Speed up lea.max_of and lea.min_of functions

[Pierre Denis]

The current implementation of lea.max_of and lea.min_of could be slightly optimized, without using fast=True (hence without loosing referential consistency).

Currently in lea 3.0.1, the call

lea.max_of(a,b,c)

builds something similar to

Flea.build(max,(a,b,c))

To calculate the resulting probability distribution, this has to call max(va,vb,vc) for each combination of values (va,vb,vc).

Now, if we build instead

Flea2(max,Flea2(max,a,b),c)

then the calculation could be faster because max(va,vb) shall be calculated only once for each (va,vb) pair. Note that Flea2 has the same semantic as Flea, i.e. functional random variable, but is optimized for functions having two arguments.

Of course, the same optimization could be done also on lea.min_of.

[Pierre Denis]

Optimize implementation of lea.max_of and lea.min_of functions (refs #46)

→ <<cset 1478f7b8cc8a>>

[Pierre Denis]

Use iter(...) to browse arguments in lea.max_of and lea.min_of functions (refs #46)

→ <<cset 896e86bb0227>>

[Pierre Denis]

Add tests for lea.max_of and lea.min_of functions (refs #46)

→ <<cset 2124bb394c8b>>

[Pierre Denis]

Optimize implementation of lea.max_of and lea.min_of functions (refs #46)

→ <<cset ff0a1909a0d0>>

[Pierre Denis]

Use iter(...) to browse arguments in lea.max_of and lea.min_of functions (refs #46)

→ <<cset 368b275158de>>

[Pierre Denis]

Add tests for lea.max_of and lea.min_of functions (refs #46)

→ <<cset 40be7499e162>>