Close

Insights From Amazon Machine Learning

A project log for Analyzing the Hackaday Prize 2015

An unofficial project for finding patterns in cool projects. If you like analyzing stuff and would like to help, please apply to join!

frank-vigilanteFrank Vigilante 06/18/2015 at 05:5910 Comments

I took the most recent CSV (which does not include the most recent data on views), removed some projects with irregular data, removed some attributes that I did not want in the analysis, and uploaded the data into a S3 bucket on Amazon Web Services.

I then created an Amazon Machine Learning model with the modified CSV. Of the attributes included in the analysis, I decided that I most wanted to analyze the factors to determine the likelihood of not just winning a prize, but winning multiple prizes. So I set the "Prize Count" attribute as the Target. Below is a summary of the stats for the numerical categories:

Attributes

Correlations to target

Range

Mean

Median

Comments 0.02909 0 - 58 5.2 2
Number of Entries 0.03509 1.595238095 1
Followers 0.07032 0 - 242 28.8952381 15
Prize Count 1.619047619 1
Skulls 0.12469 0 - 93 10.93333333 5
Views 0.22347 50 - 13300 685.4809524 73

My conclusion from the summary above is that Project Views will most likely translate to prizes on Hackaday!

...but wait, the model also included a categorical attribute:

Attributes

Correlations to target

Unique values

Most frequent categories

Least frequent

Win Date 0.28765 10 5/14/2015 4/7/2015

Wow, the "Win Date" category (which is based on the first date with which a specific project won a prize) had the highest correlation towards winning multiple prizes.

In conclusion, if you wanted to win prizes in the first 10 weeks of giveaways, your best bet was to submit your project as early as possible, and update project logs as much as possible so as to garner more views from the "trending projects" feed. Skulls matter, but not as much as "Views" and "Date Submitted". Comments don't really seem to matter.

Going forward, I am not sure how reliable this data will be as prizes will now be based on community voting as well as inclusion in sponsor lists. All I know is that I am glad that I found Hackaday as a venue for my projects. Prizes are the best!

(I will edit this log with a link to my modified CSV shortly)

Discussions

K.C. Lee wrote 06/28/2015 at 04:21 point

I found my projects in general have a view count to follower ratio almost dead on 1%. Followers to Skulls ratio is about 2:1. (My non-contest projects have much higher view count to follower ratio, but the view counts are low to start out with)

The top 10 list have a much higher view to follower ratios.  Some have similar followers: skull ratios, but a couple has lower ratios i.e. their followers are more impressed.

Looking at the top 10 list exposure, not likely lesser projects would ever get the same level of exposure once they get pushed down the list.  Other than mini-contests, I might as well give up.

  Are you sure? yes | no

frankstripod wrote 06/28/2015 at 06:55 point

I think we were trying to encourage people to enter early for the smaller prizes, and there are more prizes on the way. I am sorry you feel discouraged. Please keep in mind the data is from a list of 213 "winning" projects, that already won random prizes, some just T-shirts for entering. That list that was copied from the blog articles with prizes (there is a list with dates). At that time there were 412 entries, so if a project was advancing greatly, but didn't even win a shirt yet, it was not counted in any of the charts in this project yet. We are working on a more fair list (567 entries now). This project was done in fun, and has no influence on the winners. We just like numbers.

Big prize or not, I must insist you do not give up. IMHO You have the best vision project entered this year. If you do not win a prize at least in the THP Freescale contest and I do, I will surely poop my pants.

  Are you sure? yes | no

K.C. Lee wrote 06/28/2015 at 13:07 point

Thank for the encouragement.  I have won quiet a few of the mini prizes so far including the bulbdial by entering lots of project.  My vision project is on 3 vendor lists which some don't even have 50 entries i.e. shooting fish in a barrel.  Ditto for my 3 other projects mostly in one vendor with few entries, so the odds are pretty good. 

I do difficult projects and use non-popular (IMHO better and cheaper parts) because it is a good way to force myself to learn new things and push my limits.  Thankfully that pays off this year for not using any Atmel parts which has the lowest odd (30% ? or less by the minute) in the mini prize.  If not for those, I wouldn't have bothered.  That unfortunately means that there are little external incentive and supports to go on with the current system for the main contest.  Will alone can only go so far once the momentum runs out and hit road blocks.

Having been crushed last year by the community voting and didn't even made the semi finalist.  Quite a few of the better projects by others are in that same boat last year.  Now comparing the stats between years and they are exactly at the same ratio.  So I don't think the vision would have good odds getting to the first round.  The crowd are just not there.

Given that HaD only like to features the most popular projects by numbers which only make thing worse.  The first couple of pages in the project page shows the big discontinuity between the top projects and the rest and the numbers do not lie.  Some deserves to be on the top, but some are there because of simple concepts -  marketing success.  Simple projects even when they are not going to the final rounds are the ones that catches the public views and interest.  That unfortunately barred entries of others to the semifinalist round last year.  

I do have projects that are entered just for that crowd this year and get featured.  Didn't say I don't analyze the stats and learn to game the system.

  Are you sure? yes | no

frankstripod wrote 06/28/2015 at 21:47 point

Your right, and I can't argue with facts. Some of it is a popularity
contest. I have a lot of fun here, but you know better than I do that
there is a lot of amazing work being done around here, if your know
where to look. As for winning the big prize, I think the judges will
see how things develop and have that ability to know where to look.

If we really wanted to win this thing, its almost not too late to
start putting together a team now for the 2016 prize. I think I may
have the winning formula: One part mad scientist, one part old school geezer electric engineer, and one part under aged web app developer.

I bet your analyzing skills would be useful here, specially after the first semifinal round, if your not too busy winning.

  Are you sure? yes | no

K.C. Lee wrote 06/29/2015 at 00:03 point

My complain was about not even getting to semifinal round no matter how much effort you put in. i.e. project gets blocked just like HR blocks a "good" resume because they are looking for superficial things before it gets to the hands of the hiring manager that otherwise would have been impressed.  The selection from the staffs/community round did not reflect the criteria of the judges - very experienced one that I was trying hard to impressed.  Some of the qualifying projects did not follow the rules for last year, but they were popular and they got in.  Those were pretty much eliminated by the judges.  The damages were already been done as they blocked the more technical (IMHO better projects) from even having a chance.  Thankfully the mini-prizes this year at least make up for some of that as they were before the community voting and there are more prizes.

This is more a popularity contest - something that is simple enough that the average reader can almost replicate i.e Arduino/eval boards or two, but not quite as it would have been done already.  Just need to be hard enough to get some appreciation, but not so hard to be way over their head.  From what I have seen so far base on last year... Some pop culture reference (e.g. tricorder, 3D printer, quadcoptors , SiFi were the theme from last year) , glitters (clock/LED type of projects), flashiness, laziness and may be some outrageousness/danger element (gas powered quadcoptor, big bots). 

And if you want to win, you'll need a PR/marketing person.  There is a different type of engineering to be done for designing a project that get prizes for minimum efforts.  And don't forget about winning by volume of projects.

As for around semifinal time frame, I might have to attend to a product I am supposed to be working on (or so I am told).  I would be pleasantly surprised if I have to work on my entries project.

  Are you sure? yes | no

Eric Hertz wrote 06/29/2015 at 00:14 point

story of life right there, yo.

  Are you sure? yes | no

alpha_ninja wrote 06/19/2015 at 21:20 point

I'd like to recommend checking out correlations with the attributes you listed, but with the target "value in prizes earned". My thinking is that less projects got the more valuable prizes (for example bulbdial, ...) and thus they probably got more exposure...

  Are you sure? yes | no

Frank Vigilante wrote 06/20/2015 at 00:38 point

That sounds logical, and you are probably correct. I am more interested in conducting an analysis on the community voting data we should be getting on Monday. Have a good weekend!

  Are you sure? yes | no

alpha_ninja wrote 06/20/2015 at 16:30 point

Thanks, you too :)

  Are you sure? yes | no

alpha_ninja wrote 06/18/2015 at 14:43 point

Great analysis, I'll have to look into amazon machine learning :)

  Are you sure? yes | no