Dictum Game Data Engine Python Project Walkthrough

Summary

The video is a step-by-step Python implementation walkthrough for a Dictum game word engine. The project gathers Latin vocabulary, expands entries into inflected forms with the Omnes Formae library, filters the results, and writes the final word list in the format expected by the game.

Project setup

• Creates a project folder and opens it in VS Code.
• Sets up a Python virtual environment and installs the required dependency from requirements.txt.
• Scaffolds the package, main module, and unit-test files.
• Demonstrates running tests as a Python module and packaging a submission for GradeScope.

Vocabulary data and loading

• Uses vocabulary data organized into up to 40 chapters.
• Builds a temporary downloader to fetch all chapters, inspect the data fields, and save a local JSON copy. Caching the data avoids downloading it every time the engine runs.
• Defines a vocabulary data class that accounts for optional fields and different field types across chapters.
• Implements and tests a loader that reads the JSON file and converts entries into vocabulary objects. The walkthrough reports 880 vocabulary entries in the data file.

Processing pipeline

The engine is developed as separate modules for loading vocabulary, filtering, expansion, and formatting.

• Filtering: Adds chapter and part-of-speech filters, with tests for matching chapters, empty selections, and empty input. A conversion helper maps source vocabulary types to the part-of-speech values used by Omnes Formae.
• Expansion: Uses Omnes Formae to generate Latin forms from vocabulary entries. Because multiple entries can produce the same spelling, the expander keeps only one form per spelling.
• Tag filtering: Applies requested grammatical criteria—such as case, number, gender, tense, voice, mood, and person—to generated forms. A missing tag does not automatically disqualify a form.
• Wordle constraints: Keeps only valid five-letter alphabetic forms.
• Formatting and output: Converts the filtered results into the required JSON structure, including fields for the word, lemma or full form, translation, and type.

Testing reveals an important logic issue: part-of-speech filtering must use each generated form’s own part of speech, not just the source vocabulary entry’s general type. For example, a form marked as a participle should not pass a filter intended for verbs. The walkthrough also catches and corrects issues in tag handling, output formatting, and test data.

Testing, AI assistance, and results

• Uses Python unit tests throughout, including small fixture data and tests for the major processing steps.
• Adds documentation strings and checks test coverage; the final local report reaches 100% coverage.
• Uses GitHub Copilot for suggestions, debugging, test writing, and code review, while keeping permissions manual and reviewing its changes. The narrator emphasizes staying involved rather than handing over the implementation entirely.
• Profiles a full run at about 1.5 seconds under the profiler, with roughly 99% of the time spent in Omnes Formae generating forms. The narrator concludes that optimizing the project’s own code is unlikely to make a major difference without reducing the number of forms sent to the library.
• The final submission receives 150 points (full credit). The narrator estimates the work took about 3 hours, and the Copilot chat usage cost about $1.23, according to the plan’s credit estimate.

Main speaker and source

dr. niqwis is the video’s narrator and project walkthrough presenter.