def __init__(self,
threshold,
min_entries=10,
metric='minkowski',
k=1,
n_jobs=1):
"""Initialize the metric
Args:
threshold (float): Maximum distance for a prediction to be "trustable"
min_entries (int): Minimum number of training entries before surrogate can be evaluated
metric (string): Distance metric to use
k (int): Number of nearest neighbors to consider
n_jobs (int): Number of threads to use when computing distances
"""
super().__init__()
# Make the featurizer
# TODO (wardlt): This code is duplicated in the inference engine. Maybe we should let "featurizer" be a param
cm = CoulombMatrix(flatten=True)
cm.set_n_jobs(1)
self.cm = Pipeline([('featurizer', cm), ('scaler', RobustScaler())])
# Save the other things
self.threshold = threshold
self.min_entries = min_entries
self.metric = metric
self.k = k
self.n_jobs = n_jobs
self.nn_ = None
def __init__(self, n_neighbors: int = 5):
"""
Args:
n_neighbors (int): Number of neighboring points to use for the NN model
"""
cm = CoulombMatrix(flatten=True)
cm.set_n_jobs(1)
model = Pipeline([('featurizer', cm), ('scaler', RobustScaler()),
('model', KNeighborsRegressor(n_neighbors))])
super().__init__(model)