Python makeKPCAdata示例

示例#1

文件： dr.py 项目： kotarohara/ml2011fall

def pcaTest():
	"""
	Si = sqrtm( array([[3,2], [2,4]]) )
	x = dot( randn(1000, 2), Si )
	
	(P,Z,evals) = pca( x, 2 )
	print "Z:", Z
	print "evals:", evals
	x0 = dot(dot(x, Z[0,:]).reshape(1000,1), Z[0,:].reshape(1,2))
	x1 = dot(dot(x, Z[1,:]).reshape(1000,1), Z[1,:].reshape(1,2))
	
	plot(x[:,0], x[:,1], 'b.', x0[:,0], x0[:,1], 'r.', x1[:,0], x1[:,1], 'g.')
	input()
	"""

	#(X, Y) = datasets.loadDigits()
	#(P,Z,evals) = pca( X, 784 )
	
	"""
	esum = sum( evals )
	print esum
	print "evals", evals[0:3]
	evals = evals / esum
	print "evals", evals[0:3]
	print "evals", evals
	plot( evals, zeros([evals.shape[0]]), 'b.')
	input()
	"""
	#print cumsum( evals )

	"""acum = 0
	for i in xrange( len( evals ) ):
		acum = sum( evals[:i] )
		print acum
		if acum >= 0.90:
			print "i:", i
			break
			"""
	
	# print Z
	#x0 = dot(dot(X, Z[0,:]).reshape(1000,1), Z[0,:].reshape(1,2))
	#plot(X[:,0], X[:,1], 'b.', x0[:,0], x0[:,1], 'r.')
	#input()
	
	#drawDigits(Z[1:50], arange(50))
	#input()
	
	(a,b) = datasets.makeKPCAdata()
	
	x = vstack((a,b))
	(P,Z,evals) = pca(x,2)
	
	Pa = P[0:a.shape[0],:]
	Pb = P[a.shape[0]:-1,:]
	
	plot(Pa[:,0], randn(Pa.shape[0]), 'b.', Pb[:,0], randn(Pb.shape[0]), 'r.')
	input()
	
	return

示例#2

文件： startup.py 项目： afcarl/Unsupervised-Learning

#FOR WU3 Figure draw the top 50 eigvectors
util.drawDigits(Z[:, 0:49].T, arange(50))
#FOR WU3b
util.drawDigits(dot(P, Z.T), Y)
#FOR WU3c
(P, Z, evals) = dr.pca(X, 5)
util.drawDigits(dot(P, Z.T), Y)

# ---------- KPCA ---------- #
Si = util.sqrtm(array([[3, 2], [2, 4]]))
x = dot(randn(1000, 2), Si)
(P, alpha, evals) = dr.kpca(x, 2, kernel.linear)
evals
# do with data where vanilla PCA fails
(a, b) = datasets.makeKPCAdata()
plot(a[:, 0], a[:, 1], 'b.', b[:, 0], b[:, 1], 'r.')

x = vstack((a, b))
x_c = (x - mean(x)) / std(x)
(P, Z, evals) = dr.pca(x, 2)

Pa = P[0:a.shape[0], :]
Pb = P[a.shape[0]:, :]
plot(Pa[:, 0], randn(Pa.shape[0]), 'b.', Pb[:, 0], randn(Pb.shape[0]), 'r.')

# projection of the data with each eig vectors...
# x0 = dot(dot(x, Z[:, 0]).reshape(473,1), Z[:, 0].reshape(1,2))
# x1 = dot(dot(x, Z[:, 1]).reshape(473,1), Z[:, 1].reshape(1,2))
# pro0a = x0[0:a.shape[0],:]
# pro0b = x0[a.shape[0]:,:]

示例#3

文件： startup.py 项目： akanazawa/Unsupervised-Learning

#FOR WU3 Figure draw the top 50 eigvectors
util.drawDigits(Z[:,0:49].T,arange(50))
#FOR WU3b
util.drawDigits(dot(P,Z.T),Y)
#FOR WU3c
(P,Z,evals) = dr.pca(X, 5)
util.drawDigits(dot(P,Z.T),Y)

# ---------- KPCA ---------- #
Si = util.sqrtm(array([[3,2],[2,4]]))
x = dot(randn(1000,2), Si)
(P, alpha, evals) = dr.kpca(x, 2, kernel.linear)
evals
# do with data where vanilla PCA fails
(a,b) = datasets.makeKPCAdata()
plot(a[:,0], a[:,1], 'b.', b[:,0], b[:,1], 'r.')

x = vstack((a,b))
x_c = (x-mean(x))/std(x)
(P,Z,evals) = dr.pca(x, 2)

Pa = P[0:a.shape[0],:]
Pb = P[a.shape[0]:,:]
plot(Pa[:,0], randn(Pa.shape[0]), 'b.', Pb[:,0], randn(Pb.shape[0]), 'r.')

# projection of the data with each eig vectors...
# x0 = dot(dot(x, Z[:, 0]).reshape(473,1), Z[:, 0].reshape(1,2))
# x1 = dot(dot(x, Z[:, 1]).reshape(473,1), Z[:, 1].reshape(1,2))
# pro0a = x0[0:a.shape[0],:]
# pro0b = x0[a.shape[0]:,:]