Step 2: Package the model into an API service

step_2_package_api.py

%%writefile sentiment_analysis_service.py

import bentoml

from bentoml.frameworks.sklearn import SklearnModelArtifact
from bentoml.service.artifacts.common import PickleArtifact
from bentoml.adapters import JsonInput

@bentoml.artifacts([PickleArtifact('model')])  # picke your trained model so that it can run on the server
@bentoml.env(pip_packages=["scikit-learn", "pandas"])  # specify the packages that your model depends on
class SKSentimentAnalysis(bentoml.BentoService):

    sentiment_names = {
        0: "very negative",
        1: "somewhat negative",
        2: "neutral",
        3: "somewhat positive",
        4: "very positive",
    }
    
    @bentoml.api(input=JsonInput())
    def predict(self, parsed_json):
        """
        Sentiment prediction API service
        
        Expected input format:
        ["Some text to predict the sentiment...", "some more text to predict sentiment"]

        Output format:
        {"sentiment_score": 4, "sentiment": "Very Positive", "tweet": "Tweet text to predict the sentiment..."}
        """
        texts = parsed_json
        predictions = self.artifacts.model.predict(texts)
        res = []
        for idx, pred in enumerate(predictions):
            res.append({
                "sentiment_score": pred, 
                "sentiment": self.sentiment_names[pred], 
                "text": texts[idx]
            })

        return res