Frequently Ask Questions

Here are answers to some questions you might have. If you have a concern not covered below, feel free to get in touch.

General Questions

The KXY platform aims to Democratize Lean AI.

  • - Derisk Your Machine Learning Projects: The KXY platform empowers you to never again start a machine learning project that will come to a dead-end.
  • - Your Machine Learning Projects, 10x Cheaper & 10x Faster: The KXY platform empowers you to quantify the feasibility of your machine learning experiments (i.e. model and/or variable choice), so that you may focus on the high ROI ones. We estimate that 1 in 10 experiments are dead-ends. The KXY platform can therefore allow you to slash your compute cost and project duration tenfold.
  • - Leverage Your Data, To Its Full Potential: Don't be satisfied with the performance you get from popular machine learning models. Determine how suboptimal your production model is, and get actionable insights on how to improve it.

What makes the KXY platform work is the realization that it is possible to estimate the feasibility of a machine learning experiment without actually runnning it. A machine learning experiment is considered feasible when it can yield a performance greater than a given baseline.

As it turns out, for a variety of metrics including R-squared, Root Mean Square Error, classification accuracy, to name but a few, it is possible to estimate the highest performance achievable when using a specific set of explanatory variables to predict a business outcome, without actually training any predictive model. We wrote the first paper explaining how to do this.

A useful analogy is oil production. Running a machine learning experiment without any informed expectation of its feasibility is like drilling an oil well to extract oil without any reason to believe that the site where the well is being drilled is rich in oil. Most of the time the endeavor will fail, thereby resulting in avoidable wastes of resources.

The oil and gas industry knows it very well and has created an entire field of study, namely exploration geophysics, dedicated to estimating the amount of oil in the ground and the expected ROI of an extraction project prior to, and independently from, launching the oil extraction project. Our machine learning papers lay out how to do this for machine learning experiments and projects.

Technical Questions


We provide a Python package that can be installed from PyPi by running

pip install kxy

or from source by running

git clone https://github.com/kxytechnologies/kxy-python.git & cd ./kxy-python & pip install .

We also provide the Docker image kxytechnologies/kxy that comes with our Python package, Jupyter and a variety of popular analytics libraries pre-installed. Programmatic access to the KXY platform requires an API key. Yours is

You can find more information on how to use the KXY platform programmatically, including a documentation of our Python package here.

You will find a big part of the math behind the KXY platform in our reference section.

Additionally, we wrote two machine learning papers describing the novelities making the KXY platform possible in greater details. One of them was recently published at the top-tier machine learning conference AISTATS 2021, and can be found here. The other is currently undergoing double-blind peer review prior to publication, and as such is not yet publicly available. In the meantime, if you would like to receive a private copy, feel free to send us an email at ml@kxy.ai.

If you didn’t find the answer you were looking for please get in touch.