KTT Hierarchical Classification System

KTT Hierarchical Classification System (KTT for short) is a training and inference framework specifically designed for hierarchial multilabel text classification (HMTC) workloads. It is preloaded with seven classification models and a builtin GPU-powered inference pipeline based on the BentoML model inference framework.

Dependencies

We highly recommend using an Anaconda distribution (such as Miniconda) to install all dependencies at the correct version. Simply use the supplied environment.yaml file to create an environment for KTT:

conda env create -n ktt -f ./environment.yaml
conda activate ktt

This is a complete and minimal environment with just the dependencies needed to run KTT to its full capabilities. For building documentation, please refer to that section.

Documentation

Our documentation is hosted publicly at our GitHub Page.

Documentation source is located in onnx/doc. We use Sphinx with the RTD theme.

Additional dependencies

Building documentation requires you to install additional dependencies not covered by our environment.yaml:

• sphinx
• sphinx-click
• sphinxcontrib-bibtex
• sphinxcontrib-svg2pdfconverter
• sphinx-rtd-theme

Ensure you are in KTT's conda environment (as doc-building still needs to import KTT's Python modules and thus requires all the runtime dependencies, too) and install these via pip as follows:

# Assuming said environment was named 'ktt'
conda activate ktt-lts
pip install sphinx sphinx-click sphinxcontrib-bibtex sphinx-rtd-theme

Building

cd doc
make html

Reading the docs

Open onnx/doc/build/index.html with your browser.