Welcome to the pints documentation

Pints is hosted on GitHub, where you can find downloads and installation instructions.

Detailed examples can also be found there.

This page provides the API, or developer documentation for pints.

• Index
• Search Page

Contents

• Boundaries
• Core classes and methods
  • Forward model
  • Forward model with sensitivities
  • Problems
• Diagnosing MCMC results
• Diagnostic plots
  • Functions
  • MCMC Diagnostics
• Error measures
• Function evaluation
• I/O Helper classes
• Log-likelihoods
• Log-PDFs
• Log-priors
• MCMC Samplers
  • Running an MCMC routine
  • MCMC Sampler base classes
  • Adaptive Covariance MC
  • Differential Evolution MCMC
  • Dram ACMC
  • DreamMCMC
  • Dual Averaging
  • EmceeHammerMCMC
  • Haario ACMC
  • Haario Bardenet ACMC
  • Hamiltonian MCMC
  • Metropolis-Adjusted Langevin Algorithm (MALA) MCMC
  • Metropolis Random Walk MCMC
  • Monomial-Gamma Hamiltonian MCMC
  • No-U-Turn MCMC Sampler
  • Population MCMC
  • Rao-Blackwell ACMC
  • Relativistic MCMC
  • Slice Sampling - Doubling MCMC
  • Slice Sampling - Rank Shrinking MCMC
  • Slice Sampling - Stepout MCMC
  • MCMC Summary
• Nested samplers
  • Nested sampler base class
  • Nested ellipsoid sampler
  • Nested rejection sampler
• Noise generators
• Optimisers
  • Running an optimisation
  • Optimiser base classes
  • Convenience methods
  • Boundary transformations
  • Bare-bones CMA-ES
  • CMA-ES
  • Gradient descent (fixed learning rate)
  • Nelder-Mead
  • PSO
  • SNES
  • xNES
• Noise model diagnostics
  • Plotting functions
  • Diagnostics
• Toy problems
  • Toy base classes
  • Annulus Distribution
  • Beeler-Reuter Action Potential Model
  • Cone Distribution
  • Constant Model
  • Eight Schools distribution
  • Fitzhugh-Nagumo Model
  • Gaussian distribution
  • German Credit Hierarchical Logistic Distribution
  • German Credit Logistic Distribution
  • Goodwin oscillator model
  • HES1 Michaelis-Menten Model
  • High dimensional Gaussian distribution
  • Hodgkin-Huxley IK Experiment Model
  • Logistic model
  • Lotka-Volterra model
  • Multimodal Gaussian distribution
  • Neal’s Funnel Distribution
  • Parabolic error
  • Repressilator model
  • Rosenbrock function
  • Simple Egg Box Distribution
  • Simple Harmonic Oscillator model
  • SIR Epidemiology model
  • Stochastic degradation model
  • Stochastic Logistic Model
  • Twisted Gaussian distribution
• Transformations
• Utilities

Hierarchy of methods

Pints contains different types of methods, that can be roughly arranged into a hierarchy, as follows.

Sampling

1. MCMC without gradients
   • MetropolisRandomWalkMCMC, works on any LogPDF.
   • Metropolis-Hastings
   • Adaptive methods
     • AdaptiveCovarianceMC, works on any LogPDF.
   • PopulationMCMC, works on any LogPDF.
   • Differential evolution methods
     • DifferentialEvolutionMCMC, works on any LogPDF.
     • DreamMCMC, works on any LogPDF.
     • EmceeHammerMCMC, works on any LogPDF.
2. Nested sampling
   • NestedEllipsoidSampler, requires a LogPDF and a LogPrior that can be sampled from.
   • NestedRejectionSampler, requires a LogPDF and a LogPrior that can be sampled from.
3. Particle based samplers
   • SMC
4. Likelihood free sampling (Need distance between data and states, e.g. least squares?)
   • ABC-MCMC
   • ABC-SMC
5. 1st order sensitivity MCMC samplers (Need derivatives of LogPDF)
   • Metropolis-Adjusted Langevin Algorithm (MALA), works on any LogPDF that provides 1st order sensitivities.
   • Hamiltonian Monte Carlo, works on any LogPDF that provides 1st order sensitivities.
   • NUTS
6. Differential geometric methods (Need Hessian of LogPDF)
   • smMALA
   • RMHMC

Optimisation

All methods shown here are derivative-free methods that work on any ErrorMeasure or LogPDF.

1. Particle-based methods
   • Evolution strategies (global/local methods)
     • CMAES
     • SNES
     • XNES
   • PSO (global method)

Problems in Pints

Pints defines single and multi-output problem classes that wrap around models and data, and over which error measures or log-likelihoods can be defined.

To find the appropriate type of Problem to use, see the overview below:

1. Systems with a single observable output
   • Single data set: Use a SingleOutputProblem and any of the appropriate error measures or log-likelihoods
   • Multiple, independent data sets: Define multiple SingleOutputProblems and an error measure / log-likelihood on each, and then combine using e.g. SumOfErrors or SumOfIndependentLogPDFs.
2. Systems with multiple observable outputs
   • Single data set: Use a MultiOutputProblem and any of the appropriate error measures or log-likelihoods