min(A, k) algorithm

Author: Li Yiming

• Sort select: O(n log n)
• Linear select: 1.5n ~ 5.43n
• Lazy select: 2n + o(n)

See my report (in Chinese) for more technical information.

Run test cases and see the logs under log folder (you have to manually create it if not exist):

python test/test.py

The log file with WARNING is a simple version of log, and DEBUG is a more verbose version of log.

For example, the lazy-WARNING.log file has only the summary of the algorithm result and time used:

min([ 4  7  2 10  3  3], 0) = 2
Time used: 0.0seconds
min([ 4  7  2 10  3  3], 1) = 3
Time used: 0.0seconds
min([ 4  7  2 10  3  3], 2) = 3
Time used: 0.0seconds
min([ 4  7  2 10  3  3], 3) = 4
Time used: 0.0seconds
min([ 4  7  2 10  3  3], 4) = 7
Time used: 0.0seconds
min([ 4  7  2 10  3  3], 5) = 10
Time used: 0.0seconds

But the lazy-DEBUG.log file contains all the steps taken during the algorithm, it will be useful for you to understand the algorithm quickly.

Benchmark

You can run benchmark with command, the output will be written to files under log folder:

python test/benchmark.py

Zipf distribution seems more difficult to select, so I only tested the case when the length of A is 10,000.

Distribution	Algorithm	Length of A	Time used
Uniform	SORT-SELECT (numpy)	1,000,000	0.24885845184326172seconds
Uniform	LINEAR-SELECT	1,000,000	3.949763059616089seconds
Uniform	LAZY-SELECT	1,000,000	4.454483985900879seconds
Normalize	SORT-SELECT (numpy)	1,000,000	0.12593460083007812seconds
Normalize	LINEAR-SELECT	1,000,000	0.5047116279602051seconds
Normalize	LAZY-SELECT	1,000,000	2.378643274307251seconds
Zipf with ζ = 1.01	SORT-SELECT (numpy)	10,000	0.0010001659393310547seconds
Zipf with ζ = 1.01	LINEAR-SELECT	10,000	0.9494802951812744seconds
Zipf with ζ = 1.01	LAZY-SELECT	10,000	0.03298377990722656seconds