Figure 1
(Left) An experimentally measured noisy time series, and a simulated one. (Right) An example of an optimisation procedure with PINTS. Note that the actual simulation code is omitted from the example model wrapper at the top: this is the user-provided part, and can be written in Python or any other language that Python can interface with Python, allowing computationally heavy forward simulations to be handled entirely outside of PINTS. This image, and the full example code, can also be found in the PINTS repository.
Figure 2
An overview of the main PINTS classes used to define error measures (for optimisation) and PDFs (for optimisation or sampling). Users write a wrapper class for their model, making it available to pints, and must provide the experimental data using any Python sequence structure (for example, a list or a NumPy array). With these ingredients, a (single or multi-output) problem can be defined that can then be used with any of the available error measure or likelihood classes. Alternatively, users implement their own ErrorMeasure or LogPDF, which allows for further customisation and for problems other than time series problems to be solved.
Figure 3
The three steps iterated in an ask-and-tell interface. The stars here represent code specific to the chosen sampling or optimiser method and θ′ is the input parameter vector proposed by ask(). The state(.) of the system varies according to the method but typically holds a set of input parameter vectors and other constant or dynamic variables used by the ask() and tell() steps.