[#16] user based cross device tracking #17

gingerlime · 2017-03-19T15:17:34Z

adding an optional user_id attribute to experiments
when this attribute is present, variant selection will
be based on the user_id as part of the seed for a hash
function (along with experiment name and the type of caller)
tracking uuid for Gimel and Keen will also change to be
based on the user_id as seed, so the same user will only
get tracked once from different devices. Unless the goal
is not-unique.
updated tests and added adapter tests for Gimel and Keen.io

* adding an optional user_id attribute to experiments * when this attribute is present, variant selection will be based on the user_id as part of the seed for a hash function (along with experiment name and the type of caller) * tracking uuid for Gimel and Keen will also change to be based on the user_id as seed, so the same user will only get tracked once from different devices. Unless the goal is not-unique. * updated tests and added adapter tests for Gimel and Keen.io

joker-777

Looks in general very good. Only a few comments

joker-777 · 2017-03-20T17:50:45Z

CHANGES

+* 0.15.0-rc.1 BREAKING changes (only if you wrote your own tracking adapter)
+    - User-based / Cross-device tracking
+    - If you use your own tracking adapter, you'll have to change it.
+      * `experiment_start` now accepts the `experiment` as the first parameter


I would use the word 'expects' since you really have to pass this parameter. 'accepts' sounds quite optional.

Yes, great suggestion!

joker-777 · 2017-03-20T17:51:36Z

README.md

        },
-        goal_complete: function(experiment_name, variant, event_name) {
-            keen_client.addEvent(experiment_name, {variant: variant, event: event_name});
+        goal_complete: function(experiment, variant, event_name, _props) {


why _props when you don't use it?

by convention variables that are passed over but not used are marked with _ prefix. I think it's useful to know that it's being passed over from the experiment, even if it's not used.

joker-777 · 2017-03-20T18:02:04Z

src/adapters.coffee

-        callback = @_remove_uuid(item.properties.uuid)
-        @_ajax_get(@url, item.properties, callback)
+        callback = @_remove_quuid(item.properties._quuid)
+        @_ajax_get(@url, utils.omit(item.properties, '_quuid'), callback)


Why do you have to remove the _quuid here?

there's no need to send it over the wire.

joker-777 · 2017-03-20T19:57:57Z

src/adapters.coffee

-    @experiment_start: (experiment_name, variant) =>
-      @_track(@namespace, "#{experiment_name} | #{variant}", 'Visitors')
+    @experiment_start: (experiment, variant) =>
+      @_track(@namespace, "#{experiment.name} | #{variant}", 'Visitors')


the third parameter should be an object, no?

not in this tracking adapter... the tracking adapters don't share any code and can implement internally differently.

In general I think you should try to keep the interfaces of the adapters similar.

The interface is the same. This is internal implementation.

joker-777 · 2017-03-20T19:59:12Z

src/adapters.coffee


-    @goal_complete: (experiment_name, variant, goal) =>
-      @_track(@namespace, "#{experiment_name} | #{variant}", goal)
+    @goal_complete: (experiment, variant, goal, _props) =>


In PersistentQueueKeenAdapter the third parameter is the goal_name and here it is just goal. I think you should stay consistent with it.

Thanks for spotting. Will change this.

joker-777 · 2017-03-20T20:24:48Z

src/alephbet.coffee

+    _random: (salt) ->
+      return utils.random() unless @user_id
+      seed = "#{@name}.#{salt}.#{@user_id}"
+      utils.random(seed)


Why do you need this seed for the user_id? Can't you just always use utils.random("#{@name}.#{salt}.#{@user_id}")?

Are you asking why I'm assigning it to a variable? not strictly necessary. I can pass it directly as you suggested. But might be a bit more readable to understand what this string is used for?

No that wasn't my question. My question is why you can't use utils.random("#{@name}.#{salt}.#{@user_id}") even when there is no @user_id? Simply writing

_random: -> utils.random("#{@name}.#{@user_id}")

That won't be random though...

Can you quickly explain how does it work with the salt?

Take a look at the implementation. It's super small and simple. It uses a hash function with the seed and generates a random number based on the seed. If the seed is different for different users, the result hash will produce a different hash, which would get "translated" to a different number between 0 and 1. So for a given experiment, salt and user_id, we'll get a pseudo-random number. If you send the same user_id (or null), then you'll always get the same result, which won't be (pseudo)random...

When we don't have a user_id, using a "real" random number is preferable.

#16 has a link to planout which might explain this in more detail.

joker-777 · 2017-04-03T16:44:47Z

@gingerlime Somehow I missed it. Why do I need to specifically add the user_id check to the goals and also to the triggers. Can't you do it internally? Is there a use case where you add the user_id but want to trigger an experiment even if it doesn't exist?

gingerlime · 2017-04-03T17:49:19Z

@joker-777 I wrote about it on this wiki page https://github.com/Alephbet/alephbet/wiki/User-based-and-Cross-device-tracking

Is there a use case where you add the user_id but want to trigger an experiment even if it doesn't exist?

I don't think you can mix not-logged-in and logged-in on the same experiment, without getting unreliable results unfortunately. I tried to cover a couple of cases where this might cause inconsistent results on the wiki page.

Why do I need to specifically add the user_id check to the goals and also to the triggers. Can't you do it internally?

It might be possible to do this internally, but this definitely complicates things. You'd need to store/cache the user_id. You'd need to know when the user_id is missing, vs when it wasn't meant to be used at all... It's probably a good idea, but need to think carefully and design it to work seamlessly without creating unexpected behaviour. For now, I think it's ok to expect the caller to make sure the user_id is available, and it's not too difficult to cache it externally if you want to "remember" who the user was initially.

joker-777 · 2017-04-03T23:19:12Z

The only way to achieve this is to add another attribute, in addition to user_id. Something like cross_device: true. I don't think it is to complicated to add some small checks internally. In fact I think you only need to add return if @cross_device && !@user_id to the method which calls the trigger callback and to add_goal and add_goals methods. I didn't dig into the code though.

gingerlime · 2017-04-04T05:04:19Z

two attributes for one purpose makes the Alephbet API ugly. If the intention is to make it transparent, then this should be handled internally with only the user_id attribute. It does make things a bit more complicated however.

Yoav added 2 commits March 19, 2017 15:48

[#16] bump to 0.15.0-rc.1

da31fa6

joker-777 requested changes Mar 20, 2017

View reviewed changes

[#16] applied feedback comments + updated README

caac4ef

joker-777 approved these changes Mar 21, 2017

View reviewed changes

gingerlime merged commit b70fe7e into master Apr 4, 2017

gingerlime deleted the 16-user-based-cross-device-tracking branch April 4, 2017 11:24

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[#16] user based cross device tracking #17

[#16] user based cross device tracking #17

gingerlime commented Mar 19, 2017

joker-777 left a comment

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 21, 2017

gingerlime Mar 21, 2017

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 20, 2017

gingerlime Mar 21, 2017

joker-777 Mar 21, 2017

gingerlime Mar 21, 2017

joker-777 Mar 21, 2017

gingerlime Mar 21, 2017

gingerlime Mar 21, 2017

gingerlime Mar 21, 2017

joker-777 Mar 21, 2017

joker-777 commented Apr 3, 2017

gingerlime commented Apr 3, 2017

joker-777 commented Apr 3, 2017

gingerlime commented Apr 4, 2017

[#16] user based cross device tracking #17

[#16] user based cross device tracking #17

Conversation

gingerlime commented Mar 19, 2017

joker-777 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

joker-777 commented Apr 3, 2017

gingerlime commented Apr 3, 2017

joker-777 commented Apr 3, 2017

gingerlime commented Apr 4, 2017