def softmax_regression(size,classes,lamda,data,labels,iters,validator,a=0.3):
#initialize theta
#TODO figure out why 0.005?
theta = 0.005 * np.random.rand(size*classes,)
#wrap initail parameters
grad_func = functools.partial(softmax_gradient,input_size=size,num_classes=classes,input_labels=labels,lmd=lamda)
#lbfgs
W = minfunc.lbfgs(data,theta,grad_func,validator,iters,alpha=a)
return W
def sparseAutoencoder(vSize,hSize,spars,lamda,b,a,data,iters=400):
#step1 initialize parameters
W1,W2,b1,b2 = initializeParameters(hSize,vSize)
theta = np.r_[W1.flatten(),W2.flatten(),b1.flatten(),b2.flatten()]
#step2 wrap grad function paramameters
grad_func = functools.partial(sparseAutoencoderGradient,visibleSize=vSize,hiddenSize=hSize,sparsityParam=spars,lmd=lamda,beta=b)
#step3 lbfgs
W = minfunc.lbfgs(data,theta,grad_func,alpha=a,max_iter=iters)
return W
def softmax_regression(size, classes, lamda, data, labels, iters, a=0.3):
#initialize theta
#TODO figure out why 0.005?
theta = 0.005 * np.random.rand(size * classes, )
#wrap initail parameters
grad_func = functools.partial(softmax_gradient,
input_size=size,
num_classes=classes,
input_labels=labels,
lmd=lamda)
#lbfgs
W = minfunc.lbfgs(data, theta, grad_func, iters, alpha=a)
return W
def sparseAutoencoder(vSize, hSize, spars, lamda, b, a, data, iters=400):
#step1 initialize parameters
W1, W2, b1, b2 = initializeParameters(hSize, vSize)
theta = np.r_[W1.flatten(), W2.flatten(), b1.flatten(), b2.flatten()]
#step2 wrap grad function paramameters
grad_func = functools.partial(sparseAutoencoderGradient,
visibleSize=vSize,
hiddenSize=hSize,
sparsityParam=spars,
lmd=lamda,
beta=b)
#step3 lbfgs
W = minfunc.lbfgs(data, theta, grad_func, alpha=a, max_iter=iters)
return W