This repo is for testing three methods of distributing bloom filter size.

test.sh is a generic script that will：

1. compile the testing cpp,
2. run experiments for given dataset,
3. generate plots.

The dataset should be given by following forms:

dataset_directory
|   scale1x
|   |   test1
|   |   |   Z0.0_ZD0_query_stats.txt
|   |   |   Z0.0_ZD1_query_stats.txt 
|   |   test2
|   |   test3
|   scale2x
|   scale4x
|   ...
|   scale32x

For each testcase (query_stats), it should follow following format:

#entries(N) bpk(we will not use this) #files
fileID#1 #existing_queries #non-existing_queries
fileID#2 #existing_queries #non-existing_queries
...
fileID#N #existing_queries #non-existing_queries

---

test.py will scan every file with query_stats under directory, recording scale#x, #test, Z#_ZD#, and output result to output/stats.csv, where Z refer to portion of zero queries and ZD implies the distribution of the queries (0 for uniform, 1 for normal and 3 for zipfian).

• use -b, --bpk to specify a list of bpk to test,
• use -r to specify #rounds for each testcase,
• use -d to specify dataset directory.

plot.py will automatically gather results from same Z, ZD and bpk, and plot #files - execution_time graph for every file.

• use -b, --bpk to specify a list of bpk,
• use -d to specify output directories.

./build/main will be generated to run single testcase for 3 bpk distribution algorithms. Algorithms are implemented in cpp, see /src and /exec for details

• use -b to specify bpk,
• use -r to specify #rounds for the testcase,
• use -f to specify testcase name.