'setprecision', 24

Why are the two results below the same? I would like to simulate single precision with system_dependent('setprecision' , 24).

>> system_dependent('setprecision', 64)

>> 10000*sqrt(exp(pi))-48104

ans =

0.773809653510398

>> system_dependent('setprecision', 24)

>> 10000*sqrt(exp(pi))-48104

ans =

0.773809653510398

Answer 1

[Walter Roberson]

system_dependent('setprecision')... I seem to recall that it only ever had effect on windows and only for some of the older releases. You could find more information on Yair's undocumented-matlab site.

[Beat Aebischer]

Thanks. This means one cannot rely on 'system_depedent' or 'feature'. Anyway in the ML version I currently use (9.5.0), 'setprecision' does not seem to work - it is also asking a lot as it would have to be "system independent".

Instead I have tried casting the input with the command 'single', this seems to work.

Answer 2

[Gaurav Garg]

Hi,

On running the same code, and then verifying the variable precision being used, I could see that the variable precision was never changed.

You can verify the variable precision using the digits.

digits

In your case, if you need to change the precision, you can use the below piece of code -

digitsOld = digits(64);
vpa(10000*sqrt(exp(pi))-48104)
digits
digitsOld = digits(24);
vpa(10000*sqrt(exp(pi))-48104)
digits

You can also follow the documentation here.

[Walter Roberson]

digits only applies to the symbolic toolbox, never to floating-point work.

[Beat Aebischer]

Yes, Gaurav's suggestion is very nice, but I was trying to do it without a Symbolic Math TB.

[Walter Roberson]

The solution with vpa needs modification to prevent the argument from being evaluated in double precision before it gets to vpa.

Q = @(v) sym(v) ;
digitsOld = digits(64);
vpa(10000*sqrt(exp(Q(pi)))-48104)
digits
digitsOld = digits(24);
vpa(10000*sqrt(exp(Q(pi)))-48104)

It would need further modifications to emulate step by step evaluation at the chosen digits the way that setting system_dependent('precision') would