stedn/mixpandas

Get Mixpanel event data to Pandas for custom analysis

mixpandas

Get data from Mixpanel's Raw Data Export API into a Pandas DataFrame for powerful custom analysis. Use read_events to query your data. Specify event names, date ranges or other filters to control selection. Consult the documentation for detailed use.

Mixpandas wraps Mixpanel's API client with a few modifications documented in the code.

Example

An app tracks a submit rating event with a stars property indicating the rating. Find each user's average rating over a specified time span.

>>> keys = (MIXPANEL_API_KEY, MIXPANEL_API_SECRET)
>>> df = mixpandas.read_events(keys, events='submit rating', start='9/8/2013', end='9/15/2013')
>>> df.groupby('distinct_id')['stars'].mean()
distinct_id
10             3.00000
12             4.50000
15             4.75000
2              4.00000
23             2.66667
27             5.00000
...

The start and end dates are parsed with pandas.to_datetime and converted to the proper Mixpanel format. This allows you to think less about how to specify the dates and just get the data you want. The following are just a few ways to do the same thing:

>>> df = mixpandas.read_events(keys, events='submit rating', start='9/8/2013')
>>> df = mixpandas.read_events(keys, events='submit rating', start='2013-9-8')
>>> df = mixpandas.read_events(keys, events='submit rating', start='Sept 8, 2013')

Mixpanel includes the time as a property of each tracked event. Mixpandas converts this time property to Pandas Timestamp objects enabling powerful slicing and other time series analysis.

>>> df.set_index('time', inplace=True, drop=False)

# Grab arbitrary time spans of the data
>>> df['9/12/2013':'2013-09-15']  # Dates can be specified in many ways

# Compute the 5 day window moving average of number of ratings submissions
>>> counts = df.groupby(df['time'].map(lambda x: x.date())).size()  # Group data by day, get counts
>>> counts.index = pd.DatetimeIndex(counts.index)  # Get back the DatetimeIndex
>>> counts = counts.resample('D')  # Re-sample every day in case some days have none
>>> pd.rolling_mean(counts, 5)  # Compute the 5 day moving average