DiPS Data-informed Parameter Synthesiser

A Tool for Data-informed Parameter Synthesis for Discrete-Time Stochastic Processes from Multiple-Property Specifications

DiPS is a tool for parameter estimation for discrete time Markov chains (DTMCs) against multiple-property specifications provided in Probabilistic Computation Tree Logic (PCTL). The probability of satisfaction of each property is constrained by intervals computed from data. For a single property, the existing parametric model checking tools can compute a rational function over the chain’s parameters, which will evaluate exactly to the satisfaction probability / reward value for a given single property and chain. DiPS first computes these rational functions for each among the multiple properties by invoking the existing tools for parametric model checking PRISM and Storm.

Further, data measurements serve as an experimental estimation of the probability of satisfaction or reward value of PCTL formulae. We provide several methods for computation of confidence intervals for proportions using standard / CLT / Wald, Agresti-Coull, Clopper-Pearson, Jeffreys, Wilson, or 3/N method [1]. With intervals derived from the data, DiPS allows to further constrain the rational functions.

Finally, by coupling the obtained rational functions and data/constraints, DiPS implements four distinct methods for exploring which parameters of the chain are compliant with the data measurements:

• Space refinement - CEGAR like method splitting the parameter space. In each iteration, the result is provided by:
  • SMT solver - Z3 or dreal
  • interval arithmetic - scipy
  • Space sampling - checking satisfaction of constraints in sampled points,
  • Optimisation - searching for a parameter point with least distance to the data,
  • Bayesian inference (Metropolis-Hastings) - searching for most probable parameter points wrt. data.

The tool was primarily designed to facilitate flexible and efficient parameter search for stochastic models in computational systems biology, where scarcely available data measurements (e.g. only at the steady-state) can be naturally encoded as multiple-property specifications. However, the tool is directly applicable to any search problem, where multiple functions over unknown real variables are constrained by real-valued data.

A scheme showing tool components and communication among them:

Main GUI components, 6 tabs, in green, main functionality components in blue, and leveraged tools and libraries in red. For more information on how to use DiPS, please read this section. Feel free to leave response either via issues or an email.

HOW TO INSTALL

The tool was developed and optimised for Win10 and Ubuntu. The tool may not function properly on different operating systems.

1. INSTALL DEPENDENCIES:

• Python 3.7 or 3.8
  • Windows - just python
  • Ubuntu/Debian - Python header files should also be installed, please use sudo apt install python3-dev
  • Fedora/CentOS - Python header files should also be installed, please use sudo dnf install python3-devel
• PRISM 4.4 or 4.5
• install tkinter library
  • Windows - already done
  • otherwise go here
• other missing python packages
  • in the main directory MyDiPS run pip3 install -v .
• (Unix, Mac) - install dreal (optional)
  • pip3 install dreal
  • build dreal - info here
• Storm (optional, advanced)
• Prophesy (optional, advanced)

Are you having trouble with installing/running z3 or Storm? Read MyDiPS\README-z3.md or MyDiPS\README-storm.txt respectively. Still having trouble? Please contact us.

2. SETUP CONFIG (OPTIONAL)

In the main folder, there is config.ini file. Please fill 'prism_path' to use PRISM. Further items are optional. You can use them to navigate to I/O folders or change default settings of DiPS.

[mandatory_paths]

• prism_path: path to PRISM PRISM\bin\
• cwd: path to the main folder MyDiPS

[paths]

• models: path to PRISM models eg. MyDiPS/models
• properties: path to PRISM properties eg. MyDiPS/properties
• data: path to data eg. MyDiPS/data
• results: path to save results (all the results are saved in the subfolders) eg. MyDiPS/results
• tmp: path to save temporal/intermidiate files eg. MyDiPS/tmp

[settings]

Intervals

• number_of_samples : number of samples/observation
• confidence_level : 1 - alpha

Methods

Sampling

• grid_size : number of samples per dimension

Refinement

• max_depth : maximal number of splits
• coverage : proportion of nonwhite area to be reached
• algorithm : methods how to choose which region to pick after splitting
• solver : SMT solver - z3 or dreal
• delta : dealta for dreal option
• refine_timeout: refinement timeout

Metropolis-Hastings

• iterations : number of iterations, steps in parameter space
• mh_grid_size : number of segments/bins in the heatmap plot
• burn_in : trims the fraction of first accepted points
• mh_timeout: Metropolis-Hastings timeout

Meta settings

• autosave_figures : autosave figures
• minimal_output : set minimal console output
• extensive_output : set maximum console output
• show_mh_metadata : show metadata plot when running Metropolis-Hastings

HOW TO RUN

Now you can invoke DiPS in console, import the source code, or run the tool with GUI.

GUI

- open command line in the main DiPS directory (on Win - please open prompt with admin privileges to ensure changing the PRISM setting does not fail on permission denied)

>> cd src

>> python gui.py

Graphical User Interface should appear now (With some additional output returned in the command line).

For fully functional GUI (all the features fully visible) please set the scale of the screen to 100%.

HOW TO USE

For a brief summary, you can see our poster (CMSB 19), cmsb_poster.pdf.

To follow the main workflow of the tool and learn about each method using the graphical user interface, please see tutorial.pdf.

For more information about the methods, please follow the tool paper (submitted) or previous HSB paper [1] (if not reachable, please write us an email.)

HOW TO CITE US

[1] Hajnal, M., Šafránek, D., Petrov, T.: DiPS: A Tool for Data-informed Parameter Synthesis for Discrete-Time Stochastic Processes from Multiple-Property Specifications. In: European Performance Engineering Workshop. Springer International Publishing, (2021) (To be published)

ACKNOWLEDGMENT

I want to thank the following people who helped me with the code issues:

• Michal Ovčiarik
• Nhat-Huy Phung
• Denis Repin

and I want to thank the following people for user testing and feedback:

• Nhat-Huy Phung
• Denis Repin
• Samuel Pastva
• Stefano Tognazzi
• Morgane Nouvian
• Matej Troják

This project has been supported by the Centre for the Advanced Study of Collective Behaviour, Young Scholar Fund (YSF), project no. P83943018F P430 /18, and Max Planck Institute of Animal Behaviour.