dagma.linear.DagmaLinear¶

class dagma.linear.DagmaLinear(loss_type: str, verbose: bool = False, dtype: type = np.float64)¶

A Python object that contains the implementation of DAGMA for linear models using numpy and scipy.

Parameters:¶

loss_type : str¶

One of [“l2”, “logistic”]. l2 refers to the least squares loss, while logistic refers to the logistic loss. For continuous data: use l2. For discrete 0/1 data: use logistic.

verbose : bool, optional¶

If true, the loss/score and h values will print to stdout every checkpoint iterations, as defined in fit(). Defaults to False.

dtype : type, optional¶

Defines the float precision, for large number of nodes it is recommened to use np.float64. Defaults to np.float64.

Methods¶

_score(→ Tuple[float, numpy.ndarray])	Evaluate value and gradient of the score function.
_h(→ Tuple[float, numpy.ndarray])	Evaluate value and gradient of the logdet acyclicity constraint.
_func(→ Tuple[float, numpy.ndarray])	Evaluate value of the penalized objective function.
_adam_update(→ numpy.ndarray)	Performs one update of Adam.
minimize(→ Tuple[numpy.ndarray, bool])	Solves the optimization problem:
fit(→ numpy.ndarray)	Runs the DAGMA algorithm and returns a weighted adjacency matrix.