Skip to content

Instantly share code, notes, and snippets.

@johnsmith17th
Created May 27, 2013 03:16
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save johnsmith17th/5655013 to your computer and use it in GitHub Desktop.
Save johnsmith17th/5655013 to your computer and use it in GitHub Desktop.
Adaboost
#!/usr/bin/env python
#coding=utf-8
import sys, math, random
import numpy
import operator
from abstract_classifier import Classifier
def shuffle(ary):
a = len(ary)
b = a-1
for d in range(b,0,-1):
e = random.randint(0,d)
if e == d:
continue
ary[d],ary[e] = ary[e],ary[d]
return ary
class AdaBoost(Classifier):
def __init__(self, weak_classifier_type):
Classifier.__init__(self)
self.WeakClassifierType = weak_classifier_type
def train(self, T, k = 1):
X = self.X
Y = numpy.array(self.Y)
N = len(self.Y)
w = (1.0/N)*numpy.ones(N)
self.weak_classifier_ensemble = []
self.alpha = []
for t in range(T):
sys.stdout.write('.')
weak_learner = self.WeakClassifierType()
weak_learner.set_training_sample(X,Y)
weak_learner.weights = w
weak_learner.train()
Y_pred = weak_learner.predict(X)
# (Y=-1, Y_pred=1) False Positive
# (Y=1, Y_pred=-1) Missing should be assigned more weights
#ww = numpy.log(k)*(numpy.exp( (Y-Y_pred)>1 ) - 1)/(numpy.exp(1)-1) + 1
e = sum(0.5*w*abs((Y-Y_pred)))/sum(w)
#e = sum(0.5*w*abs(Y-Y_pred))
ee = (1-e)/(e*1.0)
alpha = 0.5*math.log(ee+0.00001)
w *= numpy.exp(-alpha*Y*Y_pred) #*ww) # increase weights for wrongly classified
w /= sum(w)
self.weak_classifier_ensemble.append(weak_learner)
self.alpha.append(alpha)
print "\n"
self.T = T
def predict(self,X):
X = numpy.array(X)
N, d = X.shape
Y = numpy.zeros(N)
for t in range(self.T):
#sys.stdout.write('.')
weak_learner = self.weak_classifier_ensemble[t]
#print Y.shape, self.alpha[t], weak_learner.predict(X).shape
Y += self.alpha[t]*weak_learner.predict(X)
return Y
def test_on_training_set(self, X, Y, T):
self.set_training_sample(X,Y)
self.train(T)
o = self.predict(X)
return o
def measure_accuracy(self, Y, o, threshold=0):
oo = o.copy()
oo[numpy.where(o>threshold)[0]] = 1
oo[numpy.where(o<threshold)[0]] = -1
d = (oo - Y)
return len(d[numpy.where(d==0)[0]])*1.0/len(Y)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment