Cloud Fx noise offset limit

[merlino]

Hi, I'm playing with the cloudfx and achieving some good result

But I've some problem animating the noise: it seems like there's a strange limit that depends on how big is the number. Yep, seems really strange but there's some example @F==1:

$T/25 = 0.0417084 (I'm using 23.976 fps, it rounds the value to 7 decimals)

-1+$T/25 = -0.998332 (rounds to 6 decimals)

-10+$T/25 = -9.99833 (rounds to 5 decimals)

-100+$T/25 = -99.9983  (rounds to 4 decimals) and so on ...

 

Any idea why it's happening?

[Jordan Walsh]

Probably just rounding for easier display

I put $T/$PI in a parm, and it displays as:

15.9818

If I evaluate it in python (hou.parmTuple('/obj/build/xform40/t').eval()) i get:

15.981808868811157

[merlino]

Thanks Jordan, but it seems it's considering the "rounded" value also for calculations: rendering, let's say, 25 frames depending on the initial number (-1 or -10 or -100) will end up with modifications just when that number changes. So rendering a -1500+$T/25 offset over the 25 frames you'll have just 3 real offset modifications. I've also tried evaluating that in phyton and that way I'll get the full number, no rounded ...

Edited March 13, 2014 by merlino

[Guest tar]

Looks like floating-point rounding. You have 6 values and a floating point

[merlino]

yep Martin, it's something like that, but the strange thing is that reading the value with python (as suggest Jordan) gives all the decimals, but inside the cloudnoise it's like it only reads 6 values including the integer part 

[Guest tar]

without checking it; is cloudfx processing at half and python at full, 16 vs 32 bit?

[merlino]

How can I check it? 

[Guest tar]

My second guess is it looks like Python is doing double precision and Houdini is Single - 32 vs 64bit.

 

Edit : always be aware that float is a logarithmic scale of precision. 

 

What Every Computer Scientist Should Know About Floating-Point Arithmetic

http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html

Edited March 13, 2014 by tar

[Guest tar]

How can I check it? 

 

Easy - check how many significants are visible for calculations

 

Try in Houdini 1.111111111/10000 and you'll see 0.000111111 , i.e. 6 significants, whereas in the Python window the answer is 0.00011111111110000001 i.e. way more than 6

 

http://en.wikipedia.org/wiki/Significand

Edited March 16, 2014 by tar

[merlino]

So, just learned something new!

And I suppose there's no simple workaround to this situation ... right?

 

EDIT:

 

right now my workaround is: set offset to 0 ... but I'd like more variety!

 

EDIT2:

 

Thanks for the usefull info!

Edited March 18, 2014 by merlino

[Guest tar]

The plot thickens!  Hidden down the bottom of the page for news for Houdini 12, this little nugget!

 

"Precision: Much of Houdini’s internal structure, including all parameter evaluation, has been converted to double-precision."

 

Edward has a good explanation of spreadsheets and precision. last post:

http://www.sidefx.com/index.php?option=com_forum&Itemid=172&page=viewtopic&t=31070

Edited March 18, 2014 by tar