Sampling from normal distribution

[mikhailsirenko]

Hi team,

Imagine the following setup. My model has a float parameter, and I have a prior that it forms a normal distribution with loc=0 and scale=1. Thus, I would like to sample it in my experiments. What are my options to sample it? Shall I specify it at the level of the RealParameter class? I see there the following:

self.dist = sp.stats.uniform(lower_bound, upper_bound - lower_bound) # @UndefinedVariable

However, I don't see an option to pass an arbitrary distribution yet. Also, the names lower_bound and upper_bound are not perfect even if there would be an option to pass, e.g., dist = sp.stats.norm.

In addition, how should this (dist of RealParameter) interact with the Sampler class? The sample function of, e.g., LHSSampler, does accept the distribution argument. However, I believe that distribution here is something else. Also, when LHSSampler does not accept arguments and perform_experiments throws an error when passing LHSSampler().sample(stats.norm(loc=0, scale=1), 1000)

Did I overlook something, or is there an example of doing this?

(Attaching an example code just in case)

from scipy import stats
from ema_workbench.em_framework.samplers import LHSSampler, AbstractSampler, UniformLHSSampler, MonteCarloSampler
from ema_workbench import Model, ScalarOutcome, RealParameter, perform_experiments
import matplotlib.pyplot as plt

# Define a simple model
def simple_model(x, y):
    return {'result': x + y}


# Create an EMA model
my_model = Model(name="SimpleModel", function=simple_model)

# Define the uncertainties
my_model.uncertainties = [RealParameter("x", lower_bound=0, upper_bound=10),
                          RealParameter("y", lower_bound=100, upper_bound=200)]

# Define the outcomes
my_model.outcomes = [ScalarOutcome("result")]

# Define the sampler
sampler = LHSSampler().sample(stats.norm(loc=0, scale=1), 1000)

# Perform experiments
results = perform_experiments(my_model, 1000, uncertainty_sampling=sampler)

[quaquel]

You can use RealParameter.from_dist. See Eigenraam example

[mikhailsirenko]

Awesome! Solved!