[connect] Optimize the connector about ScaleFreeBA, ScaleFreeBADual, PowerLaw

[Routhleck]

Description

Add new test case for ProbDist in test_random_conn.py

Create test_all_time.py to test the efficiency of all connectors and the results show that the following four connectors need to be optimized: ScaleFreeBA, ScaleFreeBADual, PowerLaw and ProbDist.

Optimize ScaleFreeBA, ScaleFreeBADual, PowerLaw's build_conn function by by preallocating repeated_nodes with array of numpy and Rename their build_conn to build_mat, also their returned values are changed accordingly

Try to optimize ProbDist, but if parallelization is performed, the results will be inconsistent which need to be solved in the future

Create test_optimized_result.py to compare the results and time needed between before and after optimized for GaussianProb, ScaleFreeBA, ScaleFreeBADual and PowerLaw

How Has This Been Tested

Test device

Hardware

• CPU: Intel Core i5-12490F 6C12T

• GPU: Nvidia Geforce RTX 2070

• RAM: 32G DDR4 3200MHz

Software

• OS: Windows10 20H2

• Python: 3.10

Below is the test result

<style> </style>

connector name	connect matrix size	build function	other parameter	time origin(ms)	time optimized(ms)	ratio	remark
GaussianProb	100x100	build_mat	sigma=1 / include_self=False	2.00	1.00	1.00
GaussianProb	500x500	build_mat	sigma=1 / include_self=False	16.00	5.00	2.20
GaussianProb	2500x2500	build_mat	sigma=1 / include_self=False	382.98	86.79	3.41
GaussianProb	12500x12500	build_mat	sigma=1 / include_self=False	6861.67	1682.14	3.08
GaussianProb	100x100	build_mat	sigma=4	2.00	1.00	1.00
GaussianProb	500x500	build_mat	sigma=4	14.00	5.00	1.80
GaussianProb	2500x2500	build_mat	sigma=4	286.29	77.00	2.72
GaussianProb	12500x12500	build_mat	sigma=4	6748.89	1350.49	4.00
GaussianProb	100x100	build_mat	sigma=4 / periodic_boundary=True	2.00	1.00	1.00
GaussianProb	500x500	build_mat	sigma=4 / periodic_boundary=True	16.00	4.00	3.00
GaussianProb	2500x2500	build_mat	sigma=4 / periodic_boundary=True	302.53	72.00	3.20
GaussianProb	12500x12500	build_mat	sigma=4 / periodic_boundary=True	7276.54	1386.13	4.25
GaussianProb	100x100	build_mat	sigma=4 / periodic_boundary=True	2.00	0.99	1.01
GaussianProb	500x500	build_mat	sigma=4 / periodic_boundary=True	16.00	5.00	2.20
GaussianProb	2500x2500	build_mat	sigma=4 / periodic_boundary=True	313.53	71.00	3.42
GaussianProb	12500x12500	build_mat	sigma=4 / periodic_boundary=True	7243.03	1391.71	4.20
ScaleFreeBA	100x100	build_mat	m=2	2.00	1.00	1.00
ScaleFreeBA	500x500	build_mat	m=2	21.00	4.00	4.25
ScaleFreeBA	2500x2500	build_mat	m=2	462.00	18.00	24.67
ScaleFreeBA	12500x12500	build_mat	m=2	11013.49	113.00	96.46
ScaleFreeBADual	100x100	build_mat	m1=2 / m2=3 / p=0.4	2.00	1.00	1.00
ScaleFreeBADual	500x500	build_mat	m1=2 / m2=3 / p=0.4	29.00	4.00	6.25
ScaleFreeBADual	2500x2500	build_mat	m1=2 / m2=3 / p=0.4	618.25	28.00	21.08
ScaleFreeBADual	12500x12500	build_mat	m1=2 / m2=3 / p=0.4	13944.25	132.00	104.64
PowerLaw	100x100	build_mat	m=3 / p=0.4	2.00	2.00	0.00
PowerLaw	500x500	build_mat	m=3 / p=0.4	38.00	7.00	4.43
PowerLaw	2500x2500	build_mat	m=3 / p=0.4	820.53	39.00	20.04
PowerLaw	12500x12500	build_mat	m=3 / p=0.4	18474.92	282.06	64.50
ProbDist	{1000}	build_coo	dist=1 / prob=0.5 / pre_ratio=0.3 / include_self=True	29.00	17.00	0.71	unparallel, result same
ProbDist	{(100, 20)}	build_coo	dist=1 / prob=0.5 / pre_ratio=0.3 / include_self=True	114.53	118.00	-0.03	unparallel, result same
ProbDist	{(4, 20, 20)}	build_coo	dist=1 / prob=0.5 / pre_ratio=0.3 / include_self=True	75.00	77.00	-0.03	unparallel, result same
ProbDist	{(4, 3, 8, 5)}	build_coo	dist=1 / prob=0.5 / pre_ratio=0.3 / include_self=True	8.00	10.00	-0.20	unparallel, result same
ProbDist	{10000}	build_coo	dist=1 / prob=0.5 / pre_ratio=0.3 / include_self=True	2745.23	461.59	4.95	parallel, result diifferent

[chaoming0625]

Great! Excellent works.