def show(self): """ .. todo:: WRITEME """ #image.imview_async(self.image) show(self.image)
def explore(X,y): for i in xrange(X.shape[0]): patch = X[i,:].reshape(28,28) print y[i] show(patch) x = raw_input('waiting...') if x == 'n': return if x == 'q': quit(-1)
def show(self): #image.imview_async(self.image) show(self.image)
r = 6 c = 6 dataset = CIFAR10(which_set = 'train', one_hot = True, gcn = 55.) ten4 = dataset.get_batch_topo(m) from pylearn2.utils import sharedX ten4th = sharedX(ten4) X = cifar10neighbs(ten4, (r,c)) from theano import function X = function([],X)() print X.shape from pylearn2.gui.patch_viewer import make_viewer from pylearn2.utils.image import show stride = (32-r+1)*(32-c+1) for i in xrange(m): ten4v =ten4[i,:,:,:] ten4v -= ten4v.min() ten4v /= ten4v.max() show(ten4v) patch_viewer = make_viewer(X[i*stride:(i+1)*stride], is_color= True) patch_viewer.show() print 'waiting...' x = raw_input()
def show(self): show(self.image)
H[i*batch_size:(i+1)*batch_size,:] = f(X) h = H.mean(axis=0) proj = np.dot(H,h) #sort examples by projection along the mean vector ranking = sorted(zip(proj,range(proj.shape[0]))) new_H = H.copy() for i, t in enumerate(ranking): new_H[i,:] = H[t[1],:] H = new_H #sort units by mean activation ranking = sorted(zip(h,range(h.shape[0]))) new_H = H.copy() for i, t in enumerate(ranking): new_H[:,i] = H[:,t[1]] H = new_H from pylearn2.utils import image image.show(H)
import numpy as np from pylearn2.utils import image imbase = '/data/lisatmp/goodfeli/esp/final_images' ims = sorted(os.listdir(imbase)) for label, im in zip(labels, ims): stem = label.split('.')[0] assert stem in im img = image.load(imbase + '/' + im) image.show(img) full_label_path = labels_dir + '/' + label print 'True labels:' fd = open(full_label_path,'r') print fd.read() fd.close() full_l2_path = l2_path + '/' + label.split('.')[0] + '.npy' l2 = np.load(full_l2_path) y = f(l2) print 'Predicted labels: ' print y.shape
f = function([X], features) for i in xrange(num_batches): X = dataset.get_batch_design(batch_size) H[i * batch_size:(i + 1) * batch_size, :] = f(X) h = H.mean(axis=0) proj = np.dot(H, h) #sort examples by projection along the mean vector ranking = sorted(zip(proj, range(proj.shape[0]))) new_H = H.copy() for i, t in enumerate(ranking): new_H[i, :] = H[t[1], :] H = new_H #sort units by mean activation ranking = sorted(zip(h, range(h.shape[0]))) new_H = H.copy() for i, t in enumerate(ranking): new_H[:, i] = H[:, t[1]] H = new_H from pylearn2.utils import image image.show(H)
l2_path = '/data/lisatmp/goodfeli/esp/final_l2' import numpy as np from pylearn2.utils import image imbase = '/data/lisatmp/goodfeli/esp/final_images' ims = sorted(os.listdir(imbase)) for label, im in zip(labels, ims): stem = label.split('.')[0] assert stem in im img = image.load(imbase + '/' + im) image.show(img) full_label_path = labels_dir + '/' + label print 'True labels:' fd = open(full_label_path, 'r') print fd.read() fd.close() full_l2_path = l2_path + '/' + label.split('.')[0] + '.npy' l2 = np.load(full_l2_path) y = f(l2) print 'Predicted labels: ' print y.shape
r = 6 c = 6 dataset = CIFAR10(which_set='train', one_hot=True, gcn=55.) ten4 = dataset.get_batch_topo(m) from pylearn2.utils import sharedX ten4th = sharedX(ten4) X = cifar10neighbs(ten4, (r, c)) from theano import function X = function([], X)() print X.shape from pylearn2.gui.patch_viewer import make_viewer from pylearn2.utils.image import show stride = (32 - r + 1) * (32 - c + 1) for i in xrange(m): ten4v = ten4[i, :, :, :] ten4v -= ten4v.min() ten4v /= ten4v.max() show(ten4v) patch_viewer = make_viewer(X[i * stride:(i + 1) * stride], is_color=True) patch_viewer.show() print 'waiting...' x = raw_input()
grad = T.grad(g, X) step_X = X + alpha * grad norm = T.sqrt(T.sqr(X).sum()) renormed_X = step_X #/ norm from theano import function f = function([], [g, norm], updates={X: renormed_X}) while True: a, b = f() print a, ' ', b if a > .75: break from pylearn2.utils import image from pylearn2.config import yaml_parse dataset = yaml_parse.load(dataset_yaml_src) X = dataset.get_topological_view(X.get_value()) X /= np.abs(X).max() image.show(X[0, ...])
act = p[0,filter_idx,i,j] obj = - act + norm_penalty * T.square(X).sum() assert obj.ndim == 0 optimizer = BatchGradientDescent(objective = obj, params = [X], inputs = None, param_constrainers = None, max_iter = 1000, verbose = True, tol = None, init_alpha = (.001, .005, .01, .05, .1)) optimizer.minimize() img = X.get_value()[0,:,:,:] print 'max mag: ',np.abs(img).max() print 'norm: ',np.square(img).sum() print 'min: ',img.min() print 'max: ',img.max() img /= np.abs(img).max() img *= .5 img += 1 show(img)
for j in xrange(mr): for k in xrange(mc): r = 0. g = 0. b = 0. r_count = 0 g_count = 0 b_count = 0 for l in xrange(ch): elem = convmap[i,j,k,l] if elem < 32: r_count += 1 r += float(elem)/31. if elem >= 32 and elem < 64: g_count += 1 g += float(elem-32)/31. if elem >= 64: b_count += 1 b += float(elem-64)/float(m) decoded_map[i,j,k,:] = np.asarray( [ r/(1e-12+float(r_count)), g/(1e-12+float(g_count)), b/(1e-12+float(b_count))]) for i in xrange(m): show(decoded_img[i,:,:,:]) show(decoded_map[i,:,:,:]) x = raw_input()
norm = T.sqrt(T.sqr(X).sum()) renormed_X = step_X #/ norm from theano import function f = function([],[g,norm],updates = { X : renormed_X }) while True: a,b = f() print a,' ',b if a > .75: break from pylearn2.utils import image from pylearn2.config import yaml_parse dataset = yaml_parse.load( dataset_yaml_src ) X = dataset.get_topological_view( X.get_value()) X /= np.abs(X).max() image.show(X[0,...])
from pylearn2.optimization.batch_gradient_descent import BatchGradientDescent bgd = BatchGradientDescent(objective=-neuron, params=[X], inputs=None, max_iter=100, lr_scalers=None, verbose=3, tol=None, init_alpha=None, min_init_alpha=1e-3, reset_alpha=True, conjugate=True, gradients=None, gradient_updates=None, accumulate=False, theano_function_mode=None, param_constrainers=None) bgd.minimize() X = normed.eval()[:,:,:,0].transpose(1,2,0) import numpy as np X /= np.abs(X).max() print (X.min(), X.max()) from pylearn2.utils.image import show show(X)