The source codes of the proposed algorithm in the paper and its competitors are provided while other auxiliary programs (like data generation, query generation) are provided in binaries for Windows.

It is easy to repeat the experiments as the authors provided some all-in-one scripts. The results in my run are more or less the same as the paper. The proposed histogram, STHistogram, is the best for more than 90\% of the cases and most of the measured relative errors are very close to the values shown in the paper. On the other hand, the performance of MinSkew is generally worse than that in the paper. In the worst case, fig 14(h) shows that MinSkew has a similar performance as STHistogram but it is, on average, 10 times worse than STHistogram in my run(s). It may be because of the new implementation of the method. The performance of the other two competitors is generally very close to the values in the paper.

For workability evaluation, I first tested the provided scripts for additional dataset / different parameters. Then, I tested running the experiment with my custom dataset. In the first test, the scripts work normally and give the results.

In the later test, I make a small 2D dataset with only 100 tuples. The file is named `R-D=2-my.data' where the prefix `R-D=?-' (? is the number of dimensions of the data) is necessary as the script generator will always add it in the generated scripts. To generate all the scripts required for the experiment, I just need to prepare a .cfg file (as shown below) and call `java -jar SimpleTest.jar my'. The steps are easy and I get the results for my custom test.

/**************
my.cfg
**************/

[dir]		real_data
[mode]		err
[method]	ST_HIST MIN_SKEW GEN_HIST RK_HIST
[pre]		mqtree

[num_queries]	1000

[dimension]	2
[real_data]	my

[type]		bucket
[bucket]
10 20 30
[query]
0.3