The EncoderVersion API provides a scriptable object used in Predictive Intelligence stores.

This API requires the Predictive Intelligence plugin (com.glide.platform_ml) and is provided within the sn_ml namespace.

This API is used for working with encoder versions based on Encoder API objects in the Encoder store.

The system activates the most recent version of the encoder when it completes training, and only allows one version to be active at a time. However, you can activate any previously trained version you want to use to make predictions.

Methods in this API are accessible using the following Encoder methods:

EncoderVersion - getProperties()

Gets encoder object properties and version number.

Table 1. Parameters
Name Type Description
None
Table 2. Returns
TypeDescription
Object Contents of the Dataset and Encoder details. Results vary by object property setup.
{
  "algorithmConfig" : {Object},
  "datasetsProperties": [Array],
  "domainName": "String",
  "isActive": "String",
  "label": "String",
  "name": "String",
  "predictedFieldName": "String",
  "processingLanguage": "String",
  "scope": "String",
  "stopwords": [Array],
  "trainingFrequency": "String",
  "versionNumber": "Number"
}
<Object>.algorithmConfig Optional. JavaScript object containing algorithm configuration properties. 
'algorithmConfig' : {
  "algorithm": "String"
}

Data type: Object.
<Object>.algorithmConfig.algorithm Name of the algorithm for training this encoder.
Possible values:
  • paravec: Paragraph vector word embedding.
  • tf-idf: Term Frequency–Inverse Document Frequency (TF-IDF)-based text.

Data type: String.
<Object>.datasetsProperties

List of DatasetDefinition() properties associated with the encoder.

{
  "encodedQuery": "String",
  "fieldDetails": [Array],
  "fieldNames": [Array],
  "tableName": "String"
}

Data type: Array.
<Object>.datasetsProperties.tableName Name of the table for the dataset. For example, "tableName" : "Incident".

Data type: String.
<Object>.datasetsProperties.fieldNames List of field names from the specified table as strings. For example, "fieldNames" : ["short_description", "priority"].

Data type: Array.
<Object>.datasetsProperties.fieldNames.fieldDetails List of JavaScript objects that specify field properties. 
[
  {
    "name": "String",
    "type": "String"
  }
]

Data type: Array.
<Object>.datasetsProperties.fieldNames.fieldDetails.<object>.name Name of the field defining the type of information to restrict this dataset to.

Data type: String.
<Object>.datasetsProperties.fieldDetails.<object>.type Machine-learning field type.

Data type: String.
<Object>.datasetsProperties.fieldDetails.encodedQuery Encoded query string in standard Glide format. See Encoded query strings.

Data type: String.
<Object>.domainName Domain name associated with this dataset. See Domain separation and Predictive Intelligence.

Data type: String.
<Object>.isActive Flag that indicates whether this version is active.
Valid values:
  • true: Version is active.
  • false: Version is not active.

Data type: String
<Object>.label Identifies the prediction task. 
{
  "label": "my first prediction"
}

Data type: String.
<Object>.name System-assigned name.

Data type: String.
<Object>.predictedFieldName Identifies a field to be trained for predictability.

Data type: String.
<Object>.processingLanguage Processing language in two-letter ISO 639-1 language code format.

Data type: String.
<Object>.scope Object scope. Currently the only valid value is global.

Data type: String
<Object>.stopwords Optional. Preset list of strings that the system automatically generates based on the language property setting. For details, see Create a custom stopwords list.

Data type: Array.
<Object>.trainingFrequency The frequency to retrain the model.
Possible values:
  • every_30_days
  • every_60_days
  • every_90_days
  • every_120_days
  • every_180_days
  • run_once
Default: run_once

Data type: String.
<Object>.versionNumber Version number of the Encoder object.

Data type: String.

Example

The following example gets properties of the active object version in the store.

// Get properties
var mlEncoder = sn_ml.EncoderStore.get('ml_incident_categorization');

gs.print(JSON.stringify(JSON.parse(mlEncoder.getActiveVersion().getProperties()), null, 2));

Output:

*** Script: {
  "datasetsProperties": [
    {
      "tableName": "incident",
      "fieldNames": [
        "assignment_group",
        "short_description",
        "description"
      ],
      "encodedQuery": "activeANYTHING"
    }
  ],
  "domainName": "global",
  "isActive": "true",
  "label": "my encoder definition",
  "name": "ml_x_snc_global_global_my_encoder_definition",
  "processingLanguage": "en",
  "stopwords": [
    "Default English Stopwords"
  ],
  "versionNumber": "1"
}

EncoderVersion - getSentenceVectors(Array input)

Returns vectors for each input sentence.

Table 3. Parameters
Name Type Description
input Array Array of strings as sentences from which to receive vectors.
Table 4. Returns
Type Description
String Array of sentence vectors.

Example

The following example shows how to return a vector for a single sentence.

var myEncoderName = 'GloVe';

var myEncoder = sn_ml.EncoderStore.get(myEncoderName);

var input = ["I like to code."];

var vectors = myEncoder.getActiveVersion().getSentenceVectors(input);

gs.print(vectors);

Output:

*** Script: [-0.16243751347064972,0.30614474415779114,0.08489049971103668,
-0.48100000619888306,-0.170997753739357,0.08779674768447876,-0.07848624140024185,-0.15123701095581055,
-0.07843250036239624,-1.9505999088287354,0.3007825016975403,-0.07804800570011139,-0.04779449850320816,
0.04803549498319626,0.09848674386739731,0.2427891194820404,-0.41138750314712524,0.10880374908447266,
 … ,
0.21227750182151794,0.18478751182556152,-0.3113832473754883,-0.16560424864292145,0.09052124619483948]

EncoderVersion - getSimilarWords(Array input, Object options)

Returns words similar to each input word in the descending rank order of similarity.

Table 5. Parameters
Name Type Description
input Array Array of words for which to find similar words.
options Object Map to refine results. 
{ "topN":"String" }
options.topN String If provided, returns the top results up to the specified number of words. For example, use "10" to return the top 10 most similar words.
Table 6. Returns
Type Description
Array List of elements containing the similar words for the input word in the corresponding position. These similar words are represented by an of array of pairs in the format [word, similarity score].

Example

The following example shows how to get similar words using the GloVe encoder.

var myEncoderName = 'GloVe';
var myEncoder = sn_ml.EncoderStore.get(myEncoderName);
var input =  ["apple"];
var options = {"topN":"5"};
gs.print(myEncoder.getActiveVersion().getSimilarWords(input, options));

Output:

*** Script: [[["iphone",0.5987],["macintosh",0.5836],["ipod",0.5761],["microsoft",0.5664],["ipad",0.5628]]]

EncoderVersion - getStatus(Boolean includeDetails)

Gets training completion status.

Table 7. Parameters
Name Type Description
includeDetails Boolean Flag that indicates whether to return status details.
Valid values:
  • true: Return additional details.
  • false: Don't return additional details.

Default: False
Table 8. Returns
TypeDescription
Object JavaScript object containing training status information for an Encoder object.
{
  "state": "String",
  "percentComplete": "Number as a String",
  "hasJobEnded": "Boolean value as a String",
  "details": {Object}
}
<Object>.state Training completion state. If the training job reaches a terminal state, the job does not leave that state. If the state is terminal, the hasJobEnded property is set to true.
Possible values:
  • fetching_files_for_training
  • preparing_data
  • retry
  • solution_cancelled (terminal)
  • solution_complete (terminal)
  • solution_error (terminal)
  • solution_incomplete
  • training_request_received
  • training_request_timed_out (terminal)
  • training_solution
  • uploading_solution
  • waiting_for_training

Data type: String
<Object>.hasJobEnded Flag that indicates whether training is complete.
Valid values:
  • true: Training is complete.
  • false: Training is incomplete.

Data type: Boolean value as a String
<Object>.percentComplete Training percent complete. If the completion percentage is less than 100, the job might be in a terminal state. For example, if training times out.

Data type: Number as a String

Range: 0 thru 100
<Object>.details Object containing a list of additional training details.

Data type: Object

Example

The following example shows a successful result with training complete.

// Get status
var mlEncoder = sn_ml.EncoderStore.get('ml_incident_categorization');

gs.print(JSON.stringify(JSON.parse(mlEncoder.getActiveVersion().getStatus(true), null, 2)));

Output:

{
 "state":"solution_complete",
 "percentComplete":"100",
 "hasJobEnded":"true",
 "details":{"stepLabel":"Encoder Complete"} // This information is only returned if getStatus(true);
}

Example

The following example shows an unsuccessful result with training complete.

// Get status
var encoderName = 'ml_x_snc_global_global_encoder';
var mlEncoder = sn_ml.EncoderStore.get(encoderName);
var trainingStatus = mlEncoder.getLatestVersion().getStatus();

gs.print(JSON.stringify(JSON.parse(trainingStatus), null, 2));

Output:

{
 "state":"solution_error",
 "percentComplete":"100",
 "hasJobEnded":"true"
}

EncoderVersion - getVersionNumber()

Gets the version number of a solution object.

Table 9. Parameters
Name Type Description
None
Table 10. Returns
Type Description
String Version number.

Example

The following example shows how to get a version number.

// Get version number
var mlEncoder = sn_ml.EncoderStore.get('ml_incident_categorization');

gs.print("Version number: "+JSON.stringify(JSON.parse(mlEncoder.getActiveVersion().getVersionNumber()), null, 2));

Output:

Version number: 1

EncoderVersion - getWordVectors(Array input)

Returns vectors for each input word.

Table 11. Parameters
Name Type Description
input Array List of strings as words from which to receive vectors.
Table 12. Returns
Type Description
Array List of vectors for each word provided.

Example

The following example shows how to get a vector from the word hello.

var myEncoderName = 'GloVe';
var myEncoder = sn_ml.EncoderStore.get(myEncoderName);
var input =  ["hello"];

gs.print(myEncoder.getActiveVersion().getWordVectors(input));

Output:

*** Script: [[-0.337119996547699,-0.2169100046157837,-0.006636499892920256,
-0.41624999046325684,-1.2554999589920044,-0.0284659992903471,-0.7219499945640564,
-0.5288699865341187,0.0072085000574588776,0.3199700117111206,0.02942500077188015,
-0.013236000202596188,0.4351100027561188,0.2571600079536438,0.3899500072002411,
 … ,
0.3384299874305725,0.4055800139904022,0.18073000013828278,0.6424999833106995]]