Classification Rule Builder leading to undefined values

[farzaanh]

Hi all, I'm building a classification in rule builder using REGEX, I have my evar set up for classifications correctly under report suites > conversion classifications. When I test my rules using values that the evar has already collected, I'm seeing my rules classifying the values correctly with no issues. However it's now been 48 hours since first activating the classification, and I'm still only seeing undefined values in Adobe Analytics. Has anyone encountered this and has any tips for troubleshooting?

[Jennifer_Dungan]

Did you run a test on your ruleset with some real data to see if your rules are properly matching your content?

Unspecified usually means that content didn't match any rules, or its newer content that hasn't yet been processed (Classification rules run every 4-6 hours, after the initial backfill that is).

If it's been 48 hours, and nothing has yet shown up, and nothing new has been processed, I suspect that your rules aren't actually working...

I see that you have hidden a bunch of values for privacy, I get that... would you be willing to send me your Regex (something I can copy and paste), and a few samples of data to test?

In the meantime, the best thing you can do is get a table of your original values, copy a bunch of examples from there, and use the "Test Rule Set" to see how that information is processing.

[farzaanh]

Hey, thanks for the response. I ran a test on my ruleset using the 'Test Rule Set' function within Adobe, and everything classified correctly there. Yeah for sure, here is my Regex: ([^,:]+)[,:]([^,:]+)[,:]([^,:]+)[,:]([^,:]+)[,:]([^,:]+)[,:]([^,:]+)[,:]([^,:]+)[,:]([^,:]+)

And here are some test values:

AAAAEEEEEE,NA:DDDDCCCCCC,FFFAEEEEE:NAN,NAN:NAN,NAN
CNT1,CNT2:CNT3,CNT4:CNT5,CNT6:CNT7,CNT8

1NA,2NA:3NA,4NA:5NA,6NA:7NA,8NA

[farzaanh]

If it helps, this is what the Test Rule Set screen looks like when I tested my real values, it's showing 0 unmatched keys

[Jennifer_Dungan]

Interesting... the so real values are matching.... 

In my experience, I will sometimes use "fake values" for initial testing (or real values), but always test with "real values" during the testing phase.

So those look like they should be working... 

[Jennifer_Dungan]

Ok,

Based on these samples, yes it works, but are these real values or fabricated values? There may be nuances to real values that aren't being accounted for...

• Do you always have 8 items in this format? Is there a chance that only 7 or 6, etc items could be sent?
• Is the dimension you are using your rules on a list variable, or a prop set up to be processed like a list, that would prevent the actual result being processed in it's full format?
  • i.e. you might be sending "CNT1,CNT2:CNT3,CNT4:CNT5,CNT6:CNT7,CNT8" to s.list1.... but if list 1 is delimited by commas, then you will actually get:
    
    • CNT1
    • CNT2:CNT3
    • CNT4:CNT5
    • CNT6:CNT7
    • CNT8
• Is there a possibility that any of your parts result in "0"... Despite being a valid character, and usable in most Regex, Adobe for some reason cannot deal with 0 in classifications

[farzaanh]

Yeah we'll always have 8 items in this exact format, and the evar is not a list variable, just a standard evar. Good point on the possible 0 values, but we won't have any 0's in any of the values in each parameter. 

[Jennifer_Dungan]

Hmmm that is very interesting then.... it should work... I wonder if Adobe has issues with the "not in set"...

I've have a few instances (0s being one), where my test passed, but the actual data did not work....

I wonder if something like this would work better??

([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)[,:]([\w\d]+)

[farzaanh]

Thanks for putting that sample REGEX value together. I haven't tested it yet, because I wanted to leave my existing classification running throughout the weekend. Seems like my classifications are finally starting to yield values in AA, which may mean Adobe just needed a few days to process the classification. Slower than expected, but glad its starting to work.

[Jennifer_Dungan]

Fair enough... now that they seem to be working.. I've just never seen rules take that long to take effect before though.... 

[bjoern__koth]

Hi @farzaanh 

quick question: what do you see if you break down the Unspecified by the eVar you are trying to classify?

Are we talking about all eVar values not classified or only some?

If you don't mind, would you share some of those values with me (you can send them as DM)?

[farzaanh]

Hi @bjoern__koth 

When I breakdown the Unspecified by the eVar I was trying to classify, I see all values of that eVar that have fired. At the time, it was all eVar values not classified. Now that the classification has run over the weekend, I'm starting to see some of the classifications have values under them, so it's possible Adobe just needed extra time to process the classifications.

I provided some example values above to Jennifer, along with my REGEX, if that works!

Thanks!

[Jennifer_Dungan]

Question... what is the retention period of your eVar, and when is it set?

Is it possible that the values were set before your look-back? I saw "1 month lookback" in your screenshot... I have never seen it take longer than 24 hours to see classified data; unless maybe that information was set before your lookback?? I don't know if Classifications are made on the "actual values, or also on the attributed values.

I am glad you are starting to see classifications starting to come through though.. but it's still odd that it took so long.

[farzaanh]

Our eVar retention period is 2 years, and this particular eVar was only launched about 1 week ago. I was considering the look-back period too, but given the eVar just launched, the 1 month lookback period shouldn't have been an issue.

I've also never seen a classification take this long to process, so I do hope this doesn't become a norm. 

[Jennifer_Dungan]

OK, that's good to know.. yes, if this is new, then 1 month should be fine....  

One suggestion I have seen when it comes to Classification Rules, is to set up one "catchall" at the end, and set that value to something like "none" or "unknown"...

That way you can see what hasn't been processed at all (unspecified) and what has been processed, but wasn't caught by your rules (none / unknown / etc).

This might help diagnose what is happening in your rules... but you will have to re-process the lot again (add a new rule will deactivate the current rules, then activate and wait for all the rules to reprocess)

[farzaanh]

That's a good idea, thanks for the suggestion. I may have to add additional rules anyways, so will look to add that as well.

[DeepikaKa4]

very useful