1 - Abstract

The bugs prioritization in open source repositories is considered an important and complex task. Mainly because, a lot of information about bugs changes over time and affects the prioritization process. Based on this dynamic characteristic, this work proposes a model to prioritize bugs as dynamic optimization problem. A preliminary empirical study was conduced comparing two dynamic evolutionary approaches and a static one. The achieved results demonstrated that a dynamic approach outperforms the static one in all evaluated scenarios.

Keywords: bugs prioritization, SBSE, evolutionary dynamic optimization.


2 - Instances

The instance used conatins data extracted of the Kate Editor Bugs Respository and is available below:

Instance Number of bugs(N) Extracted at Download
Dataset-1 543 April 9th, 2015 Download

The packages of changes generated by simulator in the dataset-1 by also are available to download.

Download Package of Change


3 - Empirical Study

The study evaluated level of change (low, medium and high), period of time (30, 60, 90) and rate of configuration (hypermutation and genetic propagate). The metric AREA, therefore, was created to identify which the better rate of configuration to a given level of change and period of time. The calcule is obtained by average of the best values of fitness found before each change, for all 30 runs.

The tables below shows the average and standard deviation of the AREA of each algorithms (GA-Static, GA-HyperMut and GA-GProp) in each scenario.

Average of the AREA - GA-Static with 0% of elitism and 5% of mutation.
30 60 90
low 1,012,778.23 2,006,125.83 3,032,840.69
medium 1,005,407.36 2,029,532.87 3,026,390.38
high 1,029,091.32 2,062,091.5 3,104,904.57
Standard Deviation of the AREA - GA-Static with 0% of elitism and 5% of mutation.
30 60 90
low 24,252.10 37,486.29 53,222.09
medium 23,341.12 53,145.81 66,640.93
high 30,606.66 57,683.25 83,017.32
Average of the AREA - GA-Hypermut.
30 60 90
0.3 0.6 0.9 0.3 0.6 0.9 0.3 0.6 0.9
low 957,639.37 904,999.31 898,094.99 2,063,173.74 2,004,010.18 2,003,525.18 3,176,014.13 3,167,268.01 3,120,031.26
medium 877,134.53 821,474.83 803,961.31 1,861,379.94 1,800,652.57 1,762,999.34 2,875,910.39 2,824,359.49 2,772,308.42
high 858,282.87 791,873.57 766,338.29 1,756,232.16 1,659,839.46 1,610,556.39 2,677,564.35 2,544,452.43 2,479,569.66
Standard Deviation of the AREA - GA-Hypermut.
30 60 90
0.3 0.6 0.9 0.3 0.6 0.9 0.3 0.6 0.9
low 31,328.40 31,833.94 33,500.51 68,469.45 59,999.74 68,745.36 95,828.07 98,957.15 91,580.13
medium 22,962.02 27,409.04 33,662.66 63,363.92 77,671.34 75,094.42 66,152.34 108,589.86 88,942.95
high 18,579.41 22,615.62 24,625.01 48,276.26 61,708.58 73,584.24 88,513 115,716.23 94,085.10
Average of the AREA - GA-GProp.
30 60 90
0.3 0.6 0.9 0.3 0.6 0.9 0.3 0.6 0.9
low 1,079,329.25 1,067,504.8 1,078,151.71 2,190,891.18 2,201,917.62 2,211,585.7 3,354,821.56 3,377,404.74 3,387,037.76
medium 1,046,152.78 1,045,530.79 1,042,046.74 2,129,243.61 2,117,994.22 2,113,143.16 3,182,264.87 3,197,500.04 3,217,737.49
high 1,041,479.41 1,044,387.43 1,041,001.76 2,091,892.26 2,080,868.68 2,081,610.21 3,128,231.22 3,154,354.45 3,145,899.82
Standard Deviation of the AREA - GA-GProp.
30 60 90
0.3 0.6 0.9 0.3 0.6 0.9 0.3 0.6 0.9
low 28,415.86 24,395.1 30,756.71 60,332.24 43,314.66 56,232.64 73,120.72 71,180.37 94,136.52
medium 24,566.54 25,126.42 26,856.24 48,295.05 54,641.17 58,771.83 83,510.23 77,025.45 89,127.54
high 30,556.85 29,532.71 29,333.67 57,896.65 49,271.61 46,503.13 90,024.45 77,759.63 61,924.72

The graphic below shows the average of the best value of fitness achieved before each change in each combination of period of change, level of change and rate of configuration. It's possible interate with graphic only selecting parameters, in order to view results alone.


4 - Source Code

The source code of the algorithms and simulator used in the empirical study is aviable to download following. Both implemented in Groovy.

Download