How kormilitzin/med7 works
Project type and purpose
• This is an open-source NLP model package (Med7) for clinical named-entity recognition, built as a spaCy v3+ pipeline component.
• Its primary function is to be installed as a Python package (via pip), loaded into spaCy, and applied to free-text electronic health records to extract seven clinical entity types.
File structure
• Top-level files:
• README.md – project overview, installation, usage.
• LICENSE – licensing.
• .DS_Store – macOS artifact.
• Directories:
• images/ – static screenshots and graphs.
• tutorial/ – reproducible Jupyter notebook (example.ipynb) and conda environment definition (environment.yml).
• No backend/frontend split or microservices; it’s a library/model distribution with a usage tutorial.
README for additional insights
• Confirms three spaCy pipeline components: tokenizer/tagger, dependency parser, and an NER component trained on MIMIC-III.
• Installation flow: creates conda environment → pip install spaCy → pip install Med7 from Huggingface.
• Usage snippet: spacy.load("en_core_med7_lg") → process text → displaCy visualization → extract .ents.
Detailed capture
• Environment.yml under a “Reproducibility” module.
• Images/ folder assets as static resources used by README but not in the runtime flow.
Putting it all together:
• At the top: “Developer / Data Scientist”
→ uses “Tutorial Notebook” (environment.yml)
→ installs via “conda → pip install”
→ fetches model wheel from “Huggingface Hub”
• Center: “Python Runtime” with two sub-layers:
• spaCy Core
• Med7 Pipeline
– tagger
– parser
– clinical NER