Hi there,

I am completely new to pure data so forgive me if I'm missing something here but I need to calculate the variance of a large list of numbers, over 5000 infact. However if I use the variance object then it gives me the wrong answer, but if I run it again without closing puredata it gives me the correct answer. Why does it need to be run twice? Am I just missing something or is there an error in pure data?

In the event I'm not just missing something and I do need to run it twice how to I go about getting it to run twice after clicking the bang only once. Is there a loop of some sort I could use?

Also, how would I write just the one final value to a file, rather than the entire list of numbers.

Thanks very much for any help

You might want to upload you patch so we can see how you have it laid out.

For the last part of your question, If you are using PD Extended, I would look at all the

objects you have. e.g [list-lastx] --- [textfile]

Thanks for the reply here is the patch. I need it to save like 1 row of means and 1 row of variances to a file at the end. But as I say the variances only work after a second run through.

EDIT: what does the number after variance actually do? like [variance 10] or [variance 1000] say?

http://www.pdpatchrepo.info/hurleur/stats_generator.pd

your patch probably has errors due to undefined order. see how you have 2 cables coming out of that gui bang object that goes into the [openpanel] ? well, if the [clear( message receives a bang AFTER the file is set, then your file might be cleared from textfile as soon as it is loaded.

similarly, you have 2 cables coming out of the [rewind( message. if those cables send messages in the wrong order, that would be a likely culprit for the trouble you describe.

the solution to both of these issues is the [trigger] object. look up its help file, hopefully that will help.

Thanks for replying. So if I reverse the message order with a trigger that may fix my problem? I still don't understand what the number after variance means or how it can get the correct answer after 2 passes but not one.

The [variance] abstraction is a running variance, meaning you can send it a stream of numbers, and it updates the variance with each new input. The argument is how many of the last values it uses to calculate the variance. So if you have [variance 10], it will give you the variance of the last ten values you send it, and will toss out the older values.

oh right, well that makes sense i guess. Any ideas why it only gets the variance correct after 2 passes through though?

Are you sure your trigger is patched the right way ? Trigger outputs from the right to the left.

@jbm1991 said:

Any ideas why it only gets the variance correct after 2 passes through though?

After taking a look at how it works, it appears the problem is with the implementation. The mean it's using to determine the deviation is not constant the way it's set up, and it takes a second pass to make it constant.

I attempted to make a working version, but I haven't tested it extensively, yet. Take a look at it and see if it works for you. It works using the same idea--the argument is the length of the running variance. It also sets its initial state to all zero, meaning that if you set the argument to 4 and only send it two numbers, it will assume the other two are zero, so it won't really be accurate until you fill up the buffer. But I think it's right other than that.

http://www.pdpatchrepo.info/hurleur/variance.mmb.pd

Thanks for that, yeah I thought it was an error in the way it was implemented as I had done some tests on the outputs of various bits of the patch. That patch you have made is great, thanks