fknn.py

import operator

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from sklearn.metrics import accuracy_score
from sklearn.base import BaseEstimator, ClassifierMixin


class FuzzyKNN(BaseEstimator, ClassifierMixin):
	def __init__(self, k=3, plot=False):
		self.k = k
		self.plot = plot


	def fit(self, X, y=None):
		self._check_params(X,y)
		self.X = X
		self.y = y

		self.xdim = len(self.X[0])
		self.n = len(y)

		classes = list(set(y))
		classes.sort()
		self.classes = classes

		self.df = pd.DataFrame(self.X)
		self.df['y'] = self.y

		self.memberships = self._compute_memberships()

		self.df['membership'] = self.memberships

		self.fitted_ = True
		return self


	def predict(self, X):
		if self.fitted_ == None:
			raise Exception('predict() called before fit()')
		else:
			m = 2
			y_pred = []

			for x in X:
				neighbors = self._find_k_nearest_neighbors(pd.DataFrame.copy(self.df), x)

				votes = {}
				for c in self.classes:
					den = 0
					for n in range(self.k):
						dist = np.linalg.norm(x - neighbors.iloc[n,0:self.xdim])
						den += 1 / (dist ** (2 / (m-1)))

					neighbors_votes = []
					for n in range(self.k):
						dist = np.linalg.norm(x - neighbors.iloc[n,0:self.xdim])
						num = (neighbors.iloc[n].membership[c]) / (dist ** (2 / (m-1)))

						vote = num/den
						neighbors_votes.append(vote)
					votes[c] = np.sum(neighbors_votes)

				pred = max(votes.items(), key=operator.itemgetter(1))[0]
				y_pred.append((pred, votes))

			return y_pred


	def score(self, X, y):
		if self.fitted_ == None:
			raise Exception('score() called before fit()')
		else:
			predictions = self.predict(X)
			y_pred = [t[0] for t in predictions]
			confidences = [t[1] for t in predictions]

			return accuracy_score(y_pred=y_pred, y_true=y)


	def _find_k_nearest_neighbors(self, df, x):
		X = df.iloc[:,0:self.xdim].values

		df['distances'] = [np.linalg.norm(X[i] - x) for i in range(self.n)]

		df.sort_values(by='distances', ascending=True, inplace=True)
		neighbors = df.iloc[0:self.k]

		return neighbors


	def _get_counts(self, neighbors):
		groups = neighbors.groupby('y')
		counts = {group[1]['y'].iloc[0]:group[1].count()[0] for group in groups}

		return counts


	def _compute_memberships(self):
		memberships = []
		for i in range(self.n):
			x = self.X[i]
			y = self.y[i]

			neighbors = self._find_k_nearest_neighbors(pd.DataFrame.copy(self.df), x)
			counts = self._get_counts(neighbors)

			membership = dict()
			for c in self.classes:
				try:
					uci = 0.49 * (counts[c] / self.k)
					if c == y:
						uci += 0.51
					membership[c] = uci
				except:
					membership[c] = 0

			memberships.append(membership)
		return memberships


	def _check_params(self, X, y):
		if type(self.k) != int:
			raise Exception('"k" should have type int')
		if self.k >= len(y):
			raise Exception('"k" should be less than no of feature sets')
		if self.k % 2 == 0:
			raise Exception('"k" should be odd')

		if type(self.plot) != bool:
			raise Exception('"plot" should have type bool')