Sample Sentiment model converted into ONNX and run in MarkLogic. Original Model : https://github.com/cocoa-ai/SentimentCoreMLDemo/blob/master/SentimentPolarity/Resources/SentimentPolarity.mlmodel

1_convertModel.py

import coremltools
# Load a Core ML model
coreml_model = coremltools.utils.load_spec('SentimentPolarity.mlmodel')
# Convert the Core ML model into ONNX
onnx_model = onnxmltools.convert_coreml(coreml_model, 'Sentiment Polarity')
# Save as protobuf
onnxmltools.utils.save_model(onnx_model, 'SentimentPolarity.onnx')

2_Execute_Model.sjs

function getWordFrequencies(sentence) {
  return cts.tokenize(sentence.toLowerCase()).toArray()
        .filter(token=>token instanceof cts.word)
        .map(token=>String(token))
        .reduce((accum,token)=> {
          if (!accum[token]) {
            accum[token] = 1
          } else {
            accum[token] = accum[token] + 1
          }
          return accum
        },{})
}

function wordFrequenciesToOnnxMap(wordFrequencies) {
  let entries = Object.entries(wordFrequencies)
  let words = []
  let frequencies = []
  for (let [word, freq] of entries) {
    words.push(word)
    frequencies.push(freq)
  }
  return ort.map(ort.string(words,[entries.length]),ort.value(frequencies,[entries.length],"float"))
}

function classProbabilityToObject(classProbability) {
  let seq = ort.getValue(classProbability,0)
  let keys = ort.stringContent(ort.getValue(seq, 0))
  let values = ort.valueGetArray(ort.getValue(seq, 1))
  let output = {}
  for (let i=0; i<keys.length;i++) {
    output[keys[i]] = values[i]
  } 
  return output
}

let sentence = "Pizza is awesome";

let wordFreq = getWordFrequencies(sentence)
let input = wordFrequenciesToOnnxMap(wordFreq)

const session = ort.session(cts.doc("/onnx/model/SentimentPolarity.onnx"));
let output = ort.run(session, { input: input })

let result = {
  "classLabel" : ort.stringContent(output.classLabel),
  "classProbability" : classProbabilityToObject(output.classProbability)
}

result

/* Sample Output

{
   "classLabel": [
                  "Pos"
                 ], 
   "classProbability": {
                  "Neg": 0.469456851482391, 
                   "Pos": 0.530543148517609
   }
}

*/