SZ_test includes evaluation scripts for ADT-FSE(lx1zhong/SZ_ADT at develop (github.com)).

All evaluations are divided into 3 parts:

• test1 for performance only of encoders.
• test2 for overall performance.
• test3 for compression ratio prediction.

Please run on linux and make sure you have 2GB of free space.

For short, you can do the main evaluation by simply running one script:

git clone https://github.com/lx1zhong/SZ_test.git
cd SZ_test
sudo ./all_in_one.sh

This will give Figure 6, 7 and 8 in the paper. （cost 3~4 compute-hours）

If you want to do the full evaluation, follow these steps:

• overview:
  1. Dependencies (5 human-minutes + 10 compute-minutes)
  2. Installation (10 human-minutes)
  3. Dataset download (0.5~2 compute-hours, depends on network)
  4. Edit script (2 human-minutes each test)
  5. Run (1 compute-hour each test)
  6. Draw (1 human-minutes each test)

1. Dependencies: if you have installed, ignore it.

   sudo apt install python3 python3-pip 
   pip install matplotlib numpy glob2 brewer2mpl
   

2. Installation: if you have installed, ignore it.

   1. Before evaluation, you should install SZ_ADT(ADT-FSE-enhanced SZ) here: (lx1zhong/SZ_ADT at develop (github.com)). The "./easytest.sh" script there may help you have a quick look of our work. Please remember your install path.

   2. Then install another tool ZFP here: (lx1zhong/zfp: Add performance statistic output (github.com)). This is for comparison with SZ_ADT.

3. Dataset: then download datasets from here (SDRBENCH) and here (small data). Make sure the directory will be in this format: [SDRBENCH/SMALL]-name-d1xd2xd3-[f/d], for example:

   test/
   -SDRBENCH-CESM_ATM-1800x3600-f/
   -SDRBENCH-EXAALT_HELIUM-7852x1037-f/
   -SDRBENCH-Miranda-256x384x384-d/
   -SMALL-dump_dense-55692-d/
   ......
   

4. Edit script: For each test, go to the specified directory, such as SZ_test/test1/1/, then scripts can be found in script directory. Before execute any script, be sure SZ/ZFP install paths and datasets path are right!

   The corresponding directory for the figures in the paper are as follows

   Figure.	Dir	Description
   Figure 5	test1/1
   Figure 6	test1/3
   Figure 7	test2/2	generated along with Figure 8
   Figure 8	test2/1
   Figure 9	test2/3
   Figure 10	test3

5. Run: First, execute shell script.

   ./script/run_multithread_abs.sh
   

   then results will be written into output directory.

6. Draw: After done, you can execute python scripts to draw the figure.

   python3 ./script/figure.py
   

We have integrated ADT-FSE into TDengine, a time-series database. You can get the full implementation here: TDengine (github.com).

The UK-DALE dataset in paper can be found here: The UK-DALE dataset, domestic appliance-level electricity demand and whole-house demand from five UK homes | Scientific Data (nature.com)