Welcome to kdescent’s documentation!

kdescent provides a general framework for comparing an N-dimensional distribution of a model population to that of a training dataset. It allows you to perform stochastic gradient descent over mini-batched KDE statistics. The code is open-source and available on GitHub.

Contents:

• Installation Instructions
  • Installation
  • Prerequisites
• Quickstart Tutorial
  • Example model
  • Define loss function comparing \({\rm PDF}(x_1, x_2)\)
  • Run gradient descent
• Advanced Usage
  • More complex example model
  • Define loss function comparing \({\rm PDF}(g-r, r-z | M_\ast)\) and its Fourier pair
  • Run gradient descent
• Upweighting Example: The Halo Mass Function (HMF)
  • Define the model
  • Define loss function comparing \({\rm PDF}(g-r, r-z | M_\ast)\) and its Fourier pair
  • Descend the gradient without upweighting
  • Descend the gradient with upweighting
  • Closing Remarks
• Integration with multigrad
• API Reference
  • kdescent

Indices and tables

• Index

• Search Page