wehub-resource-sync
75 lines
2.2 KiB
Python
75 lines
2.2 KiB
Python
# coding:utf-8
|
|
|
|
from collections import Counter
|
|
|
|
import numpy as np
|
|
from scipy.spatial.distance import euclidean
|
|
|
|
from mla.base import BaseEstimator
|
|
|
|
|
|
class KNNBase(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 = None if k == 0 else k # l[:None] returns the whole list
|
|
self.distance_func = distance_func
|
|
|
|
def aggregate(self, neighbors_targets):
|
|
raise NotImplementedError()
|
|
|
|
def _predict(self, X=None):
|
|
predictions = [self._predict_x(x) for x in X]
|
|
|
|
return np.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) for example in self.X)
|
|
|
|
# Sort all examples by their distance to x and keep their target value.
|
|
neighbors = sorted(
|
|
((dist, target) for (dist, target) in zip(distances, self.y)),
|
|
key=lambda x: x[0],
|
|
)
|
|
|
|
# Get targets of the k-nn and aggregate them (most common one or
|
|
# average).
|
|
neighbors_targets = [target for (_, target) in neighbors[: self.k]]
|
|
|
|
return self.aggregate(neighbors_targets)
|
|
|
|
|
|
class KNNClassifier(KNNBase):
|
|
"""Nearest neighbors classifier.
|
|
|
|
Note: if there is a tie for the most common label among the neighbors, then
|
|
the predicted label is arbitrary."""
|
|
|
|
def aggregate(self, neighbors_targets):
|
|
"""Return the most common target label."""
|
|
|
|
most_common_label = Counter(neighbors_targets).most_common(1)[0][0]
|
|
return most_common_label
|
|
|
|
|
|
class KNNRegressor(KNNBase):
|
|
"""Nearest neighbors regressor."""
|
|
|
|
def aggregate(self, neighbors_targets):
|
|
"""Return the mean of all targets."""
|
|
|
|
return np.mean(neighbors_targets)
|