Skip to content

About

This section provides the API documentation for the kxi.ml Python library provided by KX for use by customers of KX Insights, both for interactions with the microservices/platform, and independent use.

Presently this library provides users with the ability to interact with the ML Registry for persistence, retrieval and removal of Machine Learning models written in q or Python within the following libraries:

  • q functions generated using PyKX.
  • Scikit-learn
  • Keras
  • Theano
  • PyTorch

Installation

Note

Python packages should typically be installed in a virtual environment. This can be done with the venv package from the standard library.

kxi.ml is only intended for use on Linux based systems; it is not supported presently on Windows or MacOS

Ensure you have a recent version of pip:

pip install --upgrade pip

Then to install kxi.ml using the KX PyPI server run the following command:

pip install --extra-index-url=https://$KX_PYPI_USER:$KX_PYPI_PASS@nexus.dl.kx.com/repository/kxi/simple/ kxi.ml

Where the environment variables KX_PYPI_USER and KX_PYPI_PASS have been set as appropriate. Please contact kxpy@kx.com for information on getting access to the KX PyPI server.

Python Dependencies

kxi.ml depends on the following Python packages:

  • kxi==1.0.2
  • scipy==1.7.3
  • scikit-learn==1.0.2
  • statsmodels==0.13.1
  • pandas==1.4.1
  • joblib==1.1.0

These are installed automatically by pip when kxi.ml is installed. The following are optional dependencies which can be installed by a user to provided additional functionality:

Library Version Usage
keras 2.7.0 Save Keras models to the ML Registry
torch 1.7.1 Save PyTorch models to the ML Registry
theano 1.0.5 Save Theano models to the ML Registry

Use cases

There are many ways users can make use of the machine learning functionality provided here:

  • Train models on historical data and deploy these to real-time systems
    • Anomaly detection within manufacturing.
    • Trade surveillance.
  • Machine Learning Operations development cycle
    • Storage of frequently updated models for future deployment (trading models/recommendation engines).
    • Deployment of stored models to production settings.
    • Use of versioning to maintain control over model accountability.