nlpkit14

pip install --upgrade nlpkit14

import nlpkit14 as n
n.q1()  # Replaces this notebook cell with the full question and code

In Jupyter, run the cell containing n.q1() once to fill that same cell with editable code. Run the filled cell again to execute it. All n.q1() through n.q14() work this way. The question and expected output are comments above the program. In a plain Python shell, question functions execute directly.

Version 0.3.5 includes the Python dependencies for all fourteen exercises in its installation requirements. Each question prepares its missing NLTK data, spaCy model or Benepar model automatically before running. BERT downloads its pretrained files on first use and reuses the normal Hugging Face cache. No setup() call or manual model-extra installation is required.

Question titles, reference outputs and explanations remain comments in the bundled source, accessible using n.source(1). Questions print real computed output. Use n.questions() to list all fourteen.

Installation and first use require internet, a writable Python environment, sufficient disk space and supported dependency wheels. Large models take time to download. Existing data and models are reused. q8 always requires internet to fetch a live website. Most other questions work offline after preparation. For offline installation, transfer matching dependency/model wheels and all model/data caches in advance.

After upgrading in Jupyter, restart the kernel so it loads the new code. Print n.version to confirm 0.3.5. An already-running notebook can retain an older version even after pip succeeds.

This release is prepared locally until uploaded to PyPI. Large-model exercises and the entire dependency installation have not been tested across every platform. Model outputs may vary. See TEST_RESULTS.md.

Exact document code

All fourteen program bodies in version 0.3.5 match the supplied document code character for character after removing only leading/trailing blank lines. The question and reference output are added as comments. No helper imports, preparation calls, model.eval(), HTTP checks or algorithm corrections are inserted into the editable program. The library inserts the code immediately without importing NLP libraries or downloading data. A notebook pre-run hook prepares resources only when the exact filled code cell is executed. Editing the cell bypasses that hook; its original code and original download statements still run normally. Original download statements remain intact. Model outputs can vary because the original code is preserved, including its original training/evaluation choices.