/wiki-know

Easy way to see results of old tests

Primary LanguageJavaScriptGNU Affero General Public License v3.0AGPL-3.0

#Welcome#

The 5th biggest site on the internet runs without ads, and without government support. That site is Wikipedia (hosted by the Wikimedia Foundation), and it survives on small-dollar donations! Those donations serve 500 million users, keep servers running, and power the entire organization.

Wikipedia and its small-donor funding model is something quite special. We use testing and experimentation to figure out the best way to explain the Wikimedia story to our users, and improve our fundraising efficiency. While doing so, we hope to cause minimum disruption to any single visitor. Testing and analysis is helping us reach that goal.

We've created this web app to showcase some of our significant tests and findings (and keep track of them internally as well!)

##Some Notes:

This code specifically refers to the presentation and distribution of the analysis we gained from tests we ran. Two sample tests (one with screenshots, one without) were specially scrubbed and included as sample data to show how the app might function. In production, this app includes thousands of test results.

You can see a live demo of this codebase (drawing on two sample datasets) here, hosted on heroku.

To see the code that powers the actual analysis, see the Wiki-Crunch codebase, which, while perhaps more technically interesting, is probably less useful to you, since it assumes the specific internal configuration of the WMF fundraising team.

In this version of wiki-know, you guess before seeing the answer.

##Setup: ####To run this project, first do: make then run a virtual envirnonment (virtualenv venv, source venv/bin/activate) pip install -r requirements.txt ####Then do: python hello.py ####Note: Sorting by "interesting handpicked tests" will give an error. This is because that data is currently private and can't be copied on this public repository. Sorry.

##Adding your own tests:

This app relies on data being in a standard format in directories.

Here are the assumptions:

  1. Each test is found in it's on directory in static/reports/
  2. In each test's directory, there must be at least the following files (formatted correctly):
    • meta.csv
    • screenshots.csv
  3. In each tests directory, there should be the following files (formatted correctly):
    • val_lookup.csv
    • description.txt
    • reportA.html
    • reportB.html
    • pamplona.png
  4. In each tests directory there can be as many png files as you want.

###A. How to format meta.csv: Currently, meta.csv has an inflexible format that doesn't leave much room for changes. This will change.

####Things that have no room for error:

  1. meta.csv must be a 2-line csv file.

  2. each row must have at least 10 fields

  3. row[4] and row[5] must point to the name of the winner and loser, respectively

    • those names must be the same as the names in val_lookup and screenshots.csv

  4. row[6] must have the naive percentage winning amount

  5. row[8] and row[9] must detail the outer and lower bound of confidence, respectively ####The model meta.csv has values that look like this:

    "test_id","var","country","language","winner","loser","bestguess","p","lowerbound","upperbound","totalimpressions","totaldonations" "1366565886","Banner.design","YY","yy","Key.phrase.in.blue","All.text.in.black",2.69,0.018,0.46,4.92,95076500,31750

###B. How to format screenshots.csv: ####Things that have no room for error:

  1. screenshots.csv's top line is assumed to be column names and is ignored
  2. each row must have at least 4 fields
  3. row[3] must be a direct URL to the relevant screenshot image
  4. row[1] must be the name of the value to be tested. This should be identical to "winner" or "loser" in meta.csv and 'value' in val_lookup.csv ####Assumptions:
  5. fields 5 and 6 are for alternate screenshots. (currently not used)
  6. There will be only 2 different values. (Multiples with the same name are fine).

####The model screenshots.csv might look like this: "test_id","value","campaign","screenshot","extra.screenshot.1","extra.screenshot.2","testname" "1366565886","Key.phrase.in.blue","C13_wpnd_enWW_FR","http://i.imgur.com/32wJUQD.png",NA,NA,"1366565886Banner.design" "1366565886","All.text.in.black","C13_wpnd_enWW_FR","http://i.imgur.com/QW0FDGU.png",NA,NA,"1366565886Banner.design" ###C. How to format val_lookup.csv

  1. First row is skipped
  2. Column A is the canonical internal name of the thing being tested. Should be the same as in screenshots.csv and as "winner" or "loser" in meta.csv
  3. Column B is the text you want to display to the user as the name of the banner being tested. ####The model val_lookup.csv looks like this: "value","description" "Key.phrase.in.blue","Key phrase in blue" "All.text.in.black","All text in black" ###D. How to format description.txt Write in plain text (html works too) the description of what's going on in the test. Users will see it. ###E. How to format reportA.html
  4. ReportA.html should contain a table, and only a table.
  5. It should have these classes: "table table-hover table-bordered"
  6. reportA should include interesting data about the test ###F. How to format reportB.html see report A ###G. What is pamplona.png? It is the main chart you use to show your confidence in the test over time. It gets pride-of-the-place treatment in the report.