Re-implement Trebuchet percent strategy for better experiment analysis#6
Open
krisr wants to merge 4 commits intotemporarily-deprecate-percent-based-strategiesfrom
Open
Conversation
added 2 commits
July 3, 2013 11:27
The original Trebuchet percent strategy was not good for independent feature analysis because it tended to cluster percent launches for features to the same subsets of users. This new strategy is better because it computes a hash of the feature name along with the identifying user or visitor id and therefore distributes features randomly across their domains. This makes it more likely that the results of analysis for individual features are not biased by other features that were enabled with the percent strategy at the same time. Note that I exliminated the :to / :from options from the percent strategy since they are no longer as useful with the new implementation and seemed like interface bloat. Testing: Re-ran specs and implemented new specs for the new percent strategy
Contributor
There was a problem hiding this comment.
Our style guide allows for up to 100 characters on a line. I think this fits.
Contributor
Author
There was a problem hiding this comment.
Thanks, but I broke it up intentionally because I find it easier to read this way. Putting the whole thing on one line makes the comparison with percentage harder to see. If this goes against some other style guidelines I'm happy to change since I didn't feel too strongly and went back and forth.
added 2 commits
July 5, 2013 11:05
I skrewed this up in my previous commit by still referencing the no longer present @to instance variable.
This captures the goal of the spec more than the more exact comparison did.
Contributor
|
lgtm |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The original Trebuchet percent strategy was not good for independent feature analysis because it tended to cluster percent launches for features to the same subsets of users.
This new strategy is better because it computes a hash of the feature name along with the identifying user or visitor id and therefore distributes features randomly across their domains. This makes it more likely that the results of analysis for individual features are not biased by other features that were enabled with the percent strategy at the same time.
Note that I exliminated the :to / :from options from the percent strategy since they are no longer as useful with the new implementation and seemed like interface bloat.
Testing:
Re-ran specs and implemented new specs for the new percent strategy