that played ELP-esque music.

>>42388
Here: https://arxiv.org/abs/2404.13813
It was common anecdotally that gpt-4 blew away gtranslate and even deepl for languages like Japanese, but here's some quantitative analysis of that.

I've never thought about it before, but French food is massively overrated. Before I unquestioningly accepted that it's "good", but in retrospect, I think it's an emperor has no clothes situation. Yes it was highly influential and innovative. Yes, they have great bakery. Yes it's better than other western European cuisine(not that that's saying much). None of this changes that's it's overrated.

What are the classic, famous dishes? With Italian, I can immediately come up with stuff that's memorable, distinct, and I love. Caprese salad, gnocchi, squid ink pasta, salami, pizza of course, burrata, bruschetta. What French food can I think of? Quiche. Onion soup. Croissant. Ratatouille... Brie? None of which I particularly like. Didn't love Cajun either when I tried it in New Orleans. Even Germans have a signature dish I really like in Bratwurst.

Visually it's boring. Coq au vin is just chicken and vegetables, with some brown sauce. Reminds me of the boring home cooking I grew up with. Then you look at oyakodon, and it POPS. It's so colorful and interesting. Look at char siu. Ramen. Sushi. I see East Asian food, and immediately want to try it based on how it looks alone.

Basically, French cuisine is either fussy and massively overpriced, or boring peasant food. American and British food sucks, so influencing those doesn't count for much. Maybe I'm just ignorant, but food usually isn't a hard sell for me even if I haven't tried it. If it's so amazing, why isn't that more obvious?

Post edited on 31st May 2024, 9:37pm

>>42637
I guess when I think of french food, the only thing that comes to my uncultured (pun intended) mind is bread and pastries.

Playing MMORPGs seems like one of the best predictors for someone ending up a miserable, completely unproductive fuckup. Not only is it a stereotype in fiction, I even known somebody like that. MMOs are the original "lifestyle" games, and there's no limit to the amount of time they can suck up. On top of that, their repetitive, grindy gameplay is a precursor to mobile games, but unlike those, you also have to be shut-in to be knee-deep in an MMO.

Post edited on 2nd Jun 2024, 3:40pm

>>42645
I had it hit me a few months back about how much of my time "gaming" has really been spent in a couple of games. 6500 hours in Runescape (osrs/rs3), ~2K hours in eu4, god knows how many hours in Minecraft.

I think growing up poor wired me to isolate and seek out cheap forms of escapism.

>>42646
The guy I know is into FF14. Told me he hasn't played any other FF games, and only started this one because his online friends did too. He's also never seen any anime, because he's "too lazy" to pirate them.

Post edited on 2nd Jun 2024, 6:26pm

>>42645
>>42646
It's funny, I always had one foot out of that pond. I felt like I wasted a lot of time, but also it didn't stop me from living life, or at least getting shit I wanted done, done. Before a particular MMO I spent most (or second most) of my time on shut down I checked the hours played. 1538. Over two months of play over the span of about seven or eight years. It really felt a lot more.

I think reading for recreation hurts me much in the same way any other form of escapism. It's in many ways better than playing games or watching anime though, because it still requires you to exert some effort, use imaginations, etc. I simply feel empty to a degree after a book I was interested in, even if I've been for 10+ hours in a row. I feel outright dumb after even 6 hours of anime/games. That's how I conclude books are healthier. Doesn't prevent them from hurting me though. I think recreation is harmful, actually. Maybe I'm doing it wrong.

This thread is 12 years old. Roughly 900 posts. 75 posts per year. 6.25 posts per month. 0.2 posts per day. Which is one post every five days.

>>42653
You could scrape the actual timestamps and then do some plots! I bet if you plot the delta between posts you would see clusters where people respond to each other in a flurry then things settle back down.

I don't know how to plot, but here's sorted data as zip archive attached to picrel. A few years ago I could do it myself but now I'm beyond mediocrity and can't even figure out how to calculate amount of posts per day.

>>42654
Here ya go.

import matplotlib.pyplot as plt import numpy as np from datetime import datetime from datetime import timedelta data = [] #np.empty(0, dtype="datetime64[ns]") dt_string = '%m/%d/%y(%a)%H:%M' file = open("data", 'r') Lines = file.readlines() for line in Lines: dp = datetime.strptime(line.strip(), dt_string) data.append(dp) data = np.array(data) delta = np.diff(data) / timedelta(days=1) plt.plot(delta) plt.xlabel('Index') plt.ylabel('Time Δ(days)') for i, time_stamp in enumerate(data[1:]): if delta[i] > 200: plt.annotate(time_stamp, (i,delta[i]), fontsize=6) plt.show()

>>42656
log-scale the y-axis to magnify the low-end?

>>42657
Also I think probably a different visualization might be cooler:
x-axis as time, y-axis as time since last post. I can't remember what it's called, time-to-event / survival plot?

This should give a better sense of relative time rather than using numerical post index for x-axis.

>>42656
what exactly does it plot?

>>42658
>time-to-event / survival plot?
I looked those up, and they seem more applicable to medical or mechanical data. So I used your other suggestions and made it a scatterplot.
>>42659
The point with 2023-11-21 for instance indicates that a post was made on that date, and it was around 500 days since the last one.

import matplotlib.pyplot as plt import matplotlib.dates as dts import numpy as np from datetime import datetime from datetime import timedelta data = [] #np.empty(0, dtype="datetime64[ns]") dt_string = '%m/%d/%y(%a)%H:%M' file = open("data", 'r') Lines = file.readlines() for line in Lines: dp = datetime.strptime(line.strip(), dt_string) data.append(dp) data = np.array(data) delta = np.diff(data) / timedelta(days=1) plt.scatter(data[1:], delta) plt.grid(True) plt.yscale('symlog') plt.xlabel('Time') plt.ylabel('Time Δ(days)') for i, time_stamp in enumerate(data[1:]): if delta[i] > 200: plt.annotate(time_stamp, (data[i+1],delta[i]),fontsize=6) plt.show()

Post edited on 4th Jun 2024, 9:11pm

>>42661
>time-to-event / survival plot?
>I looked those up, and they seem more applicable to medical or mechanical dat
Yeah TTE / survial is probably not the right term for the "time vs time since last event" graph I was thinking of and which you plotted.

I've seen such a plot used before for monitoring setups (e.g. time since last failure, or cumulative delay since last event); maybe it's called a cumulative time graph or something? The idea is that it's sort of like a 1D scatter / dot plot / strip plot where you directly get a sense of the density of posts in a time interval, but then you also take advantage of the y-axis to 1) avoid too many overlapping points if the delta intervals (y) are of a different order of magnitude than the monitoring period (x), and 2) draw more attention to the pacing of events (e.g. events that deviate from a periodic pacing would be graphed as fluctuations in a horizontal line, which is easier to make out visually than seeing fluctuations in density.


So your plot is really cool. You can immediately see on X axis the time periods when TC was active (seems like TC, or at least this thread, died a bit from 2016-2020 before being revived during covid and then again this year. And likewise that most activity occurs within 1-10 days of previous post as people reply to each other.

By giving the points an alpha value, I got this. Bit easier to see the density of clusters.

>python
seethes
See how autists do it.

$ cat <<EOF > script.awk
BEGIN {
i = 0
}
{
split ($0, a, /\/|\(|\)|:/)
str = sprintf ("20%d %d %d %d %d 0", a[3], a[1], a[2], a[5], a[6]);
ts[i] = mktime (str)
i++
}
END {
d = mktime ("2024 1 2 1 1 1") - mktime ("2024 1 1 1 1 1")
diff[1] = 0
asort (ts)
for (j = 2; j <= length (ts); j++) {
diff[j] = (ts[j] - ts[j - 1]) / d
}
for (j = 1; j <= length(ts); j++) {
print strftime("%F-%H:%M", ts[j]) "," sprintf("%.10f", diff[j])
}
}
EOF
$ cat <<EOF > data.gnuplot
set terminal svg font "Liberation Serif,18" size 9600,1080 \
background "white"
set output "graph.svg"
set datafile separator comma
set xdata time
set format x "%Y"
set xlabel "Year"
set mxtics 12
set ylabel "Days since last post"
set logscale y
set grid
set timefmt "%Y-%m-%d-%H:%M"
plot 'data' using 1:2 title "" with points pointtype 7 pointsize 1 \
linecolor rgb "#c8000000"
EOF
$ regex='[0-9]{2}/[0-9]{2}/[0-9]{2}\([a-zA-Z]{3}\)[0-9]{2}:[0-9]{2}'
$ curl -sf "http://tohno-chan.com/ot/res/16448.html" | grep -E "$regex" | awk -f script.awk - > data
$ gnuplot data.gnuplot
$ inkscape -d 150 -o graph.png graph.svg

Or even
set grid mxtics xtics mytics ytics

>>42664
Done some post formatting.

$ cat <<EOF > script.awk BEGIN { i = 0 } { split ($0, a, /\/|\(|\)|:/) str = sprintf ("20%d %d %d %d %d 0", a[3], a[1], a[2], a[5], a[6]); ts[i] = mktime (str) i++ } END { d = mktime ("2024 1 2 1 1 1") - mktime ("2024 1 1 1 1 1") diff[1] = 0 asort (ts) for (j = 2; j <= length (ts); j++) { diff[j] = (ts[j] - ts[j - 1]) / d } for (j = 1; j <= length(ts); j++) { print strftime("%F-%H:%M", ts[j]) "," sprintf("%.10f", diff[j]) } } EOF $ cat <<EOF > data.gnuplot set terminal svg font "Liberation Serif,18" size 9600,1080 \ background "white" set output "graph.svg" set datafile separator comma set xdata time set format x "%Y" set xlabel "Year" set mxtics 12 set ylabel "Days since last post" set logscale y set grid set grid mxtics xtics mytics ytics set timefmt "%Y-%m-%d-%H:%M" plot 'data' using 1:2 title "" with points pointtype 7 pointsize 1 \ linecolor rgb "#c8000000" EOF $ regex='[0-9]{2}/[0-9]{2}/[0-9]{2}\([a-zA-Z]{3}\)[0-9]{2}:[0-9]{2}' $ curl -sf "http://tohno-chan.com/ot/res/16448.html" | grep -E "$regex" | awk -f script.awk - > data $ gnuplot data.gnuplot $ inkscape -d 150 -o graph.png graph.svg

Here's a comparison of /ot/ and /mai/'s activity over the years using two long-standing threads in each。Side note, the process of posting this took waaaaaay too much effort.

import matplotlib。pyplot as plt import numpy as np from datetime import timedelta from time_extract import get_timestamps def plot_data(filename, color, marker): plt。grid(True) plt。yscale('symlog') plt。xlabel('Time') plt。ylabel('Time Δ(days)') data = get_timestamps(filename) delta = np。diff(data) / timedelta(days=1) plt。scatter(data[1:], delta, alpha=0。1, c=color, marker=marker) for i, time_stamp in enumerate(data[1:]): if delta[i] > 200: plt。annotate(time_stamp。date(), (data[i+1],delta[i]),fontsize=8) plt。subplot(2, 1, 1) plt。title('/ot/') #Ponderings general plot_data('ot_16448。txt', 'steelblue', 'o') #Dream thread plot_data('ot_521。txt', 'cadetblue', 'D') plt。subplot(2, 1, 2) plt。title('/mai/') #Who your waifu is plot_data('mai_5。txt', 'orchid', 'o') #Share your daily waifu experience plot_data('mai_8705。txt', 'palevioletred', 'D') plt。gcf()。set_size_inches(15, 14) plt。savefig('scatter', dpi=500)

time_extract

import re import numpy as np from datetime import datetime ts_reg = r'<span class="postername"。*<\/span>\s*(。*)' dt_string = '%m/%d/%y(%a)%H:%M' def get_timestamps(filename): with open(filename, 'r', encoding='utf-8') as file: txt = file。read() matches = re。findall(ts_reg, txt) data = [] for match in matches: dp = datetime。strptime(match, dt_string) data。append(dp) return np。array(data)

It seems despite all my efforts I am fake. Not superficially, but with my whole being. My consciousness is fake. In retrospect I can't relate to things as if I have done them. I know I was there when they were done but it's as if the actions themselves are not mine. Just thought it's very strange.