Allow repetitions for Choice parameters

[jrapin]

Types of changes

• Docs change / refactoring / dependency upgrade
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Motivation and Context / Related issue

This seems a common scheme that could be handled more efficiently to scale better.

How Has This Been Tested (if it applies)

Checklist

• The documentation is up-to-date with the changes I made.
• I have read the CONTRIBUTING document and completed the CLA (see CLA).
• All tests passed, and additional code has been covered with new tests.

[jrapin]

30 times faster

In [16]: %cpaste -q
import time
import nevergrad as ng

t0 = time.time()
param = ng.p.Tuple(*[ng.p.Choice(range(64)) for _ in range(100000)])
param.mutate()
v = param.value
print("v1", time.time() - t0)

t0 = time.time()
param = ng.p.Choice(range(64), repetitions=100000)
param.mutate()
v = param.value
print("v2", time.time() - t0)
--
v1 104.45756006240845
v2 3.4510011672973633

[danthe3rd]

30x faster

On how many CPUs? Can you also measure using a single CPU?

[jrapin]

30x faster

On how many CPUs? Can you also measure using a single CPU?

This is on my macbook pro, there is no parallelization apart from the one that may be in numpy, i think it's 1 CPU then

[teytaud]

Waow that one is really GREAT :-)