README.md

Tensorflow metaframe for training models and getting predictions

This is a functional versioned metaframe website with matching npm module.

• Metaframe: https://metapages.github.io/metaframe-tensorflow-1d/
• Versioned metaframe e.g 0.3.0: https://metapages.github.io/metaframe-tensorflow-1d/v0.3.0/
• npm module (matches browser version): https://www.npmjs.com/package/@metapages/metaframe-tensorflow-1d

Inputs:

• training
• prediction
• model

Outputs:

• error: json
• prediction
• model

Input a list of training data consisting of n-1-dimensional series (e.g. a bunch of time series) and this will train a model:

Format of training examples:

{
    "examples": [
        {
            "label": "label1",
            "data": {
                "series": {
                    "series1": "base64-encoded-array-of-float32",
                    "series2": "base64-encoded-array-of-float32",
                    "seriesn": "base64-encoded-array-of-float32"
                }
            }
        },
        {
            "label": "label2",
            "data": {
                "series": {
                    "series1": "base64-encoded-array-of-float32",
                    "series2": "base64-encoded-array-of-float32",
                    "seriesn": "base64-encoded-array-of-float32"
                }
            }
        }
    ]
}

If there is an error (you also might see it in the metaframe itself):

The model will also be cached locally, so if the same training data is sent (compared via hashes) then the cached model will be returned, avoiding recomputing the model again.

A model can also be just passed in via the model input (same format as the output).

Once a model is loaded or computed, you can pass a prediction in, and get a prediction result out:

Example prediction input:

{
    "series": {
        "series1": "base64-encoded-array-of-float32",
        "series2": "base64-encoded-array-of-float32",
        "seriesn": "base64-encoded-array-of-float32"
    }
}

Example prediction output:

{
  "prediction": "label2",
  "predictions": { "label2": 0.94, "label1": 0.1, "labeln": 0.3 }
}