|
| 1 | +# -*- coding: utf-8 -*- |
| 2 | +""" |
| 3 | +Created on Sun Jun 21 14:06:04 2015 |
| 4 | + |
| 5 | +@author: Pavitrakumar |
| 6 | +Credits: Jason Brownlee[Machinelearningmastery.com] |
| 7 | +""" |
| 8 | + |
| 9 | +from __future__ import division |
| 10 | +import numpy as np |
| 11 | +from sklearn import datasets |
| 12 | +from sklearn import cross_validation |
| 13 | +from sklearn.metrics import mean_squared_error |
| 14 | +import math |
| 15 | +import operator |
| 16 | + |
| 17 | +""" |
| 18 | +Euclidean distance measure: This is defined as the square root of the sum of the |
| 19 | +squared differences between the two arrays of numbers |
| 20 | +""" |
| 21 | + |
| 22 | +def euclideanDistance(instance1, instance2, no_of_features): |
| 23 | + distance = 0 |
| 24 | + for x in range(no_of_features): |
| 25 | + distance += pow((instance1[x] - instance2[x]), 2) |
| 26 | + return math.sqrt(distance) |
| 27 | + |
| 28 | +""" |
| 29 | +getNeighbors function returns k most similar neighbors from the training set |
| 30 | +for a given test instance (using the already defined euclideanDistance function) |
| 31 | +""" |
| 32 | +def getNeighbors(X_train,y_train, test_instance, k): |
| 33 | + # getting the k-nearest neighbors of the data point testInsatance |
| 34 | + distances = [] |
| 35 | + no_of_features = len(test_instance) |
| 36 | + for x,y in zip(X_train,y_train): |
| 37 | + # we are finding distance from each training example to out testInstance data point |
| 38 | + # and storing it as a list of pairs i.e (ith training example's response,distance to our instance data point) |
| 39 | + dist = euclideanDistance(test_instance, x, no_of_features) |
| 40 | + distances.append((y, dist)) |
| 41 | + distances.sort(key=operator.itemgetter(1)) |
| 42 | + #sorting the list by the 2nd element in each pair - sorting by distance |
| 43 | + #extracting the top k elements from the sorted list |
| 44 | + #we only need the response |
| 45 | + neighbors = [response for (response,distance) in distances] |
| 46 | + neighbors = neighbors[0:k] |
| 47 | + return neighbors |
| 48 | +""" |
| 49 | +getReponse just returns the most commonly occuring class in the given set of neighbors |
| 50 | +""" |
| 51 | +def getResponse(neighbors): |
| 52 | + # neighbors is a vector of length k |
| 53 | + # now, all we need to do is to find the most occuring class |
| 54 | + counts = np.bincount(neighbors) |
| 55 | + max_count = np.argmax(counts) |
| 56 | + return max_count |
| 57 | + |
| 58 | +def predict(X_test,X_train,y_train,k = 5): |
| 59 | + predicted = [] |
| 60 | + for each_test_instance in X_test: |
| 61 | + neighbors = getNeighbors(X_train,y_train,each_test_instance,k) |
| 62 | + predicted.append(getResponse(neighbors)) |
| 63 | + return predicted |
| 64 | + |
| 65 | + |
| 66 | +""" |
| 67 | +testing using IRIS data set |
| 68 | +""" |
| 69 | + |
| 70 | +iris = datasets.load_iris() |
| 71 | +X = iris.data |
| 72 | +y = iris.target |
| 73 | + |
| 74 | +X_train, X_test, y_train, y_test = cross_validation.train_test_split(X, y, test_size=0.8) |
| 75 | + |
| 76 | + |
| 77 | +pred = predict(X_test,X_train,y_train) |
| 78 | + |
| 79 | +from sklearn.metrics import accuracy_score |
| 80 | + |
| 81 | +print accuracy_score(y_test,pred) |
0 commit comments