Sensitivity Dispatcher

The sensitivity dispatcher is a tool for running one or more predefined sensitivity experiments using the dataset notebooks/double-entry-data/double_entry_final.csv. It is intended to use when running experiments on a server.

See also

Documentation for Sensitivity Dispatcher Examples

Documentation for Individual Sensitivity Experiments

Sensitivity Dispatcher Usage

usage: python scripts/sensitivity_dispatcher.py [-h] [--max_processes MAX_PROCESSES]
                                                [--categories CATEGORIES [CATEGORIES ...]] [--dry_run]

Named Arguments

--max_processes

Number of processes to spawn

--categories

Run types to execute

--dry_run

Print run types selected and exit

Default: False

Sensitivity Analysis Dispatcher Run Types

Full specifications can be found in scripts/sensitivity_analysis/sensitivity_analysis.yaml, which can also be customised to your needs.

region_holdout

Mask data for one region and output prediction for the masked region. Runs once for each region.

npi_leaveout

Remove NPI indicators from dataset and fit model with the remaining set of NPIs. Runs once for each NPI, and additionally leaves out school and university closures jointly

cases_threshold

Run the model masking daily confirmed cases when a region’s total number of confirmed cases is below [10, 50, 150, 200] cases

deaths_threshold

Run the model masking daily death counts when a region’s total number of deaths is below [1, 5, 30, 50] deaths

oxcgrt

Run the model with additional featurs from OxCGRT:

[‘Travel Screen/Quarantine’, ‘Travel Bans’, ‘Public Transport Limited’,

’Internal Movement Limited’, ‘Public Information Campaigns’, ‘Symptomatic Testing’]

R_prior

Run the model with a prior mean R0 of [2.5, 4.5]

growth_noise

Add noise to growth rate

NPI_prior

Run the model with an NPI effetiveness prior of [skewed 10, Normal(0,0.2), ICL]

agg_holdout

Mask the final 20 days of data and predict this period with the model

any_npi_active

Add an additional dummy NPI representing whether any major NPI is active

delay_schools

Run the model with an additional 5 day delay to school and university closures

npi_timing

Run the model with NPI indices 0-8 representing 0,…,8 active NPIs, ignoring NPI type

structural

Run the model with alternative structures:

[additive, discrete_renewal, discrete_renewal_fixed_gi,

noisy_r, different_effects, cases_only, deaths_only]