Sensitivity analysis

A generic description of the SA pipeline implemented by EMSA is as follows:

1. Parameter Sampling: Latin Hypercube Sampling (LHS) is used to generate diverse sets of parameter values.

2. Model Evaluation: Each parameter set is evaluated using the specified epidemic model.

3. Sensitivity Measure: Partial Rank Correlation Coefficient (PRCC) is calculated to assess the sensitivity of model outputs to input parameters.

4. (Optional) Plot results: Figures can be generated using either EMSA’s built-in plotting functionality or a custom plotting method.

EMSA workflow followed in implemented examples:

Use cases requiring additional code

EMSA streamlines the sensitivity analysis by different degrees based on your use case. The simplest is when the following conditions are fulfilled:

• Basic Model structure: The analysed model has a structure that can be describe by the standard EMSA template

• Simple Sampling Process: The sampling is from uniform distribution, and there are no additional operations done to the samples or the target values.

• Simple Target Calculation: The target calculation process is in line with the one implemented in TargetCalculator. See documentation for caveats.

• Simple Model Evaluation: For a more complex ODE solver than the one in EpidemicModelBase, you will also need to implement that for your use case.

All of these use cases are discussed in the Advanced Topics section.

Basic use case

Now we will discuss the simplest use case, which is generally when the aforementioned conditions are fulfilled. The implementation (assuming the required inputs are available) is as follows:

from emsa.generics import SimulationGeneric

data = ...
model_struct_path = '...'
sampling_config_path = '...'

sim = SimulationGeneric(
    data=data,
    model_struct_path=model_struct_path,
    sampling_config_path=sampling_config_path
)

sim.run_sampling()             # Perform sampling
sim.calculate_all_prcc()       # Calculate Partial Rank Correlation Coefficients (PRCC)
sim.calculate_all_p_values()   # Calculate p-values
sim.plot_all_prcc()            # Plot PRCC results

Running this code will execute the full simulation pipeline, including sampling, PRCC calculation, and visualization using tornado plots.