Supplementary files for the dingo Python library

We compare the Multiphase Monte Carlo flux Sampling (MMCS) feature of the dingo library against the combined method of PolyRound (for rounding) followed by hopsy (for sampling) on a set of 7 models with a ranging dimension (ext_data.zip).

The simpl_transf_polytopes.zip contains the polytopes retrieved after the simplify() and transform() functions of the PolyRound library. These polytopes were used as input for the dingo implementation of the MMCS algorithm asking for an ESS of 1000. Under the dingo_samples_on_simpl_transf_polytopes.zip the resulting samples from dingo can be found using the MMCS algorithm. 

Similarly, polyrounded_polytopes.zip contains the polytopes retrieved after applying simplify(), transform() and round() functions of the PolyRound library. These polytopes were used as input for the hopsy library, again, asking for an ESS of 1000. The hopsy_samples.zip folder contains the resulting samples from hopsy library, using a thinning of 100d ; only in the case of Recon3D a thinning of 200d was used as suggested by the authors. Under the hopsy_samples_ess_1000.zip folder, we provide the hopsy samples with an ESS of 1000. Last, the samples produced using the efficient Billiard Walk implementation of dingo can be found under the BWRsamples.zip file. In this case, 20000 points were sampled for each model.

Further, the sars_samples.zip file contains dingo samples from the solution space of the SARS-CoV-2 integrated model of Renz et al (2020) for the following cases: 

unbiased; where the zero vector has been used as the objective function of the model

after maximising for the human biomass 

after maximising for the virus biomass objective function (VBOF)

The following Python scripts to perform these experiments are included:

polyround_preproces.py : runs the PolyRound functions and builds the simplified and transformed polytopes that dingo will use as well as the simplified, transformed and rounded polytopes hopsy uses

hopsy_on_polyrounded_polytopes.py : performs sampling with hopsy 

dingo_on_simpl_transf_polytopes.py : performs sampling with dingo 

run_bwr_exp.py computes samples using the efficient Billiard Walk of dingo

binary_search.py : a function to return the index in the chain where ESS becomes 1000

compute_ess.py: based on a model's hopsy samples (under the hopsy_samples.zip folder) it retrieves the samples with an ESS of 1000 and the corresponding required time for hopsy to build them. The script requires the total time of the hopsy experiment recorded in the model's corresponding .txt file (you can find this under the hopsy_samples.zip)

compute_ess_psrf_per_phase.py computes ESS and PSRF in specific indices which correspond to those when MMCS switches from a phase a to next one

A notebook is available for how the integrated model was sampled.