/sci/ - Science & Math » Thread #11149580

102KiB, 1920x1080, 0*qwkWXc-wzW2D8ggV.jpg

View Same Google iqdb SauceNAO

Anonymous Fri 15 Nov 19:46:30 2019 No.11149580 View Reply Original Report

Quoted By: >>11150922 >>11150926

What are the state-of-the-art methods or algorithms to quantify subjective things?
For example, I tried using quicksort with "manual comparisons" to sort a bunch of pictures, but shit is O(n log n), maybe its fast for computers but for manual comparisons its tiresome, its also not really consistent.
So, any decent method?

Anonymous

Anonymous Fri 15 Nov 2019 19:52:17 No.11149602 Report

Quoted By: >>11151296

I'm pretty sure it's learn-able to become at a particular sorting algorithm as human - and I'm also sure nobody did a large scale statistical experiment on what formal algorithms humans tend to exceed. I might be wrong.

Anonymous

Anonymous Sat 16 Nov 2019 04:18:25 No.11150819 Report

Quoted By: >>11151277

You could do something like playing random pairs of items off each other and adjusting something something like an elo score that will approximately rank them. That'd be $\mathcal{O}(nt)$ where t is the average number of comparisons each item is involved in. I have a feeling that you probably need at least $t = \log{n}$ to get that approximation to be any good though.

Maybe there's something you can do by binning items and then comparing bins and then splitting up bins that have low similarity.

Anonymous

Anonymous Sat 16 Nov 2019 05:06:58 No.11150922 Report

Quoted By:

>>11149580
I use a slightly modified quicksort to manually sort a deck of cards - the pivot isn't the first one picked, I make a lazy effort to find something somewhat close to the middle value. Also towards the end when the arrays get short it payes off to brute-force sort groups of three.

Anonymous

Anonymous Sat 16 Nov 2019 05:08:28 No.11150926 Report

Quoted By:

Anonymous

Anonymous Sat 16 Nov 2019 07:28:33 No.11151277 Report

Quoted By:

>>11150819
>Elo
What the fuck, how I didn't think of that? The thing is made for win/lose comparisons, and I can actually have an accurate "score" for each element. I was thinking about defining this score just sorting them and assigning the index + linear normalization for their "score", but Elo seems much better.
Just the thought of Elo scored boob pics is hilarious rofl

>at least O(n log n) to be good
Quicksort is already n log n, so no losses here.

>Maybe there's something you can do by binning items and then comparing bins and then splitting up bins that have low similarity.
I don't know if its a good idea, since I want to apply it by evaluating some elements with this subjective sorting, attributing a value to them and feeding some attributes from a dataset, and AFTER THAT do the clustering.
I'll probably introduce bias if I do it also before to assigne values to the subjective attributes.

Anonymous

Anonymous Sat 16 Nov 2019 07:34:09 No.11151289 Report

Quoted By: >>11151296

bubblesort

Anonymous

Anonymous Sat 16 Nov 2019 07:37:31 No.11151296 Report

Quoted By:

>>11149602
>>11151289
The algorithms aren't by hand, it runs in a computer but asks your IO input for comparisons(is A better than B?) . Forgot to clarify.

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches