mpi-sppy

Support for solving multi-stage stochastic programs based on a scenario discretization of the uncertainty. For parallelization, MPI is used.

Getting Started

• Quick Start
  • Installation
  • Verify Installation
  • Running the Farmer Example
  • What You Need to Provide For Your Problem
• Install mpi4py
  • no MPI?
  • AN IMPORTANT NOTE FOR MPICH USERS ON HPC PLATFORMS
  • Other pip extras

Building Your Model File

• Overview
• scenario_creator function
  • Scenario Tree Information
  • Examples
  • Scenario Probability
  • EF Supplement List
  • Surrogate Nonant List
  • Multi-stage
• Speed-up
• Helper Functions in the Model File
  • Required Functions
  • Optional Functions

Running with generic_cylinders

• generic_cylinders.py
  • Your Model File (Module)
  • Solving the Extensive Form
  • Running PH with Spokes
  • Choosing a Hub Algorithm
  • Choosing Spokes
  • Multistage Options
  • ADMM Decomposition
  • Rho Settings
  • Extensions via Command Line
  • Solution Output
  • MMW Confidence Intervals
  • Pickled Scenarios and Bundles
  • Advanced: hub_and_spoke_dict_callback
  • Advanced: Using a Class in the Module
  • custom_writer
  • config-file
  • solver-options
  • solver-log-dir
  • warmstart-subproblems
  • turn-off-names-check
  • Presolve (FBBT and OBBT)
• Examples
  • Tutorial: Farmer Example
  • aircond
  • aircondMulti
  • hydro
  • netdes
  • sslp
  • UC
  • sizes
• Solving the Extensive Form
  • mpisppy.opt.ef.ExtensiveForm Class
  • EF Extensions
  • Other method: mpisppy.utils.sputils.create_EF

Algorithms and Cylinders

• Overview
  • Roles
  • Basics
  • Cylinders
  • Solver threads
• Hubs
  • PH
  • PHPrimal
  • lshaped
  • cross_scen_hub
  • APH
  • Subgradient
  • FWPH
  • Hub Convergers
• Spokes
  • Outer Bound
  • Inner Bounds
  • General
• Extensions
  • Multiple Extensions
  • PH extensions
• Distributed Subproblem Presolve
• variable_probability
  • Objective function considerations
  • zero-probability variables
  • Scenario_lp_mps_writer_dir
• Rho Setting
  • Default Rho
  • Rho Setter Function
  • Separation-based Rho (--sep-rho)
  • Coefficient-based Rho (--coeff-rho)
  • Sensitivity-based Rho (--sensi-rho)
  • Reduced-costs-based Rho (--reduced-costs-rho)
  • Gradient-based Rho (--grad-rho)
  • Dynamic Rho Updates
  • Norm Rho Updater (--use-norm-rho-updater)
  • Primal-Dual Rho Updater (--use-primal-dual-rho-updater)
  • W-based Rho

Solutions and Confidence Intervals

• Output Solutions
  • EF (and Mid-level)
  • WheelSpinner
  • xhat for Confidence Intervals
  • Customized Writer Function
  • Amalgamator
• Programmatic Access to Solutions
  • EF
  • WheelSpinner
• MMW confidence intervals:
  • Inclusion in generic_cylinders
  • For Programmers
• zhat and Background for Confidence Intervals
  • xhat
  • zhat4xhat
  • seedoffset
• Sequential sampling
  • Examples

Advanced Topics

• Proper Bundles
  • Modules
  • Multistage
  • Notes
• Pickling Scenarios and Bundles
  • Why Pickle?
  • Basic Pickling and Unpickling
  • Pre-Pickle Preprocessing
  • Tuning Workflow with Pickling
• Jensen’s Bound as potential starting bound
  • What the options do
  • Authoring average_scenario_creator
  • Seed management
  • Worked example: farmer
  • Interaction with each spoke’s own iteration 0
  • Convexity limitations, in more detail
• Per-scenario-feasible candidate xhat
  • The contract
  • File layout
  • Why this convention exists
  • In-cylinder use: --<xhatter>-try-feasible-xhat-first flags
  • Average-data-based Methods
  • See also
  • Heuristics fixing methods
• Supplying an Initial Xhat from a File
  • When This is Useful
  • Enabling the Feature
  • File Format
  • Example
  • Scope and Limitations
  • Interaction with --*-try-jensens-first
  • Interaction with --xhat-feasibility-cuts-count
  • Follow-up Milestones
  • See Also
• SMPS Format Support
  • SMPS Files
  • Usage
  • Limitations
• AML Agnosticism
  • Loose integration
  • Tight integration
  • Special Note for developers
  • Architecture
  • Overview of Recommended Bundle Practices
  • Some notes
  • Example
• ADMM with generic_cylinders
  • Tutorial: Running the distr Example
  • Model Module Interface
  • Creating Your Own ADMM Model
  • Bundling with Stochastic ADMM
  • Reference: CLI Arguments
• AdmmWrapper
  • Usage
  • Distribution example
• StochAdmmWrapper
  • Example Driver file
  • Stochastic distribution example
• APH
  • Options
  • x-hat
  • Convergence Metric
  • Dispatch
  • farmer
  • A Peek Under the Hood
• Running without MPI and mpi4py
• Secret Menu Items
  • initial_proximal_cut_count
  • subgradient_while_waiting
• PySP Conversion (Legacy)
  • Details

For Developers and Contributors

• Writing Custom Drivers (Advanced)
  • Starting with Examples
  • Extending Examples
  • Not Using Examples Utilities
  • A Note about Options
  • Contrasting _mpisppy_node_list and all_node_names
• Internals
  • nonant_vardata_list
  • nonant_indices
• Buffer Sanity Inspector
  • When to Use This
  • What the Inspector Checks
  • Manual Use
  • Command-Line Trigger at Cylinder Shutdown
  • Extending: Adding a Field Checker
  • See Also
• Notes for Code Contributors
  • Tests
  • import mpi
• Amalgamator (Internal/Advanced)
  • The Amalgamator class
  • Create Amalgamator from a module and command line
  • Notes about inparser_adder
  • Amalgamator with EF
  • Examples
• Code Coverage
  • Running the Full Suite with run_coverage.bash
  • Using --python-args for Individual Scripts
  • Tips

References

• APIs
  • High-level
  • Mid-level
  • Low Level
• References

Indices and Tables

• Index

• Module Index

• Search Page