Skip to content

Graph format #9

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
DanielMazurkiewicz opened this issue Feb 12, 2019 · 2 comments
Closed

Graph format #9

DanielMazurkiewicz opened this issue Feb 12, 2019 · 2 comments

Comments

@DanielMazurkiewicz
Copy link

DanielMazurkiewicz commented Feb 12, 2019

As I find it a separate topic to #3, would like to continue conversations about graph format itself (not operators and not api) here.

So far we have 3 major discussions:

  1. If to include any graph format support
  2. If it should be ONNX format
  3. If it should be JSON format

For the reasons I gave in these (and other) posts:

I opt for option 3 - JSON format.

Simple example with one hidden layer, 8 inputs and 9 outputs:

{
  activation: "tanh",
  layers: [8, 14, 9]
}

Same NN as above, but with weights data provided:

{
  activation: "tanh",
  layers: [8, 14, 9],
  setupData: 'ISEhIEhlbGxvIHdlYiBNTCAhISE='
}

Some "dummy" advanced example for JSON format:

{
  domain: 'custom-domain',
  dataType: 'fp32',

  layers: [
    {
      name: "input",
      pipes: [
        {
          name: "A",
          size: 30,
          to: ['addIfAboveInputA'] 
        },
        {
          name: "B",
          size: 30
          to: ['addIfAboveInputB'] 
        }
      ]
    },
    {
      type: 'addIfAbove',
      a: 'addIfAboveInputA',
      b: 'addIfAboveInputB',
      value: 0
    },
    {
      type: 'onlyLowerThan',
      value: 50
    },
    {
      type: 'neurons',
      activation: 'tanh',
      size: 30
    },
    30
  ]

}
@anssiko
Copy link
Member

anssiko commented Feb 18, 2019

At chartering time, feedback was given it is important to define the following is out of scope:

This Community Group does not attempt to mandate a specific neural network or Machine Learning model schema or format. Other groups are expected to address these requirements of this evolving area.

Given model formats are still evolving rapidly and there exists multiple such formats (and versions of each) that are in production use, defining a new model or graph format in this group is unlikely to contribute toward a solution (mandatory xkcd).

The current market accepted solution is to use model converters, also available as open source.

To conclude, the group chose option 1 initially. After revisiting related discussion in #3 I'm not seeing adequate support to change this at this time.

The issue is open to capture any further comments on the topic.

@anssiko
Copy link
Member

anssiko commented Feb 25, 2019

Hearing no further comments, I conclude the group is not interested in opening discussion on a new model or graph format at this time. This group position is consistent with the charter so no changes there.

@anssiko anssiko closed this as completed Feb 25, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants