handler.py

from typing import Dict, List, Any
from setfit import SetFitModel
import numpy as np


class EndpointHandler:
    def __init__(self, path=""):
        # load model
        self.model = SetFitModel.from_pretrained(path)
    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
         data args:
            inputs (:obj: `str` | `PIL.Image` | `np.array`)
            kwargs
        Return:
            A :obj:`list` | `dict`: will be serialized and returned
        """

        # get inputs
        inputs = data.pop("inputs", data)
        if isinstance(inputs, str):
            inputs = [inputs]


        exerciselabels = ['positive experience',
                     'power posing',
                     'worry vs rumination',
                     'self-confidence',
                     'negative emotions',
                     'sleep',
                     'loneliness',
                     'imaginary friend',
                     'perfectionism',
                     'negative self-talk',
                     'woop',
                     'venting',
                     'worry window',
                     'act of kindness',
                     'blowing balloons',
                     'feeling on anger',
                     'power of smile',
                     'body scan',
                     'stress enhancing thoughts',
                     'anger ball of fire',
                     'emotions',
                     'lean against wall',
                     'breathing',
                     'crossed arms',
                     'energy traffic light',
                     'boundaries',
                     'Inner strength']  
                       
        # run normal prediction
        preds = self.model.predict(inputs)
        scores = self.model.predict_proba(inputs)
        
        label = [[el for el, p in zip(exerciselabels, ps) if p] for ps in preds]

        # Modify the label array
        modified_label = label[0]
        
        # Extract the positives probabilities  from each inner array
        modified_proba = [[inner[0][1]] for item, inner in zip(scores, scores)]
        
        
        score = [[el for el, p in zip(modified_proba, ps) if p] for ps in preds]
        
        # Modify the score array
        modified_score = score[0]
        
        
        combined_dict = {key: value for key, value in zip(modified_label, modified_score)}

        output_array = [combined_dict]

        # for element in combined_array:
        return label