Multiround Trustgame POMCP release This file will outline the necessary steps to run the POMCP based multiround trustgame, estimation ("IPOMCP"), classifying subjects by guilt parameter, depth of planning and theory of mind level.
Prerequesites: 1.The code requires the inclusion of the boost C++ libraries in a recent (> 1.33) version. 2. The present iteration is intended for MSVC++ compilers and has been built with SDK7.1 in 64 bit. 3.Is intended to be used in conjunction with Matlab, to prepare data and evaluate outcomes
Steps: 1.First build & run the "Trust Precalculation" project.
-This will generate a 2.3 GB large file that contains the precalculated Level 0 investor action values, to make those instantly available for the other programm parts. A copy of this file needs to be included in the build directories of the other projects. This dependency will be removed in the next iteration.
- Next move to the "Trust Generative" project:
2.1 Run the GenerateValidationInput File to create a small (~10) sample of subject parameters. Valid parameterranges are noted in the input messages.
2.2 Build the "Trust Generative" project from the contained C++ files. Include the precalculated level 0 investor values in the directory.
2.3 Run the Project and use the .bin file generated by GenerateValidationInput to convey your parameters of interest to the estimation procedure
- Now move to the "Trust Estimation" project:
3.1 Build the Trust Estimation project from the contained C++ files. Include the precalculated level 0 investor values in the directory.
3.2 Run the estimation and use the .bin file from the Trust Generative project as input. This will take roughly 2 hours per subject/
3.2 Use the output .bin file together with ReadEstimationInput in Matlab to read out the results. Be aware that certain parameter combinations will not be independent (planning 0 always implies level 0, for the trustee also the converse holds, for instance).
4.If interested you can also build the "Trust Accuracy" project. This will allow you to either generate estimated accuracies for different path numbers for a given planning horizon (discrepancy table) or produce a close to converged model investor for each planning horizon (warning: run time about 5 days due to the long planning horizons).
-NOTE: Remember to include the precalculated level 0 investor values for the accuracy project as well.