inference

README

---

title: ML Model Inference weight: 4618

Inference

This activity enables the inferencing of Machine Learning models within Flogo applications. This activity was built with a framework contribution model concept. The implemented framework is TensorFlow.

For detailed instructions, refer to the Flogo Documentation.

Installation

Flogo Web

This activity does not come pre-installed with the Web UI for a number of reasons, such as, the size, the requirement of TensorFlow lib and also the fact that it is an activity that is commonly used on a daily basis.

Flogo CLI

flogo install activity github.com/TIBCOSoftware/flogo-contrib/activity/inference

Schema

Inputs and Outputs:

{
  "input":[
    {
      "name": "model",
      "type": "string",
      "required": true
    },
    {
      "name": "framework",
      "type": "string",
      "required": true
    },
    {
      "name": "sigDefName",
      "type": "string",
      "required": false,
      "value":"serving_default"
    },
    {
      "name": "tag",
      "type": "string",
      "required": false,
      "value":"serve"
    },
    {
      "name": "features",
      "type": "array",
      "required": true
    }
  ],
  "output": [
    {
      "name": "result",
      "type": "object"
    }
  ]
}

Settings

Setting	Required	Description
model	True	The location to the archive. If using TensorFlow, the archive must contain the TensorFlow SavedModel
framework	True	The framework to use. Other frameworks can be registered at build time, the only available framework is TensorFlow
sigDefName	False	The default signature definition. This comes from the SavedModel metadata. The default value is serving_default
tag	False	The model tag. This comes from the SavedModel metadata. The default value is serve
features	true	An array of input features. Refer to the following sample.

Example

Estimators

The following example demonstrates how to invoke the inference activity and pass the input feature set for the tensor named inputs. Tensor names may vary, it is best to refer to the SavedModel metadata to identify the correct tensor name.

{
  "id": "inference_2",
  "name": "Invoke ML Model",
  "activity": {
    "ref": "github.com/TIBCOSoftware/flogo-contrib/activity/inference",
    "input": {
      "model": "Archive.zip",
      "framework": "Tensorflow"
    },
    "mappings": {
      "input": [
        {
          "type": "assign",
          "value": [
            {
              "name": "inputs",
              "data": {
                "z-axis-q75": 4.140586,
                "corr-x-z": 0.1381063882214782,
                "x-axis-mean": 1.7554575428900194,
                "z-axis-sd": 4.6888631696380765,
                "z-axis-skew": -0.3619011587545954,
                "y-axis-sd": -7.959084724314854,
                "y-axis-q75": 16.467001,
                "corr-z-y": 0.3467060369518231,
                "x-axis-sd": 6.450293741961166,
                "x-axis-skew": 0.09756801680727022,
                "y-axis-mean": 9.389463650669393,
                "y-axis-skew": -0.49036224958471764,
                "z-axis-mean": 1.1226106985139188,
                "x-axis-q25": -3.1463003,
                "x-axis-q75": 6.3198414,
                "y-axis-q25": 3.0645783,
                "z-axis-q25": -1.9477097,
                "corr-x-y": 0.08100326860866637
              }
            }
          ],
          "mapTo": "features"
        }
      ]
    }
  }
}

Documentation

Index

• func NewActivity(metadata *activity.Metadata) activity.Activity
• type InferenceActivity
  • func (a *InferenceActivity) Eval(context activity.Context) (done bool, err error)
  • func (a *InferenceActivity) Metadata() *activity.Metadata

Functions

func NewActivity

func NewActivity(metadata *activity.Metadata) activity.Activity

NewActivity creates a new InferenceActivity

Types

type InferenceActivity

type InferenceActivity struct {
	// contains filtered or unexported fields
}

InferenceActivity is an Activity that is used to invoke a a ML Model using flogo-ml framework

func (*InferenceActivity) Eval

func (a *InferenceActivity) Eval(context activity.Context) (done bool, err error)

Eval implements api.Activity.Eval - Runs an ML model

func (*InferenceActivity) Metadata

func (a *InferenceActivity) Metadata() *activity.Metadata

Metadata returns the activity's metadata