The package RMixpanel provides an interface from R to Mixpanel’s API endpoints (see https://mixpanel.com/help/reference/data-export-api and https://mixpanel.com/help/reference/exporting-raw-data). For the most frequently used API endpoints (segmentation, retention, funnel, engage, export, etc.) custom methods make the parameterization more convenient and do the conversion from JSON to a corresponding R data.frame or R matrix. Furthermore it is possible to update or delete user profiles.
mixpanelGetData).segmentation/: get the segmentation matrix using mixpanelGetSegmentation.retention/: get the retention matrix using mixpanelGetRetention.addiction/: get the addiction matrix using mixpanelGetAddiction.funnel/: get funnel data using mixpanelGetFunnel.engage/:
mixpanelGetProfiles.mixpanelUpdateProfile.stream/query/: get events of selected people profiles using mixpanelGetEventsForProfiles.export/: get event data as R matrix using mixpanelGetEvents.jql/: perform custom queries using mixpanelJQLQuery.mixpanelGetProfilesCount.export/. This allows querying data for long time spans using multiple requests.or
The package depends on - jsonlite - uuid - RCurl - base64enc
In order to use the various methods of this package, we need to save the account data of the Mixpanel Project into an R object of class mixpanelAccount. The next examples all make use of this account object.
## Fill in here the API token, key and secret as found on
## www.mixpanel.com - Account -> Projects.
> account = mixpanelCreateAccount("ProjectName",
token="c12f3...",
secret="167e7e...",
key="553c55...")
> class(account)
[1] "mixpanelAccount"> retentions <- mixpanelGetRetention(account, born_event="AppInstall", event="WatchedItem",
from=20150701, to=20151101, unit="week")
> print(retentions)
## Example output:
## Retention Matrix
## Row names are Cohort Start Dates. Column names are Periods (0 -> 0 to 1 units)
## Count 0 1 2 3 ... ...
## 2015-06-29 17 94.11765 29.411765 29.411765 29.411765 ...
## 2015-07-06 38 100.00000 31.578947 18.421050 ...
...Given the people profiles have two properties named KPI1 and KPI2, the following lines of code will load these properties for all profiles matching the query KPI1 >= 1.32 and fill an R data.frame with the corresponding data. The hist method could be used to generate a histogram of one of the KPI’s.
More complex queries including logical operators and typecasts can be generated using the syntax described on Mixpanel’s documentation.
> profiles = mixpanelGetProfiles(account, where='properties["KPI1"] > 1.32',
select=c("KPI1", "KPI2"))
> print(profiles)
## Example output:
## distinct_id KPI1 KPI2
## [1,] "D1FED2..." 1.37 1.09
## [2,] "4441C5..." 2.11 -0.12
## ...
> hist(as.numeric(profiles[, "KPI1"]))Remove property KPI1 when the value is larger than 1000:
> profiles = mixpanelGetProfiles(account, where='properties["KPI1"] > 1000')
> distinctIDs = profiles[, "distinct_id"]
> for (distinctID in distinctIDs)
> mixpanelUpdateProfile(account, distinctID, data=list("$unset"="KPI1"))Delete all profiles where KPI1 is not set:
> profiles = mixpanelGetProfiles(account, where='not properties["KPI1"]')
> distinctIDs = profiles[, "distinct_id"]
> for (distinctID in distinctIDs)
> mixpanelUpdateProfile(account, distinctID, data=list("$delete"=""))Add a random value between 1 and 10 called bucket to all people profiles:
> profiles = mixpanelGetProfiles(account)
> distinctIDs = profiles[, "distinct_id"]
> for (distinctID in distinctIDs)
> mixpanelUpdateProfile(account, distinctID,
data=list("$set"=list(bucket=jsonlite::unbox(sample(10, 1)))))The general method mixpanelGetData allows to call all available API endpoints of the export API. However, the result is not parsed into R objects. Calling jsonlite::fromJSON(data) on the resulting data would do some parsing, but usually more postprocessing is needed.
Here an example without transforming the resulting JSON into handy R objects:
## Get list of funnels.
> mixpanelGetData(account, method="funnels/list/", args=list(), data=TRUE)
## Example output:
## [1] "[{\"funnel_id\": 1011888, \"name\": \"My first funnel\"},
## {\"funnel_id\": 1027999, \"name\": \"User journey funnel\"}]"
## Get data about a certain funnel.
> mixpanelGetData(account, method = "funnels/", args = list(funnel_id="1027999", unit="week"),
data = TRUE)
## Example output:
## [1] "{\"meta\": {\"dates\": [\"2015-11-04\", \"2015-11-11\"]},
## \"data\": {\"2015-11-11\":
## {\"steps\": [
## {\"count\": 7777, \"step_conv_ratio\": 1, \"goal\": \"AppInstall\", \"overall_conv_ratio\":1,
## \"avg_time\": null, \"event\": \"AppInstall\"},
## {\"count\": 555, \"avg_time\": 111, \"goal\": \"OpenedView\", \"overall_conv_ratio\": 0.77,
## \"selector\": \"(properties[\\\"status\\\"] == \\\"loggedin\\)\",
## \"step_conv_ratio\": 0.06964335860713283, \"event\": \"OpenedView\"},
## {\"count\": 333, \"avg_time\": 222, ...
## ...The JQL Query language opens a wide spectrum of possibilities. As a simple example we extract the event count per user (‘distinct_id’). The Mixpanel JQL API Reference can be found on https://mixpanel.com/help/reference/jql/api-reference.
jqlQuery <- '
function main() {
return Events({
from_date: "2016-01-01",
to_date: "2016-12-31"
})
.groupByUser(mixpanel.reducer.count())
}'
res <- mixpanelJQLQuery(account, jqlQuery,
columnNames=c("distinctID", "Count"), toNumeric=2)
hist(res$Count)Here we show how to calculate the metric Daily Active Users (DAU) when the user ID is different from the distinct_id. First write the JQL query and save it into a file named jqlDAU.js:
function today(addDays) {
var day = new Date();
day.setDate(day.getDate() + (addDays || 0));
return day.toISOString().substr(0, 10);
}
function main() {
return Events({
from_date: today(dayFrom),
to_date: today(dayTo)
})
.groupBy(["properties.UserID", getDay], function(count, events) {
count = count || 0;
return count + events.length;
})
.groupBy(["key.1"], mixpanel.reducer.count());
}The parameters