This repository contains a minimal utility for converting firm activity descriptions to ISIC Rev.4 codes.

1. Install the openai package and set your OPENAI_API_KEY environment variable if you want to use ChatGPT for classification.
2. Run the classifier on a CSV file:

python isic_classifier.py input.csv output.csv --gpt

This will read input.csv which must contain a column d1a1x with the activity description. The classified code will be written to output.csv.

Without the --gpt flag, a simple token overlap classifier is used instead of the ChatGPT API.

Run the automated tests with:

This repository provides a simple tool for mapping free-text firm activity descriptions to four digit ISIC Rev.4 codes.

Install the required packages:

pip install streamlit pandas

pytest -q

The classifier can process a CSV file containing a column named d1a1x:

python isic_classifier.py input.csv output.csv

The output CSV will contain the predicted code and a match score for each row.

A basic Streamlit interface is provided for classifying a single description or uploading a CSV file.

streamlit run app.py

The app allows you to download the classified results as a new CSV file.

This repository contains a minimal utility to map free-text business activity descriptions to ISIC Rev.4 four digit codes.

• isic_classifier.py – script and library for classifying activities.
• isic.csv – small sample of ISIC codes used for matching.
• sample_activities.csv – example CSV with a column d1a1x containing activity descriptions.
• test_classifier.py – simple pytest verifying classification of sample data.

python isic_classifier.py input.csv output.csv

The input CSV must contain a column named d1a1x. The output file will contain the original data along with isic_code and match_score columns.

To classify the provided sample activities, run:

python isic_classifier.py sample_activities.csv classified.csv

pytest -q

The included isic.csv contains only a small subset of ISIC Rev.4 codes for demonstration purposes. For production use, replace it with the full ISIC dataset.