README.md

TfClassifier

Example package using TensorFlowJS to provide a classifier

This uses the Universal Sentence Encoder model

The Universal Sentence Encoder makes getting sentence level embeddings as easy as it has historically been to lookup the embeddings for individual words. The sentence embeddings can then be trivially used to compute sentence level meaning similarity as well as to enable better performance on downstream classification tasks using less supervised training data.

And this it does! We use this model and word vectors for passed in sentences to make an easy to use sentence classifier!

tests

Makefile has basic tasks

try make test to see an example run using test data

Have a look at the TestRunner for example usage.

training

To train the classifier, you need to pass an array of phrases:

(csv example)

tag,text
READ,   I need to read more
GYM,    Go to the gym
READ,   do more reading

• tag: a name for that 'class' of the classifier (I'll do multi-tags classes later)
• text: example phrase

You can (and should!) have multiple training phrases per tag.

Load the data, and then train the classifier:

    await testModel.loadCsvInputs('./data/inputs/train.csv')
    await testModel.trainModel({ useCache: useCache })

then you can see the best match for new phrases:

const matches: IMatch[] | undefined =
  await testModel.classify('I like reading books', { expand: true })

The returned results will be a list of matches with tag

If you passed expand: true you'll also get all the training samples which had that tag

input: 'I need to read more'

matches: [
  [
    confidence: 0.3560619056224823,
    tag: 'READ',
    pct: 36,
    sources: [
      { tag: 'READ', text: 'Read more books' },
      { tag: 'READ', text: "Read a children's book" },
      { tag: 'READ', text: 'Read the news in English Language' },
      { tag: 'READ', text: 'Finish a new book' },
      { tag: 'READ', text: 'Read a book every day' },
      { tag: 'READ', text: 'finish reading Content Marketing Part 1' }
    ]
  ],
  ...
]