Language Games Further Research Notes

Language Games Further Research Notes LOGISTIC REGRESSION ON WORD VECTORS dataset model precision recall f1 support ——— ———————— ———– ——– ——– ——— Dev BoW WVs Logistic 0.736397 0.7603 0.748157 1602 Dev Initial BoW WVs Logistic 0.75089 0.790262 0.770073 1602 Dev Diff BoW WVs Logistic 0.453061 0.41573 0.433594 1602 Dev Plus BoW WVs Logistic 0.730303 0.752185 0.741082 1602

I haven’t messed around with hyperparameters at all; we’re definitely overfitting in the last model.

BATCH_SIZE = 96 NUM_EPOCHS = 20

Testing out monthly layers with cv_3layer_monthly = { ‘feature_columns’: [activity_features, monthly_wv_features, initial_wv_features], ‘dropout’: 0.1, ‘optimizer’: tf.train.ProximalAdagradOptimizer( learning_rate=0.1,l1_regularization_strength=0.01) } and depths = [[1024, 512, 256], [512, 256], [256]]:

dataset model precision recall f1 support ——— ————— ———– ——– ——– ——— CV Monthly 3 layer 0.569815 0.458299 0.508009 1211 CV Monthly 2 layer 0.533881 0.45815 0.493125 1135 CV Monthly 1 layer 0.486653 0.429348 0.456208 1104

Using three hidden layers:

dataset model precision recall f1 support ——— ———————————- ———– ——– ——– ——— CV 2 Monthly drop 0.1 0.525667 0.433898 0.475395 1180 CV 2 Monthly drop 0.2 0.578029 0.448606 0.505159 1255 CV 2 Monthly drop 0.30000000000000004 0.706366 0.452334 0.551503 1521 CV 2 Monthly drop 0.4 0.620123 0.454135 0.524306 1330 CV 2 Monthly drop 0.5 0.602669 0.456454 0.519469 1286 CV 2 Monthly drop 0.6 0.62423 0.447059 0.520994 1360

dataset model precision recall f1 support ——— ————— ———– ——– ——– ——— DEV Initial 0.652934 0.454585 0.535998 2301 DEV Initlal+Monthly 0.502497 0.456867 0.478597 1762

It looks like what I’m seeing is that

REVISION https://web.stanford.edu/class/cs224u/materials/cs224u-2015-presenting.pdf

So I’m a bit confused now because the embeddings are definitely different but the looked-up word vectors appear to be the same. I must have looked them up incorrectly, using the same embedding.

QUALITATIVE ANALYSIS: If there isn’t much surface exclusion, then what’s going on?

Strategy: Sample 10 each from those who dropped out and those who stayed. What can we learn?

LEAVERS: 6437 markkoberlein 57 ericfranklin 5548 botswana99 9986 BrokenEnso 8413 seer 8688 jscalisi 10016 squintychino 10468 travisfischer 3257 utoku 6218 tbastos

markkoberlein

markkoberlein on Feb 5, 2010 | parent [-] | on: Forget About Photoshop: 5 More Ways to Stop Using …

I have to admit that there are some nice CSS hacks there like the gradient and drop shadow effects but it’s dependent on proprietary CSS which should be avoided if possible. If you start to rely on proprietary CSS then you are setting yourself up for problems. Isn’t this part of the reason why we hate IE because we are forced to use CSS code like this? I have to think that all these effects will eventually be incorporated into the CSS standard but until then you can achieve all of those effects with small image files and simple CSS.

(image/gif)

rimantas on Feb 5, 2010 [-]

First: it’s not “propriertary CSS” it is called “vendor extensions” and are allowed in CSS. Second, and more important:

This is an unpopular comment. He’s trying to articulate the community ideology and getting it wrong.

I have to admit that there are some nice CSS hacks there like the gradient and drop shadow effects but it’s dependent on proprietary CSS which should be avoided if

markkoberlein on Jan 31, 2010 | parent [-] | on: Steve Jobs calls Google’s “Don’t be evil” mantra “…

You know I miss the days when Apple (as in Steve) went after Microsoft and IBM with a vengeance. That battle didn’t turn out too well for him, but it was fun to watch. It would be fun to see Apple and Google go at it. The history of the PC vs. Mac wars showed us that an open platform (as in able to be installed on any non-apple device) wins over a closed platform controlled by one company. It will be interested to see what happens this time around.

(image/gif)

austinshea on Jan 31, 2010 [-]

I think the PC vs Console software sales showed us that you can make a lot more money, if you have nintendo style control.

A string of rejections of his point.

57 ericfranklin Really only interested in diamonds. His interests are peripheral to the community’s.

5548 botswana99 Basically just here to post jobs.

9986 BrokenEnso

(image/gif)

BrokenEnso on Nov 10, 2014 | parent [-] | on: If money doesn’t make you happy, you probably aren…

“if you aren’t able to meet a certain reasonable minimum standard of living, you will be absolutely miserable.” For some people their minimum standard of living dose not require money. Buddhist monks[1] come to mind. There are also people that simply want to live away from society and thus have no real need for money. [1] http://www.buddhanet.net/e-learning/buddhistworld/layguide.h…

(image/gif)

freshhawk on Nov 10, 2014 [-]

I think the discussion was assuming you didn’t have someone taking care of your material needs for you, like a monastery or a parent.

Is offering a different perspective, gets corrected in a snarky way.

BrokenEnso on July 17, 2014 | parent [-] | on: North Texas citizens organize to monitor police wi…

Not really, no. It’s just not that dangerous. We imagine it must be dangerous, and so do cops themselves. But you’re a lot more likely to be killed as a garbage collector or airline pilot, or a construction worker, or plenty of other jobs. Adding a bit of objective support to that statement is a report [1] from BLS that lists Mortality Rates by industry, where law enforcement is grouped with Government. [1] http://www.bls.gov/iif/oshwc/cfoi/cfch0011.pdf

(image/gif)

jjoonathan on July 17, 2014 [-]

Doesn’t grouping police into “Government” dilute the danger of policework by grouping it with a bunch of relatively safe office jobs? I suspect I’m looking at the wrong plot because there’s no way anyone would intentionally suggest such an approach…

(image/gif)

BrokenEnso on July 17, 2014 [-]

Very good point. There is an NIH study [1] that drills into Law Enforcement Officers (LOEs). The number in the NIH study can be contrasted with the right side of slide 14. The mortality rate does raise a but from the Government rate, from 2/100k to 5.6 when you are talking homicide and 11.8 when considering all causes of fatalities; still puts it 3rd compared to other industries. [1] http://www.ncbi.nlm.nih.gov/pubmed/20564516

(image/gif)

jjoonathan on July 17, 2014 [-]

Thanks for taking the time to dig that up! I haven’t been able to conclusively determine the units in the NIH study, but we should be careful about normalization. I don’t think these are comparable as they stand:

This kind of interaction seems like it would be reinforcing to the community: affirmation of the value of objective support.

8413 seer Didn’t actually leave! (Maybe I need to improve my filters.)

8688 jscalisi Job-posting bot.

10016 squintychino

squintychino on Apr 25, 2014 | parent [-] | on: Eating Alone by Design: An Entire Restaurant With …

I think the point here is that eating does not require a social component. Food does not taste better by virtue of having another person present.

—> Rejects others’ experiences.

squintychino on Apr 20, 2014 | parent [-] | on: Ask HN: Idea Sunday

They have this. It’s called ‘any job search engine out there’. They all let you put in these search parameters to get back jobs that meet your criteria.

—> Sarcasm.

squintychino on Apr 5, 2014 | parent [-] | on: Amazon Dash

I doubt Amazon would be stupid enough to leave this possibility unaccounted for. Even the most basic purchase on Amazon requires you to put in a password to confirm your purchase. Why would this be any different? Or are you simply being a model HN contributor and nitpicking any little thing that comes to mind? —> Snark. Identifies negative stereotype about community.

squintychino on Mar 25, 2014 | parent [-] | on: The Millionaire Fastlane

Perfect. Please keep to the right and stay in the slow lane. The last thing we need in the fast lane is people like you telling us we can’t pass others.

10468 travisfischer

travisfischer on Aug 13, 2014 | parent [-] | on: Show HN: Stressed and burned out, so I created my …

I clicked through expecting an app that offered a nostalgic walk down memory lane with screenshots and features of the Windows operating system. I’m not sure why that is the first thing that came to mind, but it did. While that would have been moderately amusing, the actual application is a much more useful and helpful concept. Great job on launching something!

—> Has a lot of pretty insubstantial affirmations. Doesn’t get a lot back.

3257 utoku

utoku on Nov 12, 2014 | parent [-] | on: Philae has landed

Its speed is most likely relative to Sun. Earth moves at about 30km/s relative to sun. So that’s your minimal starting speed.

spiritplumber on Nov 13, 2014 [-]

Speeds are generally given relative to the object being orbited, so in this case the Sun.

—> Kind of a dry, condescending reply? Unfortunately they are not roughly equivalent. A solid shell or a ring around a star is not a Dyson sphere. The correct Dyson sphere is composed of many separate orbiting elements. A solid shell or semi-shell is unstable and is bound to collide with the star or just wander away. Yes, the Ringworld is unstable. This was one of Niven’s mistakes and misunderstanding of Dyson spheres.

tutuca on Feb 5, 2014 [-]

I wouldn’t say a “mistake” he explores that inestability in “Return to Ringworld”…

—> Interested largely in mathematical ideas and astronomy—perhaps the community moves away from these over time?

6218 tbastos interests: drugs, religion, spirituality

STAYERS: 1155 jon_dahl 1509 bri3d 6768 nostrebored 5534 jdoliner 1679 jackalope 9888 AdmiralAsshat 7307 xentronium 5474 MiguelHudnandez 9948 TaylorAlexander 3438 projektir

1155 jon_dahl

jon_dahl 55 days ago | parent [-] | on: Show HN: Mux Video, a simple API to powerful video…

Not a dumb question. We’re lower-level - the biggest value Wistia brings is helping marketers be more effective with video. Mux Video is like the plumbing; we get the video from ingest to device/player efficiently with high quality, but we don’t actually offer a video player, engagement tracking, content management, metadata, CRM integrations, etc.

jon_dahl 55 days ago | parent [-] | on: Show HN: Mux Video, a simple API to powerful video…

Yes, but with a sightly different pricing structure. Send us an email - I’d be interested in hearing more about your use case.

—> Speaks with authority

1509 bri3d bri3d 13 days ago | parent [-] | on: OpenSimpleLidar

This is the same low-cost triangulation approach used in other inexpensive laser rangefinders, like the one on a Neato vacuum: https://www.diva-portal.org/smash/get/diva2:995686/FULLTEXT0… It’s dramatically different from the $20k+ LIDAR units you’d find on a self-driving car, which generally use time-of-flight.

—> Assertive, factual content

6768 nostrebored

nostrebored 7 months ago | parent [-] | on: Tesla and SpaceX Share More Than Musk

Typically giving rent seekers political office is a terrible idea.

—> pithy, specific

nostrebored 8 months ago | parent [-] | on: E-commerce will evolve next month as Amazon loses …

This is an astoundingly naive view of what credit card fraud costs. The merchants are not doing this charitably. Your cost will be amortized somewhere.

nostrebored 8 months ago | parent [-] | on: E-commerce will evolve next month as Amazon loses …

You say this like it’s somehow a bad thing? Amazon has built an extremely efficient distribution network comprised of all major shipping companies and a vast array of third party sellers. The price you’re seeing is the result of billions of dollars of investment paying off.

5534 jdoliner

(image/gif)

jdoliner 12 days ago | parent [-] | on: CometML wants to do for machine learning what GitH…

Pachyderm, a project I work on, is probably as close as you’ll find to something that ties all 3 together. In my mind the major unsolved problem here was the data versioning so that’s the first thing we tackled. Code versioning is already quite well solved so we just integrate with existing tools for that. I’m not convinced that model versioning is actually distinct from data versioning, models are just data after all. So I think without an established system for versioning models, such as Git + Github is for code, treating models as data and versioning them that way is good enough for government work. From what I can tell CometML isn’t quite versioning models so much as tracking versions of models. It expects that models to be stored and versioned elsewhere but it gives you a way to get deeper insight into how those models are performing, how they’re changing, the hyper-parameters used to train them etc. Tracking this is also a very important problem that CometML seems to solve quite elegantly.

—> jargon

jdoliner 58 days ago | parent [-] | on: Fake news is not the real problem

This comment is peak HN arrogance. People rally behind charlatans because they don’t have enough proximity to us engineers to think sanely about things? Maybe we could make them an app where they can get one of us engineers on the line so they’re finally able to use their brains again and realize that they need to vote democrat.

—> positions self as within the community; makes claims about community. Deictic differences?

1679 jackalope

jackalope on Nov 20, 2014 | parent [-] | on: EIZO Announces Monitor with 1920x1920 Resolution

I’ve just set up 3 different work areas consisting of laptops with external second monitors. I’ve settled on the configuration having the external monitor above the laptop display as being the most optimum. I find that I never need to move my neck to glance from one screen to the other, and only require a slight tilt at the waist to comfortably switch for longer periods. Contrast this to a side-by-side configuration, where it feels awkward to merely shift my eyes sideways, so I move my neck more. This in itself isn’t so bad, until I need to focus on one screen that’s off-center for prolonged periods, keeping my head in an awkward angle that doesn’t seem healthy. Naturally, any configuration should still be augmented with regular stretching/activity breaks. tl;dr: Vertical: Mostly eye movement. Horizontal: Lots of neck movement.

9888 AdmiralAsshat

AdmiralAsshat 1 day ago | parent [-] | on: Should I just stop attempting to make these trivia…

Please add (2004) to the title.

—> Enforcing discourse norms

AdmiralAsshat 2 days ago | parent [-] | on: Guild Wars 2 distributed spyware in order to detec…

We understand that your Guild Wars 2 accounts are important to you and we take that trust very seriously. Our goal is to continue to foster a safe and fair community for all. We believe that everyone deserves transparency, which is why we’re providing this additional information. Wow, what a non-apology.

—> Emoting/responding in a particular way.

AdmiralAsshat 2 days ago | parent [-] | on: OLPC’s $100 laptop was going to change the world

Was pretty excited at the pi-top on first glance, but, $319 for what’s basically a screen and a clamshell enclosure to the Pi? Yikes. Surely we can do better. The Superbook is a clamshell laptop powered by your smartphone and costs half that.[0] [0]http://www.sentio.com/

—> Wow, Yikes.

7307 xentronium

(image/gif)

xentronium 3 months ago | parent [-] | on: Regarding the Em Dash

Depends on typography traditions. In Russian, em-dash in the middle of a sentence is always typeset with short spaces. We also rarely use en-dashes (I think there are maybe two cases when it is allowed: in date ranges like 1940–1945 and as a bullet point in lists).

—> INterests —> Interests: Typography. em-dash en-dash bullet point

5474 MiguelHudnandez

TaylorAlexander 17 days ago | parent [-] | on: Biologist E.O. Wilson suggests ‘moon shot’ conserv…

Whoa, whoa, whoa, hold on there. You disagree with a claim in the article. Someone replies asking you to cite evidence for your opinion. And you respond telling them you’re not going to do the research for them? Why would you even make the claim of disagreement in the first comment if you can’t already provide information to back it up? I like that after you refuse to support your own claims, you then tell the person to grow up. Awesome. —> Enforcing community norms.

9948 TaylorAlexander

TaylorAlexander 2 days ago | parent [-] | on: Amazon warehouse workers ‘peed in bottles’ fearing…

My fear is that the future of work for most people is not unemployment, but shitty jobs like this. My feeling is that in order to avoid this and prosper, the people need to own the robots that do all the work (Marxism heyy). If robots provide huge productivity gains, then they will provide those gains for their owners. If it’s the big corporations that own them, they will see the gains. But as long as people are treated like machines, they will be left out of this prosperity. Instead, if the people own the machines collectively, then they can enjoy the productivity gains themselves as a group. What do you all think of this?

3438 projektir

TAKE AWAYS

I have some bad data that I should clean up.
Remove users from [label=0] if they have posted in the last 6? months.
Remove bots. They can be identified by replicated posts. (Jaccard similarity.)
What predicts people leaving the community?
Rejection.
Contradiction (negative sentiment replies)
No response
Downvotes
Particular interests
Drugs, spirituality, math, astrophysics
What predicts people staying in the community?
Authoritative voice
Performs emotional responses like wow, yikes.
pithy, specific, confident
Quotes parent
Substantive, engaged responses

Jargon Software: versioning, snapshot, DevOps, OCR, Statistics: sample, likelihood,

Emoting on behalf of the audience whoa, yikes, wow