CK new year wishes - feel free to add your own!

[gfursin]

Hi all,

I started collecting different CK wishes for 2019 here:

1. start preparing different "How to guides": "how to create a module", "how to create a software detection plugin", "how to create different types of packages (code, data sets, models)", "how to create a program workflow", etc.
2. update CK contributor guide
3. start standardizing all APIs of shared modules: http://cKnowledge.org/shared-modules.html
4. add continuous testing of shared CK packages across different compilers, platforms and sub-dependencies: http://cKnowledge.org/shared-packages.html
5. add continuous testing of all program workflows: http://cKnowledge.org/shared-programs.html
6. improve soft/package modules and meta descriptions (make them simpler and move different functionality from OS-dependent scripts to Python)
7. improve support for spack and easybuild
8. improve automatic generation of reproducible articles based on http://cKnowledge.org/rpi-crowd-tuning
9. improve CK workflows to crowd-source AI/ML/SW/HW autotuning and co-design: https://github.com/ctuning/ck-mlperf , http://cKnowledge.org/request and http://cknowledge.org/android-apps.html

Please feel free to add your own wishes with some ideas about to achieve them! Also please indicate if you can help yourself or provide resources - as a non-profit community project, we have limited resources and rely on your help!

Looking forward to further collaboration to improve our open R&D platform in 2019!

[gfursin]

Improve high-level overview of CK capabilities and program workflows: http://cknowledge.org/high-level-overview.pdf - feedback is welcome!

[kaul84]

• Look into ONNX runtimes and other similar inference optimized runtimes for heterogeneous hardware (e.g. the Microsoft ONNX Runtime - https://github.com/Microsoft/onnxruntime
  AWS Neo based on TVM - https://aws.amazon.com/sagemaker/neo/

[gfursin]

After the 1st Quantum Machine Learning hackathon in Paris, we got a feedback to provide a simpler installer: #98

[psyhtest]

For automatic testing, it would be handy to check out a particular revision of a repository or branch e.g.:

$ ck pull repo:ck-tensorflow --git_hash=655a20f99af32926cbf6d8fab092506ddd70e49c

or:

$ ck pull repo:ck-tensorflow --git_branch=object-detection-tflite

[psyhtest]

Another handy feature would be to search preprocessing/postprocessing scripts by tags, so, for example, instead of hard-coding:

        "pre_process_via_ck": {
          "data_uoa":       "b98ee24399ef4c3a",
          "module_uoa":     "script",
          "script_name":    "preprocess"
        },

one could use:

        "pre_process_via_ck": {
          "tags": "image-classification,imagenet,tf",
          "module_uoa":     "script",
          "script_name":    "preprocess"
        },

[gfursin]

We already addressed some of these issues and together with OctoML we started adding support for the ONNX runtime.
I close this ticket and move some ideas to #150 .