Features/54 distributed random numbers

[TheSlimvReal]

Description

Target issue #54.

This builds on @Markus-Goetz implementation of the threefry64 algorithm. The __counter_sequenze function was completed which returns initial values for the "random" number calculation. It is currently only tested for torch.int64 because the threefry32 is not implemented yet.

• New feature (non-breaking change which adds functionality)

[codecov]

Codecov Report

Merging #362 into master will increase coverage by 0.09%.
The diff coverage is 98.88%.

@@            Coverage Diff             @@
##           master     #362      +/-   ##
==========================================
+ Coverage   97.14%   97.23%   +0.09%     
==========================================
  Files          53       53              
  Lines        9659    10093     +434     
==========================================
+ Hits         9383     9814     +431     
- Misses        276      279       +3

Impacted Files	Coverage Δ
heat/core/manipulations.py	98.11% <ø> (ø)	⬆️
heat/core/tests/test_statistics.py	100% <100%> (ø)	⬆️
heat/core/random.py	100% <100%> (+10%)	⬆️
heat/core/operations.py	95.2% <100%> (ø)	⬆️
heat/core/tests/test_manipulations.py	100% <100%> (ø)	⬆️
heat/core/tests/test_communication.py	97.95% <100%> (ø)	⬆️
heat/ml/cluster/kmeans.py	100% <100%> (ø)	⬆️
heat/core/tests/test_random.py	100% <100%> (ø)	⬆️
heat/core/statistics.py	95.53% <86.36%> (-1.13%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8e6d44c...e774c95. Read the comment docs.

[Markus-Goetz]

matplotlib just simply needs to go still

[Markus-Goetz]

What is the state regarding the 32-bit Threefry? Have you tried the upper bit removal?

[TheSlimvReal]

What is the state regarding the 32-bit Threefry? Have you tried the upper bit removal?

I tried to use the 64-Bit logic for the random numbers and cut off the upper 32-Bit but this resulted in a similar probability for equal values compared to the threefry-32 algorithm. (Around 10 per 10k values)

[Markus-Goetz]

Right, so let's do the following, lets use the regular threefry32 and just live with the fact that there is repetition. As for your unit tests, just set a slightly higher upper bound for the number of allowed repetitions than 1, e.g. 10?