Classifier Service For artyins deployment architecture

This is a submodule for the artyins architecture. Please refer to main module for full build details.

Refer to Trello Task list for running tasks.

---

Table of Contents (Optional)

• Usage
• Virtualenv
• Tests

---

Usage

The classifier service can be called by a HTTP POST call. Primarily on http://artyins-classifierservice:9891/infer_content. It expects a json of the following format

[{"id":1,"content":"adfsfswjhrafkf"},{"id":2,"content":"kfsdfjsfsjhsd"}]

After the content is successfully extracted, it will return a json of the following format

{"results":[{"id":1,"class":"DOCTRINE"},{"id":2,"section":"class":"DOCTRINE"}]}

Contributing

All model inferencing classes needs to implement the abstract Model class from score/model.py. An example is created in score/testmodel.py. All configuration should be set in config.py. Finally, add your model into the web service flask_app.py. Contributors, please ensure that you add your test codes into test.py

config.py

class InferenceConfig():
    MODEL_SAMPLE_INPUT=dict(sepal_length=1.0,sepal_width=2.2,petal_length=3.3,petal_width=4.4)
    MODEL_MODULE="score.testmodel"
    MODEL_CLASS="IrisSVCModel"
    MODEL_DIR = "model_files"
    MODEL_FILE = "svc_iris_model.pickle"

Abstract Model Class

from abc import ABC, abstractmethod
from schema import Schema
class ModelReport(ABC):
    """  An abstract base class for ML model prediction code """
    @property
    @abstractmethod
    def input_dataschema(self):
        Schema({'sentence': str})

    @property
    @abstractmethod
    def output_dataschema(self):
        Schema({'category': And(str,lambda s: s in ('doctrine', 'training','personnel'))
               ,'prob': And(float,lambda n: 0 <= n <= 100)})

    @abstractmethod
    #Implement the loading of model file here
    def __init__(self):
        raise NotImplementedError()

    @abstractmethod
    def predict(self, data):
        self.input_dataschema.validate(data)

An example on how to implement the Abstract model class

from score.model import ModelReport
from schema import Schema
from schema import Or
import os
import numpy as np
from config import InferenceConfig
import pickle
class DummyModel(ModelReport):
    # Note that this is overridden cos the one defined is meant for Keng On's use case 
    input_dataschema = Schema({'sepal_length': float,
                           'sepal_width': float,
                           'petal_length': float,
                           'petal_width': float})
    # Note that this is overriden cos the one defined is meant for Keng On's use case
    output_dataschema = Schema({'species': Or("setosa", 
                                          "versicolor", 
                                          "virginica")})
    def __init__(self,config=None):
        if config == None:
           config = InferenceConfig() 

    def predict(self, mydata):
        return ("TRAINING")

if __name__=="__main__":
    mymodel =DummyModel()
    classification=mymodel.predict(data)
    print(classification)

---

Virtualenv

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt`

---

Tests

This repository is linked to Travis CI/CD. You are required to write the necessary unit tests and edit .travis.yml if you introduce more model classes.

Unit Tests

import unittest

class TestModels(unittest.TestCase):

    def test_testmodel(self):
        print("Running TestModel Loading and Prediction")
        from score.testmodel import IrisSVCModel
        mymodel = IrisSVCModel()
        data=dict(sepal_length=1.0,sepal_width=2.0,petal_length=3.0,petal_width=4.0)
        classification=mymodel.predict(data)

    def test_modifiedtopicmodel(self):
        print("Running empty Modified Topic Model -  Sure pass")
        pass #Wei Deng to insert


    def test_bertmodel(self):
        print("Running empty Bert Model - Sure pass")
        pass #Kah Siong to insert

if __name__ == '__main__':
    unittest.main()

Web Service Test

#Start gunicorn wsgi server
gunicorn --bind 0.0.0.0:9898 --daemon --workers 1 wsgi:app

Send test POST request

import requests   
# api-endpoint
URL = "http://localhost:9891/test"
r = requests.get(url = URL) 
print(r) 

---