#coding=UTF-8
'''
Created on 2014年8月3日
@author: Administrator
'''
"""
NaiveBayes1 class
Chapter "Naive Bayes: a Primary Course"
"""
# imports
import numpy as np
# public symbols
__all__ = ['NaiveBayes1','self.pXgY_','self.pY_']
class NaiveBayes1(object):
"""
Naive Bayes class (1)
Attributes
----------
`pY_` : array_like, shape=(n_classes), dtype=float
pmf of a class
`pXgY_` : array_like, shape(n_features, n_classes, n_fvalues), dtype=float
pmf of feature values given a class
"""
def __init__(self):
self.pY_ = None
self.pXgY_ = None
def fit(self, X, y):
"""
Fitting model
Parameters
----------
X : array_like, shape=(n_samples, n_features), dtype=int
feature values of training samples
y : array_like, shape=(n_samples,n_classes), dtype=int
class labels of training samples
"""
# constants
n_samples = X.shape[0]
n_features = X.shape[1]
n_classes = 5
n_fvalues = 2
# check the size of y
if n_samples != y.shape[0]:
raise ValueError('Mismatched number of samples.')
# count up n[yi=y]
nY = np.zeros(n_classes, dtype=np.int)
for i in xrange(n_samples):
nY[y[i]] += 1
# calc pY_
self.pY_ = np.empty(n_classes, dtype=np.float)
for i in xrange(n_classes):
self.pY_[i] = nY[i] / np.float(n_samples)
# count up n[x_ij=xj, yi=y]
nXY = np.zeros((n_features, n_fvalues, n_classes), dtype=np.int)
for i in xrange(n_samples):
for j in xrange(n_features):
nXY[j, X[i, j], y[i]] += 1
# calc pXgY_
self.pXgY_ = np.empty((n_features, n_fvalues, n_classes),
dtype=np.float)
for j in xrange(n_features):
for xi in xrange(n_fvalues):
for yi in xrange(n_classes):
self.pXgY_[j, xi, yi] = nXY[j, xi, yi] / np.float(nY[yi])
def predict(self, X):
"""
Predict class
Parameters
----------
X : array_like, shape=(n_samples, n_features), dtype=int
feature values of unseen samples
Returns
-------
y : array_like, shape=(n_samples), dtype=int
predicted class labels
"""
# constants
n_samples = X.shape[0]
n_features = X.shape[1]
# memory for return values
y = np.empty(n_samples, dtype=np.int)
# for each feature in X
for i, xi in enumerate(X):
# calc joint probability
logpXY = np.log(self.pY_) + \
np.sum(np.log(self.pXgY_[np.arange(n_features), xi, :]),
axis=0)
print logpXY
# predict class
y[i] = np.argmax(logpXY)
return y
- 1
- 2
- 3
前往页