Training Text Classifier with spaCy#s TextCategorizer

This code uses spaCy's TextCategorizer component to train a text classifier. It takes in texts and corresponding labels as input and trains the classifier over a given number of iterations.

Technology Stack : spaCy, TextCategorizer, TextBlob

Code Type : Function

Code Difficulty : Intermediate

                
                    
import spacy
from spacy.pipeline import TextCategorizer
from spacy.util import minibatch

def train_text_categorizer(texts, labels, n_iter=100):
    """
    Train a text categorizer using spaCy's TextCategorizer component.
    """
    nlp = spacy.load('en_core_web_sm')
    textcat = nlp.create_pipe('textcat')
    nlp.add_pipe(textcat, last=True)
    
    # Add labels to the text categorizer
    textcat.add_labels(labels)
    
    # Train the text categorizer
    optimizer = nlp.begin_training()
    for itn in range(n_iter):
        losses = {}
        batches = minibatch([(text, label) for text, label in zip(texts, labels)], size=8)
        for texts, labels in batches:
            texts = [nlp(text) for text in texts]
            texts, labels = zip(*texts, labels)
            nlp.update(texts, labels, sgd=optimizer, drop=0.2, losses=losses)
    
    return textcat

# Example usage
texts = ["This is a good product", "I hate this product", "The quality is great", "It's not what I expected"]
labels = ["positive", "negative", "positive", "negative"]

textcat = train_text_categorizer(texts, labels)