mixpanel-jql

A small Python library for running JQL queries against Mixpanel's JQL API. The data returned from the API is automatically decompressed as it arrives, making it available for processing as soon as the first row arrives. This is to avoid buffering large result sets in memory.

Installation

To install the mixpanel-jql library, simply run the following in your terminal:

pip install mixpanel-jql

Simple example

Let's do a simple count of our number of 'X' events over each day of May 2016. Our key for grouping will be the date the event was sent to Mixpanel in the format YYYY-MM-DD. We can get that from our event's time property by specifying our key as new Date(e.time).toISOString().split('T')[0].

This is simple and fast to do with this library.

from mixpanel_jql import JQL, Reducer

api_secret = '...'

params = {
    'event_selectors': [{'event': "X"}],
    'from_date': '2016-05-01',
    'to_date': '2016-05-31'
}

query = JQL(api_secret, params)\
          .group_by(
              keys=[
                  "new Date(e.time).toISOString().split('T')[0]",
              ],
              accumulator=Reducer.count())

for row in query.send():
    date = row['key'][0]
    value = row['value']
    print("[%s] => %d" % (date, value))
# [2016-05-01] => 302
# [2016-05-02] => 1102
# ...
# [2016-05-31] => 120

But what if we only want to count unique events? That is to say, what if we care about how many users spawned each event per day and not just the overall number of times the event occurred?

With some minor modification to our previous code, we can achieve this:

query = JQL(api_secret, params)\
          .group_by_user(
              keys=[
                  "new Date(e.time).toISOString().split('T')[0]",
              ],
              accumulator="function(){ return 1;}")\
          .group_by(
              keys=["e.key.slice(1)"],
              accumulator=Reducer.count())

We replace our accumulator keyward argument with a JavaScript function returning 1, since each user will only be counted for once. group_by_user also adds the user ID into the key of our results. We can regroup our results by slicing that detail off with e.key.slice(1) and recounting.

More advanced examples

Let's assume we want to count all events 'A' with a property 'B' that is equal to 2 and a property F that is equal to "hello". Events 'A' also have a property 'C', which is some random string value. We want the results grouped and tallied by values of 'C' to see how many property 'C' events occurred over each day in the month of April 2016.

from mixpanel_jql import JQL, Reducer

api_secret = '...'

params = {
    'event_selectors': [{'event': "A"}],
    'from_date': '2016-04-01',
    'to_date': '2016-04-30'
}

query = JQL(api_secret, params)\
          .filter('e.property.B == 2')\
          .filter('e.property.F == "hello"')\
          .group_by(
              keys=[
                  "new Date(e.time).toISOString().split('T')[0]",
                  "e.property.C"
              ],
              accumulator=Reducer.count())

for row in query.send():
    date, c = row['key']
    value = row['value']
    print("[%s] %s => %d" % (date, c, value))
# [2016-04-01] abc => 3
# [2016-04-01] xyz => 1
# ...

If we wanted to count only unique events (i.e. count each user causing the event only once), we can change our query to group by user, to reduce the number of times they caused a particular e.properties.C to just 1.

query = JQL(api_secret, params)\
          .filter('e.property.B == 2')\
          .filter('e.property.F == "hello"')\
          .group_by_user(
              keys=[
                  "new Date(e.time).toISOString().split('T')[0]",
                  "e.property.C"
              ],
              accumulator="function(){ return 1;}")\
          .group_by(
              keys=["e.key.slice(1)"],
              accumulator=Reducer.count())

Why are your filters not joined with `&&`?

We could have also combined our .filter(...) methods into 1 method by doing, .filter('e.property.B == 2 && e.property.F == "hello"'). Successive .filter(...) expressions are automatically &&'ed. The method of expression you choose is stylistic.

What is that `Reducer` thing?

The Reducer class is for convenience and contains shortcuts to all the reducer functions (e.g. Reducer.count() returns mixpanel.reducer.count(), and Reducer.top(limit) returns mixpanel.reducer.top(limit)). Refer to the code for a list of all reducer shortcuts.

To write your own reducer, make sure to include a full JavaScript function body (i.e. function(){ ... }).

How do I see what the final JavaScript sent to Mixpanel will be?

Use the .query_plan() method on your JQL query to view what the equivalent JavaScript will be.

>>> query.query_plan()
'function main() { return Events(params).filter(function(e){return e.property.B == 2}).filter(function(e){return e.property.F == "hello"}).groupByUser([function(e){return new Date(e.time)).toISOString().split(\'T\')[0]},function(e){return e.property.C}], function(){ return 1;}).groupBy([function(e){return e.key.slice(1)}], mixpanel.reducer.count()); }'

This can be quite helpful during debugging.

Caveats

.filter(...) automatically transforms whatever is within the parenthesis' into function(e){ return ... }. This library does not support the properties.x shortcut syntax and requires e.properties.x.

This library cannot easily express everything possible in Mixpanel's JQL language, but does try to simplify the general cases. If you have some ideas for making this library more user friendly to a wider range of potential queries, please submit a pull request or create an issue.

Contributions are very welcome!

Where can I learn more about Mixpanel's JQL?

For more information on what you can do with JQL, refer to Mixpanel's documentation here.

caux/mixpanel-jql