Welcome to madupite’s documentation!

madupite is a Python package for solving large-scale Markov decision processes (MDP) using distributed inexact policy iteration (iPI). The library is written in C++ and uses distributed sparse linear solvers from PETSc. And, in case you did not know, madupite is also the name of a volcanic rock!

• Installation
  • Dependencies
  • Python package
  • C++
  • Euler (ETH Zurich Cluster)
• Tutorials
  • Writing and running your code using madupite
  • Loading and reading data
  • Generating data
  • Matrix preallocation
  • Data format
  • The MDP-class
• Examples
  • Toy-Example
  • Inverted Pendulum
  • Further examples
• Algorithm
  • Does madupite still support standard dynamic programming algorithms such as VI and PI?
• Options
  • Mandatory Options
  • Facultative Options
  • Useful PETSc Options
  • How To Correctly Set the Options
• API Reference
  • madupite.createTransitionProbabilityTensor
  • madupite.createStageCostMatrix
  • madupite.MDP
  • madupite.MatrixPreallocation
  • madupite.getCommWorld
  • madupite.mpi_rank_size
  • madupite.writePETScBinary
  • madupite.MatrixType
  • madupite.MatrixCategory
  • madupite.Matrix
  • madupite.initialize_madupite

Index

• Index