Figure 4 from the paper (real-time estimates from different groups) can be generated using compare_real_time_estimates.R. Figure 5 (within-method consistence) results from running within_method_agreement.R. First the consistence measures are calculated. The resulting .csv files are already contained in this repository (for days_until_final = 70) and can be used directly for plotting (see bottom of the script).
To plot the estimates as in Section 4 of the paper run plots_between_method_agreement.R.
To run the estimations based on the incidence data run parameter_influence_in_consensus_model.R and align_parameters_step_by_step.R. The Script parameter_influence_in_consensus_model.R runs the estimations using EpiEstim varying parameters and data between the choices of the research groups. The Script align_parameters_step_by_step.R runs the estimations using the different estimation methods with different levels of parameter adjustment. Some estimations have to be performed outside of this script, either in Python (globalrt, rtlive) or in a more time-intensive script (epiforecasts). Those estimates are contained in this repo and do not have to be calculated again. For all but rtlive and HZI, it is possible, though.
- epiforecasts: Run epiforecasts/epinow_estimation.R. Takes multiple hours depending on the resources and the length of input data.
- globalrt: Run ArroyoMarioli/input_output_dataset/estimate_R_filter_and_smoother.py (Might want to change the import between dataset.csv and dataset_final.csv. The former is the data used originally for the globalrt estimates. The latter is the correctly formatted aggregated version of the RKI line list data.)
- rtlive: The data on the number of tests performed is missing and not publicly available. Thus, the estimation cannot be performed on the basis of the repository.
- ETH: Before loading ETH data run the script ETH/otherScripts/format_linelist_data.R to calculate the delays. Then load the data with new_deconvolution = TRUE in the function load_incidence_data() with method = 'ETHZ_sliding_window'
- globalrt: Run the scripts from ArroyoMarioli/input_output/dataset/ in the following order: format_data_from_various_sources.R -> construct_dataset_adj.py -> construct_STAN_models.py -> estimate_R_filter_and_smoother.py
- rtlive: The incidence data used for the main comparison results from running the script rtlive/important_files_from_repo/notebooks/data_aggregation.ipynb.
- SDSC: The smoothed data can be constructed using the script SDSC/smoothing_example.ipynb.
Note that for the execution the RKI's line list data might be necessary. These are to large too be saved to this repository. Thus, they have to be sourced from https://hub.arcgis.com/datasets/dd4580c810204019a7b8eb3e0b329dd6_0/explore. Older versions of these data can be obtained through filtering by "Meldedatum".
- KITmetricslab/reproductive_numbers repository must be stored in the same directory as this repository. It contains the realtime estimates as published by the resepective teams until early 2022.