knn.py

# coding:utf-8

fromcollectionsimportCounter

importnumpyasnp
fromscipy.spatial.distanceimporteuclidean

frommla.baseimportBaseEstimator


classKNNBase(BaseEstimator):
def__init__(self, k=5, distance_func=euclidean):
"""Base class for Nearest neighbors classifier and regressor.

 Parameters
 ----------
 k : int, default 5
 The number of neighbors to take into account. If 0, all the
 training examples are used.
 distance_func : function, default euclidean distance
 A distance function taking two arguments. Any function from
 scipy.spatial.distance will do.
 """

self.k=Noneifk==0elsek# l[:None] returns the whole list
self.distance_func=distance_func

defaggregate(self, neighbors_targets):
raiseNotImplementedError()

def_predict(self, X=None):
predictions= [self._predict_x(x) forxinX]

returnnp.array(predictions)

def_predict_x(self, x):
"""Predict the label of a single instance x."""

# compute distances between x and all examples in the training set.
distances= (self.distance_func(x, example) forexampleinself.X)

# Sort all examples by their distance to x and keep their target value.
neighbors=sorted(((dist, target) for (dist, target) inzip(distances, self.y)), key=lambdax: x[0])

# Get targets of the k-nn and aggregate them (most common one or
# average).
neighbors_targets= [targetfor (_, target) inneighbors[: self.k]]

returnself.aggregate(neighbors_targets)


classKNNClassifier(KNNBase):
"""Nearest neighbors classifier.

 Note: if there is a tie for the most common label among the neighbors, then
 the predicted label is arbitrary."""

defaggregate(self, neighbors_targets):
"""Return the most common target label."""

most_common_label=Counter(neighbors_targets).most_common(1)[0][0]
returnmost_common_label


classKNNRegressor(KNNBase):
"""Nearest neighbors regressor."""

defaggregate(self, neighbors_targets):
"""Return the mean of all targets."""

returnnp.mean(neighbors_targets)