causally.scm.context.SCMContext.measure_err_model

SCMContext.measure_err_model(gamma: float)

Make the assumption of model with measurement error.

Rather than observing perfectly measured random variables \(x_i\) (where \(i\) is the index of a node), in the dataset we observe \(\tilde{x}_i := x_i + \epsilon_i\), where \(\epsilon_i\) is a Gaussian random variable centered at zero, whose variance is parametrized by the inverse signal to noise ratio \({\gamma} := \frac{\operatorname{Var}(\operatorname{error})}{\operatorname{Var}(\operatorname{signal})}\).

Parameters:

gamma (Union[float, List[float]]) – The inverse signal to noise ratio \({\gamma} := \frac{\operatorname{Var}(\operatorname{error})}{\operatorname{Var}(\operatorname{signal})}\) parametrizing the variance of the measurement error proportionally to the variance of the signal. If a single float is provided, then gamma is the same for each node in the graph. Else, gamma is a vector of shape (num_nodes, ).