topic Re: how to sum several variables by group faster than proc means in SAS Procedures

how to sum several variables by group faster than proc means

asdf0990 — Sat, 15 Sep 2018 14:11:23 GMT

I am working with time use data. 3 Days have been tracked. I'd like to get the total time spent on those three days on each activity.

my data looks something like this:

1	60	0	0	3
1	50	0	0	3
1	0	0	0	3
2	10	30	0	2
2	10	30	0	2
2	0	0	5	2
3	0	0	5	4
3	0	0	5	4

I'd like a result something like this:

1	100	0	0	3
2	20	60	5	2
3	0	0	10	4

somewhere else on this board I found code working for my usecase, but it is very slow, also SAS Student edition warns me no to show the output.

proc means data = timeuse nway;
class id;
var activity1  activity5  activity6 activity761;
output out=timeuse_added sum =;
run;

Any suggestions on how to speed up this step? (to keep the quarters I managed to merge the old dataset by id)

Re: how to sum several variables by group faster than proc means

data_null__ — Sat, 15 Sep 2018 14:59:08 GMT

Turn off the printed output. Use PROC statement option NOPRINT or use PROC SUMMARY. Same mostly but no printed output.

Re: how to sum several variables by group faster than proc means

Reeza — Sat, 15 Sep 2018 15:31:01 GMT

How many observations do you have? Why do you need PROC MEANS to run faster, and what do you consider slow?

I made two changes to your code below. If you presort and switch to a BY instead of CLASS it may be faster if this isn’t fast enough.

@asdf0990 wrote:

I am working with time use data. 3 Days have been tracked. I'd like to get the total time spent on those three days on each activity.

my data looks something like this:

id| activity1 |activity5 |activity6 |activity761 |quarter

1 60 0 0 0 3

1
50 0 0 0 3

1 0 0 0 0 3

2 10 0 30 0 2

2 10 0 30 0 2

2 0 0 0 5 2

3 0 0 0 5 4

3 0 0 0 5 4

I'd like a result something like this:

id |activity1 |activity5| activity6 |activity761 |quarter

1 100 0 0 0 3

2 20 0 60 5 2

3 0 0 0 10 4

somewhere else on this board I found code working for my usecase, but it is very slow, also SAS Student edition warns me no to show the output.
proc means data = timeuse nway noprint;
class id;
var activity1  activity5  activity6 activity761;
output out=timeuse_added sum = / auto name autolabel;
run; out 
Any suggestions on how to speed up this step? (to keep the quarters I managed to merge the old dataset by id)

Re: how to sum several variables by group faster than proc means

asdf0990 — Sat, 15 Sep 2018 15:51:52 GMT

@Reeza wrote:
How many observations do you have? Why do you need PROC MEANS to run faster, and what do you consider slow?

I have 32105 rows and 705 variables in the dataset. I want to add up about 100 variable values. I consider 5 minutes for that step slow, given that I want to 20x the amount of calculations to be done.

thank you for the tips, I sorted allready, will look into the rest

Re: how to sum several variables by group faster than proc means

Reeza — Sat, 15 Sep 2018 17:59:32 GMT

Make sure to increase the cores available to SAS UE in the VirtualBox settings, to 2. You can also increase the RAM that would help, but to be honest, SAS UE is not designed for this type of work. It's for learning mostly. You could try using a full edition if you have access or using SAS AoD.

Re: how to sum several variables by group faster than proc means

asdf0990 — Sat, 22 Sep 2018 16:51:17 GMT

using noprint did increase the calculation speed substantially, all good now.