Stats Junky vs. Prosper202

johnysc430 said:
give me a specific example of how you would want multivariate applied.
What would be the variables?
Click to expand...

Multivariate split testing for flog:

3 different headlines

4 different body types

2 different testimonial types

.. all at the same time. Implementation would need to be with three sets of codes. One code for headlines, one for body one for testimonial. Basically these could be used for anything - what you're doing is testing three variables each with it's own variables.

Check out Split test accelerator.
 


JoseArmando said:
Multivariate split testing for flog:

3 different headlines

4 different body types

2 different testimonial types

.. all at the same time. Implementation would need to be with three sets of codes. One code for headlines, one for body one for testimonial. Basically these could be used for anything - what you're doing is testing three variables each with it's own variables.

Check out Split test accelerator.
Click to expand...


This is a great Idea, let me see what I can do. I wouldn't rule this out.
 
JoseArmando said:
Multivariate split testing for flog:

3 different headlines

4 different body types

2 different testimonial types

.. all at the same time. Implementation would need to be with three sets of codes. One code for headlines, one for body one for testimonial. Basically these could be used for anything - what you're doing is testing three variables each with it's own variables.

Check out Split test accelerator.
Click to expand...


This one is super easy to use:
Home - genetify - GitHub
 
I'm thinking of appending, a code to each sub id for each variation. because there can be so many variations, you can upload the appended codes back into some type of script and figure out which variation converted.

like title 1,2,3,

ad copies 1,2,3

images 1,2,3

would append to the sub id like 231subid 322subid

is this what you guys would want?
 
Prosper is open source. When is that badass motherfucker going to integrate it with the Yahoo/MSN API? (Dont mind paying for Goog API costs either)
 
johnysc430 said:
I'm thinking of appending, a code to each sub id for each variation. because there can be so many variations, you can upload the appended codes back into some type of script and figure out which variation converted.

like title 1,2,3,

ad copies 1,2,3

images 1,2,3

would append to the sub id like 231subid 322subid

is this what you guys would want?
Click to expand...

this sounds good but if you have a shitton of campaigns won't there be duplicate subids eventually?

well i dunno. you figure this shit out and ill pay for it
 
johnysc430 said:
I'm thinking of appending, a code to each sub id for each variation. because there can be so many variations, you can upload the appended codes back into some type of script and figure out which variation converted.

like title 1,2,3,

ad copies 1,2,3

images 1,2,3

would append to the sub id like 231subid 322subid

is this what you guys would want?
Click to expand...


sounds good
 
johnysc430 said:
I'm thinking of appending, a code to each sub id for each variation. because there can be so many variations, you can upload the appended codes back into some type of script and figure out which variation converted.

like title 1,2,3,

ad copies 1,2,3

images 1,2,3

would append to the sub id like 231subid 322subid

is this what you guys would want?
Click to expand...

yeah but you'd want db table entries I'd imagine- so that you can easily sort to look at CTR and CR from say ad copies only. and then images only, etc.

Ideally break this down onto one screen that easily and quickly shows the best combo or if this hasn't been established clear winners/losers (ad 3, or images 2, etc)

I currently track that all w/ prosper via multiple layers of the lp split test script and many advanced landing pages .js. I pull in variables like this, along with link placement (since most lp's have many) all hyphenated into the subid.

Problem is I have to fire up excel and bust out pivot tables to decipher the info, and this is annoying.
 
johnysc430 said:
I'm thinking of appending, a code to each sub id for each variation. because there can be so many variations, you can upload the appended codes back into some type of script and figure out which variation converted.

like title 1,2,3,

ad copies 1,2,3

images 1,2,3

would append to the sub id like 231subid 322subid

is this what you guys would want?
Click to expand...

flysarescary said:
sounds good
Click to expand...

Sonny Forelli said:
yeah but you'd want db table entries I'd imagine- so that you can easily sort to look at CTR and CR from say ad copies only. and then images only, etc.

Ideally break this down onto one screen that easily and quickly shows the best combo or if this hasn't been established clear winners/losers (ad 3, or images 2, etc)

I currently track that all w/ prosper via multiple layers of the lp split test script and many advanced landing pages .js. I pull in variables like this, along with link placement (since most lp's have many) all hyphenated into the subid.

Problem is I have to fire up excel and bust out pivot tables to decipher the info, and this is annoying.
Click to expand...

Listen to Sonny, if you try and piece together an identifier from a group of primary keys and then parse out the data after the fact to correlate your winners you're gonna murder the shit out of your server when MySQL (or whatever you use) starts kicking it in the nuts aggregating and reporting the data.

Each variable should be it's own column in an aggregate type table that the SubId derives from. If you have only two variables your table schema would look like:

Code: 
SubId (PK) | AdId | OfferId
1                1        1
2                2        1
3                1        2
4                2        2
Your dataset size is going to grow from there depending on the number of factors you test horizontally (offers/ads/lps/etc...) and vertically (variations on a single variable like different ad copies).

So if you added LP variations to the test above your schema and your SubIds become:

Code: 
SubId (PK) | AdId | OfferId | LPId
1                1        1           1
2                2        1           1
3                1        2           1
4                2        2           1
5                1        1           2
6                2        1           2
7                1        2           2
8                2        2           2
9                1        1           3
10              2        1           3
11              1        2           3
12              2        2           3

Not normalzing your data to begin with is asking for heartache later.

m2c
 
justo_tx said:
Listen to Sonny, if you try and piece together an identifier from a group of primary keys and then parse out the data after the fact to correlate your winners you're gonna murder the shit out of your server when MySQL (or whatever you use) starts kicking it in the nuts aggregating and reporting the data.

Each variable should be it's own column in an aggregate type table that the SubId derives from. If you have only two variables your table schema would look like:

Code: 
SubId (PK) | AdId | OfferId
1                1        1
2                2        1
3                1        2
4                2        2
Your dataset size is going to grow from there depending on the number of factors you test horizontally (offers/ads/lps/etc...) and vertically (variations on a single variable like different ad copies).

So if you added LP variations to the test above your schema and your SubIds become:

Code: 
SubId (PK) | AdId | OfferId | LPId
1                1        1           1
2                2        1           1
3                1        2           1
4                2        2           1
5                1        1           2
6                2        1           2
7                1        2           2
8                2        2           2
9                1        1           3
10              2        1           3
11              1        2           3
12              2        2           3
Not normalzing your data to begin with is asking for heartache later.

m2c
Click to expand...


lol i was gonna post "nice relational table skills" but thought id just lead him down the wrong path just to see what the abomination of a finished product would look like lols. fuuuuuuuuuuuuuuu
 
ditto on above ^^^

Will scale more effectively and make discrete analysis far easier down the line. Don't cram variables together and try to parse out after the fact. That is just poor data design.
 
flysarescary said:
lol i was gonna post "nice relational table skills" but thought id just lead him down the wrong path just to see what the abomination of a finished product would look like lols. fuuuuuuuuuuuuuuu
Click to expand...

ROFL, I actually considered doing that too, but I figured for all the stupid shit I post here I should contribute something worthwhile just to see what it felt like.
 
Garrett said:
can Prosper be rebuilt in a more server friendly manner? without MySQL?
Click to expand...

What would you use in place of MySQL? I personally use SQLServer, but I take it you are asking if you can do away with a relational database completely?

I wouldn't think you'd be able to build anything that would be able to scale once you start running a lot of traffic through it, and aggregating flat file or XML data into usable reports that you can make decisions on would be horribly inefficient and hard as hell to build. Relational databases were built for OLTP (online transactional processing) systems like those used in ad serving, it is far and away the most appropriate choice of data store.

If prosper doesn't give you what you want in an open source product and you dont want to use Stats Junky or Bevo or whatever because it's not self hosted you may have to breakdown and write your own. It's really not that difficult, you can get a bare bones db schema and some sort of listener/redirector written in an evening or two (I know, I've done it), and add to the interface as needed. And if you start with a properly normalized schema adding on features like split testing variables becomes pretty easy. Now if you need all the pretty graphs and stuff and don't want to be mucking around in the database to get your data in the beginning the level of effort to get a minimum amount of functionality goes up.
 
flysarescary said:
MySQL is faster than sqlite, postgresql and oracledb in my tests on several apps. Mysql isn't the problem.
Click to expand...

hmmmm... but databases and using PHP to reach into it is? What is the ideal programming language for a tracking script? I'm dying to know
 
You must log in or register to reply here.
Top Bottom