We will be performing a major infrastructure and web site ugprades on Friday, September 27th 8:00 am - 5:00 pm MST (UTC -7).

The website will be down during that time.

Stochastic analysis - 1,000 samples breaks regression, 999 does not

  • klucas@energy.state.md.us
  • Topic Author
More
20 Sep 2017 15:03 #5738 by klucas@energy.state.md.us
Hi,

When running a stochastic analysis with 999 samples, the process runs fine. When running with 1,000 samples, it produces the output but then 'stepwise.exe' crashes for each output variable during the regression step. The regression output is blank. Validated this on two different project files.

This is the error message for each output variable:

Error running stepwise regression for 'Annual energy': Could not open result.txt file for reading.
Error running stepwise regression for 'Net present value': Could not open result.txt file for reading.


Separately, the progress bars stop tracking the progress of the 'Calculating' step somewhere between 800 and 900 samples. At 800, they work properly, but at 900 they are stuck on 0% until they jump to 100%.

Thanks,
Kevin

Please Log in or Create an account to join the conversation.

  • pgilman
More
20 Sep 2017 20:58 #5739 by pgilman
Hi Kevin,

There may be an imposed limit on the number of samples -- I will have to investigate

What is the largest number of samples you would like to work with?

Best regards,
Paul.

Please Log in or Create an account to join the conversation.

  • klucas@energy.state.md.us
  • Topic Author
More
21 Sep 2017 16:27 #5740 by klucas@energy.state.md.us
Replied by klucas@energy.state.md.us on topic Stochastic analysis - 1,000 samples breaks regression, 999 does not
Probably less than 10,000. I have a number of different inputs (for instance, system size, load scaling factor, metering option, battery size, battery cost), so I want to be able to have good sets for each combination.

The core part of the model works fine, and I don't necessarily need the regression set. If it bypasses that part for large samples, that is fine, but it takes a while for the process to crash, for me to cancel, and for it to continue for each output variable. I'd rather just let it run and be done, even if i don't get the regression stats.

That said, the progress bars would be very useful to get a sense of how long it will take.

Thanks,
Kevin

Please Log in or Create an account to join the conversation.

  • pgilman
More
22 Sep 2017 00:24 #5741 by pgilman
Thanks Kevin,

I've added this issue and your other suggestion to show computed samples with results to our Issues list, which is now available to the public at:

github.com/NREL/SAM/issues/22

Best regards,
Paul.

Please Log in or Create an account to join the conversation.

Moderators: pgilman
Powered by Kunena Forum