def oneVsall(X,y,num_labels,lamdba):
from batch_gradient_update import batch_gradient_update
import numpy as np
m,n=X.shape
all_theta=np.zeros([num_labels,n+1])
X0=np.ones([m,1])
X=np.concatenate((X0,X),axis=1)
for c in range(num_labels):
print 'c:',c
initial_theta=np.zeros([n+1,1])
#print y%10
theta=batch_gradient_update(initial_theta,X,np.array(((y%10)==c),dtype=int))
all_theta[c,:]=np.reshape(theta,[n+1,])
return all_theta
def oneVsall(X, y, num_labels, lamdba):
from batch_gradient_update import batch_gradient_update
import numpy as np
m, n = X.shape
all_theta = np.zeros([num_labels, n + 1])
X0 = np.ones([m, 1])
X = np.concatenate((X0, X), axis=1)
for c in range(num_labels):
print 'c:', c
initial_theta = np.zeros([n + 1, 1])
#print y%10
theta = batch_gradient_update(initial_theta, X,
np.array(((y % 10) == c), dtype=int))
all_theta[c, :] = np.reshape(theta, [
n + 1,
])
return all_theta
X, y = make_blobs(n_samples=400, centers=2, random_state=0, cluster_std=1) # after featureNormalize it accuarcy could get 89% X,X_mu,X_sigma=featureNormalize(X) #plot_data(X,y) y=np.reshape(y,(y.size,1)) m,n=X.shape X=np.concatenate((np.ones([len(X[:,0]),1]),X),axis=1) initial_theta=np.zeros([n+1,1]) #initial_theta=np.array([1,1,1]) # test is the cost_function ok? cost,grad=cost_function(initial_theta,X,y) # batch_gradient_update error!!! wrong theta theta=batch_gradient_update(initial_theta,X,y) print theta prob=sigmoid_function(np.dot(X,theta)) print prob prob[prob>0.5]=1.0 prob[prob<0.5]=0.0 print prob y=np.reshape(y,prob.shape) print "accuracy:",tuple(1-sum(abs(prob-y))/100)
import numpy as np
from sklearn.linear_model import SGDClassifier
from sklearn.datasets.samples_generator import make_blobs
import scipy as sp
#X,y=read_data("ex2data1.txt")
X, y = make_blobs(n_samples=400, centers=2, random_state=0, cluster_std=1)
# after featureNormalize it accuarcy could get 89%
X, X_mu, X_sigma = featureNormalize(X)
#plot_data(X,y)
y = np.reshape(y, (y.size, 1))
m, n = X.shape
X = np.concatenate((np.ones([len(X[:, 0]), 1]), X), axis=1)
initial_theta = np.zeros([n + 1, 1])
#initial_theta=np.array([1,1,1])
# test is the cost_function ok?
cost, grad = cost_function(initial_theta, X, y)
# batch_gradient_update error!!! wrong theta
theta = batch_gradient_update(initial_theta, X, y)
print theta
prob = sigmoid_function(np.dot(X, theta))
print prob
prob[prob > 0.5] = 1.0
prob[prob < 0.5] = 0.0
print prob
y = np.reshape(y, prob.shape)
print "accuracy:", tuple(1 - sum(abs(prob - y)) / 100)
