Listing, counting factors of unique Mongo DB values over all keys - mongodb

Question

I'm preparing a descriptive "schema" (quelle horreur) for a MongoDB I've been working with.
I used the excellent variety.js to create a list of all keys and show coverage of each key. However, in cases where the values corresponding to the keys have a small set of values, I'd like to be able to list the entire set as "available values." In R, I'd be thinking of these as the "factors" for the categorical variable, ie, gender : ["M", "F"].
I know I could just use R + RMongo, query each variable, and basically do the same procedure I would to create a histogram, but I'd like to know the proper Mongo.query()/javascript/Map,Reduce way to approach this. I understand the db.collection.aggregate() functions are designed for exactly this.
Before asking this, I referenced:
http://docs.mongodb.org/manual/reference/aggregation/
http://docs.mongodb.org/manual/reference/method/db.collection.distinct/
How to query for distinct results in mongodb with python?
Get a list of all unique tags in mongodb
http://cookbook.mongodb.org/patterns/count_tags/
But can't quite get the pipeline order right. So, for example, if I have documents like these:
{_id : 1, "key1" : "value1", "key2": "value3"}
{_id : 2, "key1" : "value2", "key2": "value3"}
I'd like to return something like:
{"key1" : ["value1", "value2"]}
{"key2" : ["value3"]}
Or better, with counts:
{"key1" : ["value1" : 1, "value2" : 1]}
{"key2" : ["value3" : 2]}
I recognize one problem with doing this will be any values that have a wide range of different values---so, text fields, or continuous variables. Ideally, if there were more than x different possible values, it would be nice to truncate, say to no more than 20 unique values. If I find it's actually more, I'd query that variable directly.
Is this something like:
db.collection.aggregate(
{$limit: 20,
$group: {
_id: "$??varname",
count: {$sum: 1}
}})
First, how can I reference ??varname? for the name of each key?
I saw this link which had 95% of it:
Binning and tabulate (unique/count) in Mongo
with...
input data:
{ "_id" : 1, "age" : 22.34, "gender" : "f" }
{ "_id" : 2, "age" : 23.9, "gender" : "f" }
{ "_id" : 3, "age" : 27.4, "gender" : "f" }
{ "_id" : 4, "age" : 26.9, "gender" : "m" }
{ "_id" : 5, "age" : 26, "gender" : "m" }
This script:
db.collection.aggregate(
{$project: {gender:1}},
{$group: {
_id: "$gender",
count: {$sum: 1}
}})
Produces:
{"result" :
[
{"_id" : "m", "count" : 2},
{"_id" : "f", "count" : 3}
],
"ok" : 1
}
But what I don't understand is how could I do this generically for an unknown number/name of keys with a potentially large number of return values? This sample knows the key name is gender, and that the response set will be small (2 values).

Answer 1

If you already ran a script that outputs the names of all keys in the collection, you can generate your aggregation framework pipeline dynamically. What that means is either extending the variety.js type script or just writing your own.
Here is what it might look like in JS if passed an array called "keys" which has several non-"_id" named fields (I'm assuming top level fields and that you don't care about arrays, embedded documents, etc).
keys = ["key1", "key2"];
group = { "$group" : { "_id" : null } } ;
keys.forEach( function(f) {
group["$group"][f+"List"] = { "$addToSet" : "$" + f }; } );
db.collection.aggregate(group);
{
"result" : [
{
"_id" : null,
"key1List" : [
"value2",
"value1"
],
"key2List" : [
"value3"
]
}
],
"ok" : 1
}