Floating point comparisons

[meilof]

Floating point comparisons

[meilof]

I had a look. Indeed this is not implemented for fixed point numbers. Thank you for reporting this, indeed I should either implement this or document that this is not supported. This is not hard to fix. If you need this functionality, let me know and I will prioritize it. Note to myself: the comparison operators are not defined for fixed-point numbers (https://github.com/meilof/pysnark/blob/master/pysnark/fixedpoint.py), so the implementation falls back to the default implementation for integers, but that does not work because addition of fixed-point numbers is different from addition of integers. Specifically, the computation of self+ LinComb.from_bits(bits)) in check_positive will not give the right result. This would be pretty easy to fix by overloading check_positive in LinCombFxp or by changing the behaviour of adding a LinComb to a LinCombFxp by not converting the LinComb. Op za 15 mei 2021 om 21:59 schreef Glenn Xavier ***@***.***>

When I try to compare fixed point values like in the following code: from pysnark.fixedpoint import PrivValFxp PrivValFxp(1.0) >= PrivValFxp(2.0) whenever the comparison evaluates to False I get an error: ValueError: constraint did not hold This seems to happen for all comparison operators other than == and != and only when the result is False. The comparisons work fine for regular integer PrivVals. Any ideas why this is happening? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#14>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACAJUU2SVVCX2XVKMKDWTQ3TN3HCBANCNFSM446GMA4A> .

-- Meilof -- ***@***.***

[gxavier38]

I do require this functionality. It would be great if you could implement it!

[meilof]

Just added this to pysnark master, seems to work:

>> from pysnark.fixedpoint import PubValFxp as PV >> PV(1)

{1.0}

>> PV(1)==PV(1)

{1}

>> PV(1)==PV(2)

{0}

>> PV(1)<PV(2)

{1}

>> PV(1)<2

{1}

>> PV(1)<0.5

{0}

>> PV(1)==0.5

{0}

>> PV(1)

{1.0}

>> PV(0.5)

{0.5}

>> PV(0.5)==0.5

{1}

…

On Mon, May 17, 2021 at 3:48 PM Glenn Xavier ***@***.***> wrote: I do require this functionality. It would be great if you could implement it! — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACAJUU4HMO6OWPLL7IKPIB3TOENC7ANCNFSM446GMA4A> .

-- Meilof -- ***@***.***

[gxavier38]

The comparisons work now, but removing the inheritance is causing some problems for me.
For example, floating points no longer have assert_eq.
Did you have an alternative planned?

[meilof]

You can of course still access these via the .lc member, this will work. But you're right, I suppose it would be good to see which operations work and add them to LinCombFxp. I do think I prefer the non inheritance approach. Op di 18 mei 2021 22:36 schreef Glenn Xavier ***@***.***>:

…

The comparisons work now, but removing the inheritance is causing some problems for me. For example, assertions on LinCombFxp objects no longer work. Did you have an alternative planned? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACAJUU4H2J4G3WV6QXG3HEDTOLFWBANCNFSM446GMA4A> .

[gxavier38]

It also looks like removing the inheritance also means losing a lot of floating point functionality:

PrivVal(0) == PrivValFxp(0)
AttributeError: 'LinCombFxp' object has no attribute 'check_zero'

if_then_else(PrivValFxp(1), PrivVal(1), PrivVal(0))
AttributeError: 'LinCombFxp' object has no attribute 'value'

PrivValFxp(1).check_positive()
AttributeError: 'LinCombFxp' object has no attribute 'check_positive'

[meilof]

I went through the LinComb operations and transferred the relevant once to LinCombFxp. Not all operations actually make sense, e.g., if_then_else(PrivValFxp(1), PrivVal(1), PrivVal(0)) should not be used like this since it requires a boolean condition as guard, e.g., PrivValFxp(1)==1 or >1 or similar, not a numeric value. So that is one of the operations that I have not implemented. But the meaningful operations that were easy to translate are there now.

…

On Wed, May 19, 2021 at 5:20 PM Glenn Xavier ***@***.***> wrote: It also looks like removing the inheritance also means losing a lot of floating point functionality: PrivVal(0) == PrivValFxp(0) AttributeError: 'LinCombFxp' object has no attribute 'check_zero' if_then_else(PrivValFxp(1), PrivVal(1), PrivVal(0)) AttributeError: 'LinCombFxp' object has no attribute 'value' PrivValFxp(1).check_positive() AttributeError: 'LinCombFxp' object has no attribute 'check_positive' — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACAJUUZ5INHEXC4TQEMIBS3TOPJLTANCNFSM446GMA4A> .

-- Meilof -- ***@***.***

[gxavier38]

Thank you!