scripts/predict.py

"""
Script file used for performing inference with an existing model.
"""

import torch
import nltk
from nltk.tokenize import sent_tokenize

from transformers import (
    AutoTokenizer,
    AutoModelForSequenceClassification
)

from scripts.config import (
    BIN_REPO,
    ML_REPO
)

class InferenceHandler:
    """A class that handles performing inference using the trained binary classification and multilabel regression models."""

    def __init__(self, api_token: str):
        """Constructor for instantiating an InferenceHandler object.

        Parameters
        ----------
        api_token : str
            A Hugging Face token with read/write access privileges to allow exporting the trained models (default is None).
        """

        self.api_token = api_token
        self.bin_tokenizer, self.bin_model = self._init_model_and_tokenizer(BIN_REPO)
        self.ml_regr_tokenizer, self.ml_regr_model = self._init_model_and_tokenizer(ML_REPO)
        nltk.download('punkt_tab')

    def _init_model_and_tokenizer(self, repo_id: str):
        """Initializes a model and tokenizer for use in inference using the models path.

        Parameters
        ----------
        repo_id : str
            The repository id (i.e., <owner username>/<repository name>).

        Returns
        -------
        tuple[PreTrainedTokenizer | PreTrainedTokenizerFast, PreTrainedModel]
            A tuple containing the tokenizer and model objects.
        """

        tokenizer = AutoTokenizer.from_pretrained(repo_id, token=self.api_token)
        model = AutoModelForSequenceClassification.from_pretrained(repo_id, token=self.api_token)
        model.eval()
        return tokenizer, model

    def _encode_binary(self, text: str):
        """Preprocesses and tokenizes the input text for binary classification.

        Parameters
        ----------
        text : str
            The input text to be preprocessed and tokenized.

        Returns
        -------
        BatchEncoding
            The preprocessed and tokenized input text.
        """

        bin_tokenized_input = self.bin_tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
        return bin_tokenized_input

    def _encode_multilabel(self, text: str):
        """Preprocesses and tokenizes the input text for multilabel regression.

        Parameters
        ----------
        text : str
            The input text to be preprocessed and tokenized.

        Returns
        -------
        BatchEncoding
            The preprocessed and tokenized input text.
        """

        ml_tokenized_input = self.ml_regr_tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
        return ml_tokenized_input

    def _encode_input(self, text: str):
        """Preprocesses and tokenizes the input text sentiment classification (both models).

        Parameters
        ----------
        text : str
            The input text to be preprocessed and tokenized.

        Returns
        -------
        tuple[BatchEncoding, BatchEncoding]
            A tuple containing preprocessed and tokenized input text for both the binary and multilabel regression models.
        """

        bin_inputs = self._encode_binary(text)
        ml_inputs = self._encode_multilabel(text)
        return bin_inputs, ml_inputs
    
    def classify_text(self, input: str):
        """Performs inference on the input text to determine the binary classification and the multilabel regression for the categories.

        Determines whether the text is discriminatory. If it is discriminatory, it will then perform regression on the input text to determine the
        assesed percentage that each category applies.

        Parameters
        ----------
        input : str
            The input text to be classified.

        Returns
        -------
        dict[str, Any]
            The resulting classification and regression values for each category.
        """

        result = {
            'text_input': input,
            'results': []
        }

        sent_res_arr = []
        sentences = sent_tokenize(input)
        for sent in sentences:
            text_prediction, pred_class = self.discriminatory_inference(sent)

            sent_result = {
                'sentence': sent,
                'binary_classification': {
                    'classification': text_prediction,
                    'prediction_class': pred_class
                },
                'multilabel_regression': None
            }

            if pred_class == 1:
                ml_results = {
                    "Gender": None,
                    "Race": None,
                    "Sexuality": None,
                    "Disability": None,
                    "Religion": None,
                    "Unspecified": None
                }

                ml_infer_results = self.category_inference(sent)
                for idx, key in enumerate(ml_results.keys()):
                    ml_results[key] = min(max(ml_infer_results[idx], 0.0), 1.0)

                sent_result['multilabel_regression'] = ml_results
            sent_res_arr.append(sent_result)

        result['results'] = sent_res_arr
        return result
    
    def discriminatory_inference(self, text: str):
        """Performs inference on the input text to determine the binary classification.

        Parameters
        ----------
        text : str
            The input text to be classified.

        Returns
        -------
        tuple[str, Number]
            A tuple consisting of the string classification (Discriminatory or Non-Discriminatory) and the numeric prediction class (1 or 0).
        """

        bin_inputs = self._encode_binary(text)

        with torch.no_grad():
            bin_logits = self.bin_model(**bin_inputs).logits

        probs = torch.nn.functional.softmax(bin_logits, dim=-1)
        pred_class = torch.argmax(probs).item()
        bin_label_map = {0: "Non-Discriminatory", 1: "Discriminatory"}
        bin_text_pred = bin_label_map[pred_class]

        return bin_text_pred, pred_class
    
    def category_inference(self, text: str):
        """Performs inference on the input text to determine the regression values for the categories of discrimination.

        Parameters
        ----------
        text : str
            The input text to be classified.

        Returns
        -------
        list[float]
            A tuple consisting of the string classification (Discriminatory or Non-Discriminatory) and the numeric prediction class (1 or 0).
        """

        ml_inputs = self._encode_multilabel(text)

        with torch.no_grad():
            ml_outputs = self.ml_regr_model(**ml_inputs).logits
        
        ml_op_list = ml_outputs.squeeze().tolist()

        results = []
        for item in ml_op_list:
            results.append(max(0.0, item))

        return results