Page 3 of 14 FirstFirst 1234513 ... LastLast
Results 21 to 30 of 136

Thread: INtpxQUISITION

  1. #21
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,544
    Just woke up covered in mosquito bites itching like a mofo. Counted six so far. Allergic to the fuckers and they love me, a little Irish tipple.

    Update: got at least one of the mutherfuckers. Protip: when you finally see the little fucker on something, don't try to swat it with the sock because they're super quick, instead bring the sock slowly closer to them until it's about two inches away then press quickly.

    Quote Originally Posted by Lud View Post
    This seems like a nerd version of "fuck, marry, kill"
    "cosplay with, be dungeon master for, ..." it's difficult translate those to nerd.

    Quote Originally Posted by Deckard View Post
    Nice idea. I'm curious about how the scoring works. Could you post the code in a different way? The attachment doesn't seem to be working.

    It might be because of how people answered questions on the advanced version, but I expected there to be more of a disparity in scores. Maybe the "----" rating could be weighted higher since it represents a deal-breaker answer.
    The scoring at the mo is pretty simple. Not sure what's up with the attachments not working but the code is here. Ideally I should github it or something but I don't have a github account (yet) that's not linked to my rl meatbag identity.

    I think there's more disparity than might be immediately evident since about ~0.4 in similarity would be expected at random. I could bump "----" a bit, yup, but I think it would not affect too much the scores (I didn't look at the responses in too much detail but I think there were few "----" or "++++".

    Also updated results with your answers (see post 2)!! ()

    Quote Originally Posted by Dot View Post
    I looked back at my answers and realized that I accidentally reversed the +'s and -'s for at least one (the armpits one). So I don't know how accurate my answers are.
    I changed your answer on that swapping the "-" and "+". () (see post 2)

    Quote Originally Posted by ACow View Post
    In practice people cluster a lot more within a possible measurement scale than they realise. I think i joked in one of the other political threads that as far as the political compass goes, practically the entirety of the online community (or at least the ones I involve myself with) cluster the in the bottom left quadrant.
    I guess we're all INTPs too. Actually I sorta cleaned up the quiz in that respect, like I removed a lot of questions that were very redundant with MBTI.

    I wonder could some actual clustering algorithm be applied. There's definitely some clusters of similar users.

    Depending on how much @Utisz likes coding, he could standardise/normalise based on the entire distribution of scores received.
    Was thinking about that! Like weighting unusual answers that coincide more than answers that are very common. I've no problem coding that but the first issue is how to do that weighting cleanly and the second is that I'm not really sure I want the answers changing between an existing pair of users every time a new user posts.

    I also like the idea of scoring the preferences in a variety of ways: say negatives push people apart but strength of positives are all just absolute matches and then the scores are standardised to place relative weight on strength of disagreements (and vice versa).

    And then there's the possibility of reporting on some of the correlations and clustering within the questions themselves :P
    Some good ideas! I like the idea of clustering on the questions themselves. (I prefer that to varying the metrics like having analysis that does not compete.) Something like association rule learning maybe.

    The other idea was to score questions into several categories and give similarities on that. I might do that.

    Also at the moment the similarity is completely binary, even when >2 answers are available (coincides or not). I could weight more similar answers. I guess though it won't have a major effect.

  2. #22
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,544
    Missed this post earlier ...

    Quote Originally Posted by Deckard View Post
    Yeah that's true for basic correlations, I guess I just assumed the +s and -s would count for more. For instance, if I have a strong preference that a question is answered one way or another, that should count for a lot more than an ordinary match / lack of match. If someone hits 2 or more answers I specify as deal-breakers, that could begin to outweigh any other accumulated matches. Something like:

    Unweighted match = 1

    + = 2
    ++ = 4
    +++ = 6
    ++++ = 10

    - = 2
    -- = 5
    --- = 10
    ---- = 25

    So for example if someone hits 2 of my dealbreaker questions where there would otherwise be a 50% overall match, it'd change the score from:

    56/112 = 50%

    to

    56/(110+25+25) = 35%
    Perhaps that makes more sense and maxing out at 4 "-" or 4 "+" is too weak. Though I'm a bit reluctant to change officially since the values feel a bit arbitrary (somehow 2, 4, 8, 16 would feel more satisfying for some reason /typical cs guy ) it's easy to change these values in the code. In fact here's the results for your values:

    Spoiler: results with new pref vals

    Preferences [Deckard Version]


    ######### ######### ######### ######### ######### ######### ######### ######### ######### #########
    Deckard
    Dot
    gator
    Guess Who
    Lud
    Madrigal
    oxyjen
    Ptah
    TeresaJ
    Utisz
    Deckard
    1.000
    0.825
    0.889
    0.437
    0.873
    0.733
    0.832
    0.727
    0.738
    0.905
    Dot
    0.975
    0.942
    0.880
    0.422
    0.824
    0.812
    0.873
    0.726
    0.880
    0.915
    gator
    0.729
    0.647
    0.830
    0.654
    0.755
    0.765
    0.811
    0.623
    0.820
    0.660
    Lud
    0.956
    0.880
    0.883
    0.612
    0.988
    0.911
    0.912
    0.842
    0.707
    0.850
    Madrigal
    0.757
    0.695
    0.776
    0.732
    0.715
    0.993
    0.754
    0.479
    0.624
    0.687
    Utisz
    0.855
    0.774
    0.804
    0.608
    0.755
    0.679
    0.818
    0.578
    0.775
    0.961


    Spoiler: standard version for comparison

    Preferences [Standard Version]


    ######### ######### ######### ######### ######### ######### ######### ######### ######### #########
    Deckard
    Dot
    gator
    Guess Who
    Lud
    Madrigal
    oxyjen
    Ptah
    TeresaJ
    Utisz
    Deckard
    1.000
    0.744
    0.837
    0.395
    0.814
    0.625
    0.750
    0.610
    0.628
    0.860
    Dot
    0.959
    0.898
    0.874
    0.405
    0.787
    0.770
    0.804
    0.630
    0.798
    0.874
    gator
    0.761
    0.673
    0.863
    0.680
    0.745
    0.796
    0.804
    0.608
    0.813
    0.686
    Lud
    0.933
    0.879
    0.848
    0.574
    0.980
    0.860
    0.878
    0.758
    0.723
    0.788
    Madrigal
    0.719
    0.662
    0.745
    0.732
    0.686
    0.985
    0.719
    0.423
    0.641
    0.642
    Utisz
    0.838
    0.751
    0.784
    0.587
    0.749
    0.660
    0.799
    0.574
    0.755
    0.956



    (actually not a huge amount of difference, or in fact bunching the scores even more: not 100% too sure why but perhaps having "----" be so extreme, it pushes the possible min range down a lot)



    [edit] @Utisz With the preferences results grid, I assume the vertical axis is people who did the advanced questionnaire, and each cell in my row represents how well they matched to my preferences (as opposed to how well I matched to theirs)? Is that right?
    Yup, that's it! (While the Similarity results are symmetric (sim(a,b)=sim(b,a)), the Preferences are not.)

  3. #23
    (╯°□°)╯︵ ┻━┻ Deckard's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    1,085
    Quote Originally Posted by Utisz View Post
    (actually not a huge amount of difference, or in fact bunching the scores even more: not 100% too sure why but perhaps having "----" be so extreme, it pushes the possible min range down a lot)
    Values seem to have gone up across the board. I think it's just a reflection on the fact that we're a relatively homogeneous group and probably didn't hit any ----s. Also INTPs tend to be pretty easygoing and probably aren't setting stringent requirements on potential partners. It would get a bit more interesting in a more diverse group. Maybe you should try this experiment over at TypoC.

  4. #24
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,544
    Quote Originally Posted by Deckard View Post
    Values seem to have gone up across the board. I think it's just a reflection on the fact that we're a relatively homogeneous group and probably didn't hit any ----s. Also INTPs tend to be pretty easygoing and probably aren't setting stringent requirements on potential partners. It would get a bit more interesting in a more diverse group. Maybe you should try this experiment over at TypoC.
    I looked at it a bit more and there were a few "----" hit, but the thing is that with the current metric missing a "----" and hitting a "----" would balance back out to 0.5 no matter what strength the values assigned. The only way to get 0 is to hit all "----" perfectly. (The same thing happens for "++++" and all other evaluations.) The metric is: (s - min)/(max - min) where s is the sum of all values for all questions, max is the maximum possible score on all questions answered by the second user, and min is the corresponding minimum possible score.

    What I could do to is normalise the results to a scale of −1/1 (where 0 is neutral) instead of 0/1 (where 0.5 is neutral). The substantive interpretation of the results don't change but the psychology of them would.

    Agreed on the INTP thing; it's definitely a factor. With regards the homogeneity of the group, a possible workaround is something like ACow says: normalise by distributions of other peoples' answers, so for example questions where everybody answers the same are weighted lower than questions, for example, where two members agree on a rare answer. This could be used to factor out the homogeneity in the group. This would seem more appropriate for similarity though than preferences (I don't think it would be a good idea at all for preferences). The one thing I don't like is that the answers for existing pairs would change as new results are added (that to me seems a bit weird).

    Posting on TypoC might be interesting yep ... not sure. I might do it if we could use it to drum up some interest on this site. (I wonder would they be averse to me linking to here?) Anyways, that's not something for right now.

  5. #25
    (╯°□°)╯︵ ┻━┻ Deckard's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    1,085
    Also, I reckon the questions & scoring might need a bit of tweaking if it's to predict compatibility. At the moment I think it'd be more skewed towards similarity matching (even with the prefs weighted higher) so INTPs will tend to match highest with INTPs and so on. To make it more insightful about actual compatibility I think you'd need to get deeper into relationship voodoo, maybe try to divine which traits complement others and which repel. You could even integrate some MBTI-ish questions and partially model the scoring on real data of MBTI pairings (there's probably research on that?).

  6. #26
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,544
    Quote Originally Posted by Deckard View Post
    Also, I reckon the questions & scoring might need a bit of tweaking if it's to predict compatibility. At the moment I think it'd be more skewed towards similarity matching (even with the prefs weighted higher) so INTPs will tend to match highest with INTPs and so on. To make it more insightful about actual compatibility I think you'd need to get deeper into relationship voodoo, maybe try to divine which traits complement others and which repel. You could even integrate some MBTI-ish questions and partially model the scoring on real data of MBTI pairings (there's probably research on that?).
    Yeah, the self-scores on preferences indicate that the questions are not hitting the aspects of one's self that one would not like in a partner and vice versa (or maybe it's just that INTPs are really into other INTPs). It would be an interesting exercise to think of questions that one would answer one way but prefer the other way ... the confidence-based ones I thought might swing that way; I guess also for INTPs the E/I, T/F dimensions would have been good to emphasise a bit more in the questions in that respect.

    When designing the compatibility aspect, I thought about some voodoo of trying to learn higher-level features and then mapping those back to individual questions (i.e., correlating questions that probably say something about each other), but without any way to really test the model "scientifically", it would be shooting in the dark or would rely a lot on my intuition and would be a complex array of numbers I come up with. Perhaps there's a way to extract something from the data, but for that we'd need a gold standard of which users are actually compatible. I figured the easiest and best way to figure out compatibility would be to just ask people directly.

    It might be fun to still try to conjure up some intuitive voodoo, but it would be impossible to say what the results really mean. I guess that's what OKcupid does, though maybe they do have a gold standard to know which relationships worked out, or how long messages were exchanged, etc., to actually evaluate variations. Also I like to think that compatibility between two people (within a population like this where we're more or less likeminded) is probably not so measurable like this. It's more interesting to me to see how members match up against the preferences they state themselves rather than trying to sell an arbitrary notion of compatibility.

    On a side note, I noticed that OKc asks inverses of questions, like "do you smoke", and "what do you think of people who smoke". That's already generalised above.

  7. #27
    malarkey oxyjen's Avatar
    Type
    INtP
    Join Date
    Dec 2013
    Posts
    1,322
    Quote Originally Posted by Utisz View Post
    Also, given that y'all are sending me potentially "intimate" answers, to keep things reciprocal, I'm happy to PM my own extended answers to anybody who already sent me answers for most/all questions in the extended version so long as they don't share them on.

    So if you already did the advanced version, just let me know (PM, rep, post, whatever) if you wanna see my answers.


    Also code is here (Java):

    Attachment 1088
    It was necessary to see our answers for scoring, so I don't feel like you need to reciprocate. It was weird though, to see you with dissimilarity scores in the same ranges as Guess Who and Ptah. I already knew those guys think differently than a lot of people here.
    ---
    Also I'm Madrigal's top similarity match? Hey girl hey
    Last edited by oxyjen; 01-28-2017 at 05:51 PM.

  8. #28
    (╯°□°)╯︵ ┻━┻ Deckard's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    1,085
    Quote Originally Posted by Utisz View Post
    I looked at it a bit more and there were a few "----" hit, but the thing is that with the current metric missing a "----" and hitting a "----" would balance back out to 0.5 no matter what strength the values assigned. The only way to get 0 is to hit all "----" perfectly. (The same thing happens for "++++" and all other evaluations.)
    Oh ok, that's a bit different to the scoring system I was thinking of. What I was suggesting was pretty basic: ordinary misses/hits count for 0/1 votes, ranked +s count for 2-10 votes, ranked -s count for 0 votes. The -s have an effect by increasing the total vote count which you're eventually dividing by. That means missing a ---- has a different effect to hitting a ----, and you get a 0 by missing everything (it doesn't matter if those misses were ranked -s).

    Bit of a messy explanation so here's a quick example:

    Code:
    q1: match (+1/+1)
    	1/1
    q2: mismatch (+0/+1)
    	1/2
    q3: match (+1/+1) & ranked hit++ (+4/+4)
    	6/7
    q4: mismatch (+0/+1) & ranked hit---- (+0/+25)
    	6/33
    q5: match (+1/+1) & ranked hit- (+0/+2)
    	7/36
    
    total: 19.4%
    [edit] I changed the example a bit for clarity. Q5 shows how you can get a match but still hit a ranked "-" score, which occurs if you preferenced a "-" value for your own answer (i.e. you want your partner to answer differently to you on that question).
    Last edited by Deckard; 01-28-2017 at 05:36 PM.

  9. #29
    Senior Member
    Type
    intp
    Join Date
    May 2014
    Posts
    1,402
    How is it that in preferences, people aren't 1.000 with themselves? I mean, except for Deckard. If possible, please explain in non-mathsciency-witchcraft terms.
    "Doesn't matter what a man has if he doesn't have purpose. You take that away from him, man usually goes with it." -Beau

  10. #30
    Merry Christmas Dot's Avatar
    Type
    INFP
    Join Date
    Dec 2013
    Location
    near a castle
    Posts
    3,423
    Quote Originally Posted by Lud View Post
    How is it that in preferences, people aren't 1.000 with themselves? I mean, except for Deckard. If possible, please explain in non-mathsciency-witchcraft terms.
    I don't put things back on their original shelves in grocery stores. I would never willfully associate with an asshole who behaves like I do.
    "Better not to feel too much until the crisis ends—and if it never ends, at least we’ll have suffered a little less, developed a useful dullness...The constant—and very real—fear of being hurt, the fear of death, of intolerable loss, or even of “mere” humiliation, leads each of us, the citizens and prisoners of the conflict, to dampen our own vitality, our emotional and intellectual range, and to cloak ourselves in more and more protective layers until we suffocate." - Toni Morrison

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •