DF variations in different runs

Dear @mostapha,

While preparing a simple DF example (with all the default HB Radiance parameters), I realised that repeating the simulation changes the DF values as high as 80% for specific points. I would have two questions on this:

  1. Am I right in assuming that this is due to a random component in the Radiance sampling procedures? If yes, do you have any reference in mind to suggest?

  2. Based on your experience, how can one make the Rad parameters more strict, so that such variations become negligible?

As always, many thanks for the great support.

Can your upload some data or pictures about your opinions?

@minggangyin, if you mean my GH defintion, here it is. Thanks for taking time.

DF_test.gh (538.2 KB)

@minggangyin, I figured out that increasing AD noticeably reduces the variations in different runs.

As stated in Radiance Primer, β€œThe error in the Monte Carlo calculation of indirect illuminate will be inversely proportional to the square root of this number.”


Yes. This is very common with Radiance. Here is one of the similar posts on the radiance mailing list which discusses the reasoning: https://radiance-online.org/pipermail/radiance-general/2010-January/006465.html

Parameters depends on your specific case. As you have figured out in this case -ad was helpful. If you have a complicated scene with small pieces then -aa and -ar will be critical. Also in many cases, you need to make sure number of bounces (-ab) is set correctly to address the needs of your case. This is a very good presentation about ambient calculation and ambient parameters in Radiance: https://www.radiance-online.org/community/workshops/2011-berkeley-ca/presentations/day1/JM_AmbientCalculation.pdf

Greg put this document together that we used for preset Radiance Parameters for Honeybee: SETTING RENDERING OPTIONS

Finally @MingboPeng has generated a parametric run that shows the effect of different parameters and put them together. See here: HB+ run time for imagebase Daylight coefficient - #8 by MingboPeng