Zero-shot Classification with Hugging Face Transformer Library

Zero-shot classification is a technique that allows us to classify text without having pre-labeled training data. This is particularly useful in real-world scenarios where labeling data can be time-consuming and expensive.

Human-Written Content: This article was demonstrated by a human, not AI.

Please support independent contributors like Code2care by donating a coffee.

Setting up Hugging Face Library: https://code2care.org/python/2-ways-to-setup-hugging-face-transformer-library-python/

The Hugging Face Transformer Library provides an easy-to-use pipeline for zero-shot classification. Let's explore this with a few examples:

Example 1: Basic Classification

from transformers import pipeline

classifier = pipeline("zero-shot-classification")
result = classifier(
    "Do you think Trump will win this election?",
    candidate_labels=["education", "politics", "business"],
)

print(result)

Example 2: Multi-label Classification

from transformers import pipeline

classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli", revision="c626438")

result = classifier(
    "The new AI policy will impact both the tech industry and academic research.",
    candidate_labels=["technology", "politics", "education", "economics"],
    multi_label=True
)

print(result)

Example 3: Classifying with Custom Labels

result = classifier(
    "The spacecraft successfully landed on Mars, marking a new era in space exploration.",
    candidate_labels=["astronomy", "technology", "history", "geology"],
    multi_label=True
)

print(result)

These examples demonstrate the flexibility and power of zero-shot classification using the Hugging Face Transformer Library. It allows us to classify text into any set of categories we define, without the need for specific training data for those categories.