Free Energy Minimization

Quick Start

Installation

• To install fem:

  pip install fem
  

  Note

  This package requires a fortran compiler to compile an external module; a popular, free choice is gfortran. This package also depends on the the lapack development files.

• Load fem in your Python script:

  import fem
  

Interactive notebook

Launch an interactive Jupyter notebook using Binder to run and edit the examples in the documentation:

Links

Online documentation:

http://joepatmckenna.github.io/fem

Source code repository:

https://github.com/joepatmckenna/fem

Python package index:

https://pypi.python.org/pypi/fem

Introduction

Free energy minimization (FEM) is a method for learning from data the probability distribution \(p\), with a form inspired by statistical physics, of an output variable \(y\) given input variables \(x_i\). We use \(p\) to both 1) understand the relations among the data variables \(x_i,y\), for example, to identify pairs or groups of variables that vary together and 2) to predict the output given new inputs. We are actively developing variations of the method that are conducive to modeling different types of data.

Table of Contents

• FEM for discrete data
  • Model
  • Method
    • Consistency and convergence
• Discrete data examples
  • Simulated time series
  • Direct information from a multiple sequence alignment
  • Comparing currency trading strategies
• Continuous data examples
  • Simulated continuous time series
  • vim2 data

Source Code

• fem package
  • Subpackages
    • fem.continuous package
    • fem.discrete package
  • Module contents
• fortran_module extension