How do I use SALib with batch_run?

[edday1]

I am trying to run a sensitivity analysis using SALib. I found some old code (below) that does this and I am trying to adapt it to use batch_run.
I don't know what to do with replacing BatchRunnerMP. I figure maybe I could use batch_run with one iteration but how do I pass variable_parameters into that?

Anyone got a modern example of how to do this?

  # Variable parameters
    problem = {'num_vars': 3,
               'names': ['spawn', 'agression', 'min_gap'],
               'bounds': [[0.3, 0.8], [0.2, 0.8], [0.5, 2.0]]}
    
    replicates = 5
    max_steps = 5000
    distinct_samples = 5
    
    param_values = saltelli.sample(problem, distinct_samples)
    
    model_reporters={'Final_avg_speed': avg_speed,
                     'Final_Cars_in_lane': cars_in_lane,
                     'Data Collector': lambda m: m.datacollector,
                     'Total_Avg_speed': avg_speed,
                     'Total_Cars_in_lane': cars_in_lane,
                     'Variance_speed': avg_speed,
                     'Variance_car': cars_in_lane}
    
    batch = BatchRunnerMP(RoadSim,
                          nr_processes=8,
                          max_steps=max_steps,
                          variable_parameters={name:[] for name in problem['names']},
                          model_reporters=model_reporters)
    
    count = 0
    for i in tqdm(range(replicates)):
        for vals in tqdm(param_values):
            vals = list(vals)
            variable_parameters = {}
            for name, val in zip(problem['names'], vals):
                variable_parameters[name] = val
            batch.run_iteration(variable_parameters, tuple(vals), count)
            count += 1
    
    data_original = batch.get_model_vars_dataframe()
    data = data_original.copy()
    print(data.shape)
    
    for i in tqdm(range(len(data["Data Collector"]))):
        if isinstance(data["Data Collector"][i], DataCollector):
            data_speed = data["Data Collector"][i].get_model_vars_dataframe()['Avg_speed']
            data_cars = data["Data Collector"][i].get_model_vars_dataframe()['Cars_in_lane']
            tenproc = int(0.2 * (len(data_speed)))
            data['Total_Avg_speed'][i] = np.average(data_speed[tenproc:])
            data['Total_Cars_in_lane'][i] = np.average(data_cars[tenproc:])
            data['Variance_speed'][i] = np.var(data_speed[tenproc:])
            data['Variance_car'][i] = np.var(data_cars[tenproc:])
    
    
    data.to_csv('Sobol_result.csv', sep=',', index=False)
    
    print(data)
    
    Si_Speed = sobol.analyze(problem, data['Total_Avg_speed'].as_matrix(), print_to_console=False)
    print("\n")
    Si_Cars = sobol.analyze(problem, data['Total_Cars_in_lane'].as_matrix(), print_to_console=False)

[tpike3]

@edday1 to update this code to batch_run you can just pass parameters in via a dictionary.

Here is an overview form the tutorial, that will also run through each parameter of batch_run.

The amount of code should be significantly less as things like tqdm are already accounted for.

[edday1]

Thanks. I'll give it a go.

[edday1]

@tpike3 Problem is batch_run generates combinations of the parms given to it and saltelli.sample generates something similar. So I think i need to run batch_run a single time with each combination generated, which seems inefficient.