Pandas groupby agg - how to get counts?

Question

I am trying to get sum, mean and count of a metric

df.groupby(['id', 'pushid']).agg({"sess_length": [ np.sum, np.mean, np.count]})

But I get "module 'numpy' has no attribute 'count'", and I have tried different ways of expressing the count function but can't get it to work. How do I just an aggregate record count together with the other metrics?

Constraint

Do you just want len? Not sure what you mean about different ways of expressing the count function - numpy certainly doesn't have np.count, as you've seen. What is this function expected to do? – Douglasdouglashome

you can use np.size – Simoneaux

@Simoneaux size will count nan as a row, count will exclude nan – Hammerhead

Answer 1

You can use strings instead of the functions, like so:

df = pd.DataFrame(
    {"id": list("ccdef"), "pushid": list("aabbc"), 
     "sess_length": [10, 20, 30, 40, 50]}
)

df.groupby(["id", "pushid"]).agg({"sess_length": ["sum", "mean", "count"]})

Which outputs:

           sess_length
                   sum mean count
 id pushid
 c  a               30   15     2
 d  b               30   30     1
 e  b               40   40     1
 f  c               50   50     1

Tiddlywinks

Answer 2

just use np.size

Not sure why the answer needs to be 30 chars long, when the answer is straightforward

Extreme

I was looking for that exactly. np.count_nonzero didn't look suitable. – Atropos

Answer 3

This might work:

df.groupby(['id', 'pushid']).agg({"sess_length": [ np.sum, np.mean, np.**size**]})

Funderburk

Is there a benefit to this syntax over the use of [ 'sum', 'mean', 'count'], as described in the accepted answer from last year? If so, it'd be useful to edit your answer to include that. – Dwyer

Answer 4

I think you mean :

df.groupby(['id', 'pushid']).agg({"sess_length": [ 'sum', 'count','mean']})

As mentioned in documentation of pandas, you can use string arguments like 'sum','count'. TBH It's more preferable way of doing these aggregations.

Dogcatcher