Home > General > Affiliate Marketing Forum

Data science in AM/PM - what data sources are out there? (5)

08-06-2018 09:50 AM #1 lootnicki (Member)
Data science in AM/PM - what data sources are out there?

Hi guys,

I have some data science skills including some basic statistical programming, analytics and machine learning skills. I believe I can use it in the AM/PM journey and want to prepare to it a little bit.

I think the main point here is to have raw data from the tracker (like .csv files including every user/hit row by row). Does it that mean that I should have self-served tracker to have it or I can get raw data from cloud tracker somehow like via FTP? Do someone have some experience on that?

What kind of other possible valuable data sources I should think of?

Thanks,
Lootnicki

08-07-2018 01:17 PM #2 theboss (Member)

Your ML skills are better used on the lander. Start with something simple like a boosting model to optimize a stylesheet.

For example:

conversion (target), stylesheet, domain, browser, etc.
0, a.css, somedomain, chrome
1, b.css, otherdomain, firefox

Use this to create a simple boosting model. You will be amazed by the results.

The problem with getting data at the tracker level is that once you build your model then what do you do with it?

Also, a good practice set for our industry is the Avazu competition on Kaggle.

https://www.kaggle.com/c/avazu-ctr-prediction

If you can become proficient in this you will be well beyond basic A/B, or even multivariate testing.

08-08-2018 12:33 PM #3 matuloo (Legendary Moderator)

Does it that mean that I should have self-served tracker to have it or I can get raw data from cloud tracker somehow like via FTP?
Trackers offer the options to export data in csv format mostly ... then you can play with it.

08-16-2018 02:17 PM #4 lootnicki (Member)

Quote Originally Posted by theboss View Post
Your ML skills are better used on the lander. Start with something simple like a boosting model to optimize a stylesheet.

For example:

conversion (target), stylesheet, domain, browser, etc.
0, a.css, somedomain, chrome
1, b.css, otherdomain, firefox

Use this to create a simple boosting model. You will be amazed by the results.

The problem with getting data at the tracker level is that once you build your model then what do you do with it?

Also, a good practice set for our industry is the Avazu competition on Kaggle.

https://www.kaggle.com/c/avazu-ctr-prediction

If you can become proficient in this you will be well beyond basic A/B, or even multivariate testing.
Hi theboss

Sorry for a late response, I got no notification from the forum

It's a good idea to do some modelling on landers to optimize them. I don't really know what data can I get from the tracker, but having an adserver data for every campaign event on cookie level we can learn some interesting stuff about the campaigns and optimize them (I mean some data like that: https://i.imgur.com/o9T3j2O.png)

Can you share some other ideas how you use ML to optimize anything? I see a big potential in ecomms for quite a scaled businesses

Quote Originally Posted by matuloo View Post
Trackers offer the options to export data in csv format mostly ... then you can play with it.
You don't mean a raw, cookie level data but the data we can see on the tracker dashboards right?

08-16-2018 02:32 PM #5 matuloo (Legendary Moderator)

You don't mean a raw, cookie level data but the data we can see on the tracker dashboards right?
Yup, the options to configure the CVS feeds are a bit limited unfortunately.

Home > General > Affiliate Marketing Forum