Unigram Tagger and Evaluation

This repository contains the implementation of a Unigram Tagger and its evaluation as part of the "Programmieren II: Fortgeschrittene (Python)" course in the University of Heidelberg. The tasks involve reading text files, training a Unigram Tagger, evaluating it, and addressing method resolution order (MRO) concepts.

Features

• Custom Corpus Handling: Efficiently read and process text data using custom classes.
• Unigram Tagger Implementation: Train a Unigram Tagger from scratch, handling unseen words with a basic strategy.
• Evaluation Framework: Evaluate the tagger's performance with a custom evaluation method, providing detailed accuracy metrics.
• Object-Oriented Design: Utilize classes and methods to structure the code, ensuring maintainability and scalability.

Project Structure

• corpus_utils.py: Defines classes to read and handle the corpus data.
• tagger_models.py: Contains the Unigram Tagger implementation.
• eval_utils.py: Includes the evaluation logic for the tagger.
• main.py: The main script to execute training and evaluation.
• train.tsv: Training dataset.
• test.tsv: Test dataset.

Corpus Handling

Implemented in corpus_utils.py, the Corpus, Sentence, and Token classes provide:

• Reading the Corpus: Load and process the TSV file format.
• Corpus Statistics: Methods to get the number of sentences and access specific sentences.
• Iteration Support: Iterate through sentences in the corpus seamlessly.

Unigram Tagger

Implemented in tagger_models.py, the UnigramTagger class provides:

• Training: Train the tagger using the most frequent PoS tags for each word form.
• Tagging: Predict PoS tags for new sentences, handling unseen words efficiently.
• Callable Instances: Use the tagger instance as a function to tag sentences.

Evaluation

Implemented in eval_utils.py, the TaggerTester class provides:

• Performance Evaluation: Compare predicted tags with actual tags and calculate accuracy.
• Detailed Metrics: Output accuracy and other relevant metrics to assess performance.

Code Highlights

• No External Libraries: The implementation avoids libraries like collections, nltk, and pandas, showcasing pure Python solutions.
• Modular Design: Each class and method is designed for reusability and clarity.
• Error Handling: Robust error handling for unseen words and other edge cases.