[REVIEW] RF: fix stream bug causing performance regressions

[venkywonka]

This nanoPR fixes performance regression caused due to improper stream assignments to the decision trees.

Before fix:

sno	algo	input	cu_time	cpu_time	cuml_acc	cpu_acc	speedup	n_samples	n_features	max_depth	n_estimators	n_bins	n_streams	n_jobs	n_classes
0	RandomForestClassifier	numpy	32.635321855545044	0.0	0.99468	0.0	0.0	800000	64	8	500	128	4	-1	2
1	RandomForestClassifier	numpy	40.36453413963318	0.0	0.994855	0.0	0.0	800000	64	10	500	128	4	-1	2
2	RandomForestClassifier	numpy	61.35148477554321	0.0	0.99504	0.0	0.0	800000	64	16	500	128	4	-1	2

After fix:

sno	algo	input	cu_time	cpu_time	cuml_acc	cpu_acc	speedup	n_samples	n_features	max_depth	n_estimators	n_bins	n_streams	n_jobs	n_classes
0	RandomForestClassifier	numpy	28.637776374816895	0.0	0.99468	0.0	0.0	800000	64	8	500	128	4	-1	2
1	RandomForestClassifier	numpy	34.11380743980408	0.0	0.994855	0.0	0.0	800000	64	10	500	128	4	-1	2
2	RandomForestClassifier	numpy	47.153409481048584	0.0	0.99504	0.0	0.0	800000	64	16	500	128	4	-1	2

Command run in cuml/

python python/cuml/run_benchmarks.py--num-rows 800000 --num-features 64 --skip-cpu --test-split 0.2 --cuml-param-sweep "n_bins=[128]" "n_streams=[4]" --cpu-param-sweep "n_jobs=[-1]" --param-sweep "max_depth=[8,10,16]" "n_estimators=[500]" --n-reps 1 --csv pool-2112-cls-800000.csv --dataset-param-sweep "n_classes=[2]" --dtype "fp32" --dataset classification -- RandomForestClassifier

[dantegd]

rerun tests

[dantegd]

@gpucibot merge

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (branch-22.04@48120ce). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-22.04    #4644   +/-   ##
===============================================
  Coverage                ?   83.86%           
===============================================
  Files                   ?      251           
  Lines                   ?    20293           
  Branches                ?        0           
===============================================
  Hits                    ?    17018           
  Misses                  ?     3275           
  Partials                ?        0           

Flag	Coverage Δ
dask	44.92% <0.00%> (?)
non-dask	77.02% <0.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out 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 48120ce...835c043. Read the comment docs.