Interactive articles

[ Home ]

Please, do not forget to check Getting Started Guides to understand CK concepts!

We attempt to unify various plotting mechanisms in the CK to reduce researchers burden. We have many examples of unified 1D, 2D and 3D plotting in the CK repository ck-analytics via JSON API:

 $ ck pull repo:ck-analytics
 $ ck list ck-analytics:demo:graph*

This approach together with distributed Unique IDs for each entry also makes it relatively straightforward to implement interactive graphs, articles and other objects integrated within HTML pages.

At the moment this functionality is working well but on a low SDK level, i.e. creating such content is not yet user-friendly. However, it should not be difficult to provide extra high-level modules to simplify creation of CK-powered widgets. If you are interested to help, please contact the CK community via this mailing list.

Here, we will just demonstrate a few examples of creating such interactive content based on autotuning examples of an image processing benchmark. These live examples can be found on the following pages:

• interactive autotuning and crowdtuning report
• interactive graph embedded to the HTML page
• live results during program crowd-tuning
• all shared graphs at cknowledge.org/repo
• all shared reports at cknowledge.org/repo

Above examples are available in the ck-autotuning repository which can be obtained via

$ ck pull repo:ck-crowdtuning

It is possible to perform compiler flag autotuning of cbench-automotive-susan benchmark via scripts from demo:cbench-automotive-susan entry:

$ cd `ck find demo:cbench-automotive-susan`
$ ./_clean_program_pipeline.sh
$ ./_setup_program_pipeline.sh
$ ./autotune_program_pipeline_i100.sh

During this autotuning, experimental results are recorded in the experiment:demo-autotune-flags-susan-linux-i100 entry.

It is now possible to produce and record a graph in PNG format (other formats such as jpeg and PDF are also supported) into graph:demo-autotuning-compiler-flags-susan-linux entry using another script:

$ ./plot_as_points_with_pareto_save_to_file.sh

This script is customized via plot_as_points_with_pareto_save_to_file.json:

 ...
  "out_repo_uoa":"ck-autotuning",
  "out_data_uoa":"demo-autotuning-compiler-flags-susan-linux",
  "out_to_file":"2d_points_time_vs_size_with_pareto.png",
  "save_table_to_json_file":"2d_points_time_vs_size_with_pareto.json",
  "save_table_to_csv_file":"2d_points_time_vs_size_with_pareto.csv"
 }

It specifies where to record PNG file (repo and entry UOA) as well as experimental table and CSV file (useful to rebuild graphs interactively) or allowing the community to process table via their own third-party tools such as Microsoft Excel, MATLAB, OpenOffice, etc.

The non-user friendly part is to prepare a meta description of the graph entry graph:demo-autotuning-compiler-flags-susan-linux to tell CK how to produce interactive HTML - we skip this part and let you check it yourself via:

$ ck load graph:demo-autotuning-compiler-flags-susan-linux

You can now start you local CK web server:

$ ck start web

And then visualize this graph via:

$ firefox http://localhost/ck/repo/web.php?wcid=graph:demo-autotuning-compiler-flags-susan-linux

It is also possible to visualize above experiments as table:

$ firefox localhost/ck/repo/web.php?wcid=experiment:demo-autotune-flags-susan-linux-best

Note, that both modules graph and experiment has an action html_viewer which is used to customize such views, otherwise a universal CK entry viewer will be used such as for datasets:

$ firefox http://localhost/ck/repo/web.php?wcid=8a7141c59cd335f5:b2130844c38e4a56
$ firefox http://localhost/ck/repo/web.php?wcid=dataset:image-pgm-0001

In the table view, it is also possible to customize raws using experiment.view module. Entries has a very simple meta-format, describing which flat keys to show and what is their description:

$ ck load experiment.view:autotuning-basic

 {
  "flat_keys": [
    "##choices#host_os", 
    "##choices#target_os", 
    "##features#platform#cpu#name", 
    "##features#platform#cpu#sub_name"
    ...
  ], 
  "flat_keys_desc": [
    {
      "desc": "Host OS", 
      "module_uoa": "0440cb72c2bc5cc6", 
      "type": "uoa"
    }, 
    {
      "desc": "Target OS", 
      "module_uoa": "0440cb72c2bc5cc6", 
      "type": "uoa"
    }, 
    {
      "desc": "CPU name"
    }, 
    {
      "desc": "CPU sub-name"
    }, 
    ...

This shows that you can not only share artifacts and workflows via CK repositories but also interactive graphs and reports making it much more convenient for the community or reviewers to test/validate your approach.

Furthermore, it is possible even to push output of benchmarks (such as slambench) directly to pages thus sharing efficient live demos and live reports as briefly described here:

• http://cknowledge.org/repo/web.php?wcid=1e348bd6ab43ce8a:3d487daf942ff64b

This documentation should be improved - please contact the CK community if you would like to help!