Examples/fix knn

[ClaudiaComito]

Description

The knn demo stopped working at some unspecified time.

Here's a version that runs. However, it returns a different accuracy depending on the number of MPI processes I run it on:

(heat-dev3.8) Claudias-MBP:heat c.comito$ mpirun --tag-output --use-hwthread-cpus -n 1 python demo_knn.py 
[1,0]<stdout>:Accuracy: [0.9333333373069763]
(heat-dev3.8) Claudias-MBP:heat c.comito$ mpirun --tag-output --use-hwthread-cpus -n 2 python demo_knn.py 
[1,0]<stdout>:Accuracy: [0.5333333611488342]
[1,1]<stdout>:Accuracy: [0.5333333611488342]
(heat-dev3.8) Claudias-MBP:heat c.comito$ mpirun --tag-output --use-hwthread-cpus -n 3 python demo_knn.py 
[1,1]<stdout>:Accuracy: [0.9833333492279053]
[1,0]<stdout>:Accuracy: [0.9833333492279053]
[1,2]<stdout>:Accuracy: [0.9833333492279053]
(heat-dev3.8) Claudias-MBP:heat c.comito$ mpirun --tag-output --use-hwthread-cpus -n 4 python demo_knn.py 
[1,0]<stdout>:Accuracy: [0.7666666507720947]
[1,1]<stdout>:Accuracy: [0.7666666507720947]
[1,2]<stdout>:Accuracy: [0.7666666507720947]
[1,3]<stdout>:Accuracy: [0.7666666507720947]

~~Help would be great @mtar @coquelin77 @Inzlinger ~~

EDIT Feb 8.: The unstable accuracy is due to a getitem call within KNearestNeighbours.predict(), where the indices are distributed and not ordered, see #914. Since it's not a specific knn problem, this PR is ready for review.

Issue/s resolved: #

Changes proposed:

• update demo script

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Documentation update

Memory requirements

N.A.

Performance

N.A.

Due Diligence

• All split configurations tested
• Multiple dtypes tested in relevant functions
• Documentation updated (if needed)
• Updated changelog.md under the title "Pending Additions"

Does this change modify the behaviour of other functions? If so, which?

no

skip ci

[codecov]

Codecov Report

Merging #903 (6a0b2de) into master (c6ebf00) will decrease coverage by 4.42%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #903      +/-   ##
==========================================
- Coverage   95.49%   91.07%   -4.43%     
==========================================
  Files          64       64              
  Lines        9793     9793              
==========================================
- Hits         9352     8919     -433     
- Misses        441      874     +433     

Flag	Coverage Δ
gpu	?
unit	91.07% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
heat/optim/dp_optimizer.py	24.19% <0.00%> (-71.89%)	⬇️
heat/core/stride_tricks.py	81.53% <0.00%> (-12.31%)	⬇️
heat/core/devices.py	86.66% <0.00%> (-11.12%)	⬇️
heat/nn/data_parallel.py	84.13% <0.00%> (-10.35%)	⬇️
heat/cluster/spectral.py	85.71% <0.00%> (-8.58%)	⬇️
heat/core/communication.py	89.90% <0.00%> (-6.86%)	⬇️
heat/core/tests/test_suites/basic_test.py	91.26% <0.00%> (-4.86%)	⬇️
heat/core/linalg/qr.py	97.30% <0.00%> (-2.70%)	⬇️
heat/utils/data/partial_dataset.py	92.30% <0.00%> (-2.06%)	⬇️
heat/core/linalg/basics.py	94.22% <0.00%> (-1.31%)	⬇️
... and 7 more

---

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 c6ebf00...6a0b2de. Read the comment docs.

[coquelin77]

NOTE: the accuracy bug is still open, however this PR fixes the issue that the classifier in the example was not called previously (as noted in the first comment)

see #914