pun_nlp

Overview

pun_nlp is a robust NLP abstraction layer designed to simplify text processing and vectorization. It handles dependency management, resource downloading, and text preprocessing automatically, so you don't have to write boilerplate code.

It solves common issues with NLTK downloads and path errors by implementing a robust, lazy-loading resource manager that works in restricted environments like Kaggle and corporate servers.

Features

• Robust Resource Management: Automatically handles NLTK/Spacy downloads and SSL errors.
• Lazy Loading: Resources are only loaded into memory when needed.
• Type Safety: Prevents invalid combinations of operations (like vectorizing POS tuples).
• Unified API: Process single strings, lists, or 2D arrays of text with one method.
• Seamless Vectorization: Integrates directly with Scikit-Learn's TF-IDF and Count vectorizers.

Installation

pip install pun_nlp

Usage

Basic Pipeline

from pun_nlp import NLPProcessor

# Initialize with desired flags
p = NLPProcessor(
    tokenize=True, 
    stem=True, 
    remove_stopwords=True,
    normalize=True
)

text = "The QUICK brown foxes are running fast!"

# Automatically handles downloads and processing
print(p.process(text))
# Output: ['quick', 'brown', 'fox', 'run', 'fast']

NER & POS Tagging

# NER (Case sensitive checking happens before normalization)
p_ner = NLPProcessor(ner=True)
print(p_ner.process("Apple Inc. is hiring in California."))
# Output: [('Apple Inc.', 'ORG'), ('California', 'GPE')]

# POS Tagging (Tags tokens correctly before stemming)
p_pos = NLPProcessor(pos_tagging=True, stem=True)
print(p_pos.process("The boys are likely running."))
# Output: [('the', 'DT'), ('boy', 'NNS'), ('are', 'VBP'), ('like', 'RB'), ('run', 'VBG')]

Vectorization

p_vec = NLPProcessor(vectorize="tfidf", stop_words=True)
corpus = [
    "Machine learning is fascinating.",
    "Natural language processing is a subset of AI."
]

p_vec.fit_vectorizer(corpus)
vectors = p_vec.transform_texts(corpus)
print(vectors.shape)

Configuration

Parameter	Description
stem	Enable stemming (PorterStemmer).
lemmatize	Enable lemmatization (WordNet).
vectorize	"tfidf", "count", or None.
tokenize	Force return of token list.
remove_stopwords	Remove English stopwords (Case-insensitive).
pos_tagging	Return (Word, Tag) tuples.
ner	Return Entity tuples (Uses Spacy).
normalize	Lowercase & remove punctuation.
backend	"nltk" (default) or "spacy".

License

MIT License. See LICENSE for details.