我们从Python开源项目中，提取了以下10个代码示例，用于说明如何使用theano.tensor.nnet.conv()。
def __init__ (self, model_params, nkerns=[1,8,4], ckern=10, filter_sizes=[5,5,5,7]): """Initializes the architecture of the discriminator""" self.num_hid, num_dims, num_class, self.batch_size, self.num_channels = model_params self.D = int(np.sqrt(num_dims / self.num_channels)) numpy_rng=np.random.RandomState(1234) self.nkerns = np.asarray(nkerns) * ckern # of constant gen filters in first conv layer self.nkerns[0] = self.num_channels self.filter_sizes=filter_sizes num_convH = self.nkerns[-1]*filter_sizes[-1]*filter_sizes[-1] self.W = initialize_weight(num_convH, self.num_hid, 'W', numpy_rng, 'uniform') self.hbias = theano.shared(np.zeros((self.num_hid,), dtype=theano.config.floatX), name='hbias_enc') self.W_y = initialize_weight(self.num_hid, num_class, 'W_y', numpy_rng, 'uniform') self.L1 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[0], bnkern=self.nkerns[1] , bfilter_sz=filter_sizes[0], tfilter_sz=filter_sizes[1]) self.L2 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[1], bnkern=self.nkerns[2] , bfilter_sz=filter_sizes[1], tfilter_sz=filter_sizes[2]) self.num_classes = num_class self.params = [self.W_y, self.W, self.hbias] + self.L1.params + self.L2.params
def propagate(self, X, num_train=None, atype='relu'): """Propagate, return binary output of fake/real image""" image_shape0=[X.shape[0], self.num_channels, self.D, self.D] ConX = X.reshape(image_shape0) H0 = self.L1.conv(ConX, atype=atype) H1 = self.L2.conv(H0, atype=atype) H2 = self.L3.conv(H1, atype=atype) H3 = self.L4.conv(H2, atype=atype) H4 = self.L5.conv(H3, atype=atype) H4 = H4.flatten(2) H5 = activation_fn_th(T.dot(H4, self.W) + self.hbias, atype='tanh') y = T.nnet.sigmoid(T.dot(H5, self.W_y)) return y
def __init__ (self, model_params, nkerns=[1,8,4,2,1], ckern=172, filter_sizes=[5,5,5,5,4]): """Initializes the architecture of the discriminator""" self.num_hid, num_dims, num_class, self.batch_size, self.num_channels = model_params self.D = int(np.sqrt(num_dims / self.num_channels)) numpy_rng=np.random.RandomState(1234) self.nkerns = np.asarray(nkerns) * ckern # of constant gen filters in first conv layer self.nkerns[0] = self.num_channels self.filter_sizes=filter_sizes num_convH = self.nkerns[-1]*filter_sizes[-1]*filter_sizes[-1] self.W = initialize_weight(num_convH, self.num_hid, 'W', numpy_rng, 'uniform') self.hbias = theano.shared(np.zeros((self.num_hid,), dtype=theano.config.floatX), name='hbias') self.W_y = initialize_weight(self.num_hid, num_class, 'W_y', numpy_rng, 'uniform') self.L1 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[0], bnkern=self.nkerns[1] , bfilter_sz=filter_sizes[1], tfilter_sz=filter_sizes[0]) self.L2 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[1], bnkern=self.nkerns[2] , bfilter_sz=filter_sizes[2], tfilter_sz=filter_sizes[1]) self.L3 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[2], bnkern=self.nkerns[3] , bfilter_sz=filter_sizes[3], tfilter_sz=filter_sizes[2]) self.L4 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[3], bnkern=self.nkerns[4] , bfilter_sz=filter_sizes[4], tfilter_sz=filter_sizes[3]) self.num_classes = num_class self.params = [self.W_y, self.W] \ + self.L1.params + self.L2.params + self.L3.params + self.L4.params
def __init__(self, model_params, nkerns=[1,8,4,2], ckern=128, filter_sizes=[5,5,5,5,4]): """Initializes the architecture of the discriminator""" self.num_hid, num_dims, num_class, self.batch_size, self.num_channels = model_params self.D = int(np.sqrt(num_dims / self.num_channels)) numpy_rng = np.random.RandomState(1234) self.nkerns = np.asarray(nkerns) * ckern # of constant gen filters in first conv layer self.nkerns[0] = self.num_channels self.filter_sizes = filter_sizes num_convH = self.nkerns[-1]*filter_sizes[-1]*filter_sizes[-1] self.W = initialize_weight(num_convH, self.num_hid, 'W', numpy_rng, 'uniform') self.hbias = theano.shared(np.zeros((self.num_hid,), dtype=theano.config.floatX), name='hbias_enc') self.W_y = initialize_weight(self.num_hid, num_class, 'W_y', numpy_rng, 'uniform') self.L1 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[0], bnkern=self.nkerns[1] , bfilter_sz=filter_sizes[0], tfilter_sz=filter_sizes[1]) self.L2 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[1], bnkern=self.nkerns[2] , bfilter_sz=filter_sizes[1], tfilter_sz=filter_sizes[2]) self.L3 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[2], bnkern=self.nkerns[3] , bfilter_sz=filter_sizes[2], tfilter_sz=filter_sizes[3]) self.num_classes = num_class self.params = [self.W_y, self.W, self.hbias] + self.L1.params + self.L2.params + self.L3.params
def propagate(self, X, num_train=None, atype='relu'): """Propagate, return binary output of fake/real image""" image_shape0=[X.shape[0], self.num_channels, self.D, self.D] ConX = X.reshape(image_shape0) H0 = self.L1.conv(ConX, atype=atype) H1 = self.L2.conv(H0, atype=atype) H1 = H1.flatten(2) H2 = activation_fn_th(T.dot(H1, self.W) + self.hbias, atype='tanh') y = T.nnet.sigmoid(T.dot(H2, self.W_y)) return y
def __init__ (self, model_params, nkerns=[1,8,4,2,1,1], ckern=128*3, filter_sizes=[5,5,5,5,5,4]): """Initializes the architecture of the discriminator""" self.num_hid, num_dims, num_class, self.batch_size, self.num_channels = model_params self.D = int(np.sqrt(num_dims / self.num_channels)) numpy_rng=np.random.RandomState(1234) self.nkerns = np.asarray(nkerns) * ckern # of constant gen filters in first conv layer self.nkerns[0] = self.num_channels self.filter_sizes=filter_sizes num_convH = self.nkerns[-1]*filter_sizes[-1]*filter_sizes[-1] self.W = initialize_weight(num_convH, self.num_hid, 'W', numpy_rng, 'uniform') self.hbias = theano.shared(np.zeros((self.num_hid,), dtype=theano.config.floatX), name='hbias') self.W_y = initialize_weight(self.num_hid, num_class, 'W_y', numpy_rng, 'uniform') self.L1 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[0], bnkern=self.nkerns[1] , bfilter_sz=filter_sizes[1], tfilter_sz=filter_sizes[0]) self.L2 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[1], bnkern=self.nkerns[2] , bfilter_sz=filter_sizes[2], tfilter_sz=filter_sizes[1]) self.L3 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[2], bnkern=self.nkerns[3] , bfilter_sz=filter_sizes[3], tfilter_sz=filter_sizes[2]) self.L4 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[3], bnkern=self.nkerns[4] , bfilter_sz=filter_sizes[4], tfilter_sz=filter_sizes[3]) self.L5 = BN_Conv_layer(self.batch_size, numpy_rng, tnkern=self.nkerns[4], bnkern=self.nkerns[5] , bfilter_sz=filter_sizes[5], tfilter_sz=filter_sizes[4]) self.num_classes = num_class self.params = [self.W_y, self.W] \ + self.L1.params + self.L2.params + self.L3.params + self.L4.params + self.L5.params
def propagate(self, X, num_train=None, atype='relu'): """Propagate, return binary output of fake/real image""" image_shape0=[X.shape[0], self.num_channels, self.D, self.D] ConX = X.reshape(image_shape0) H0 = self.L1.conv(ConX, atype=atype) H1 = self.L2.conv(H0, atype=atype) H2 = self.L3.conv(H1, atype=atype) H2 = H2.flatten(2) H3 = activation_fn_th(T.dot(H2, self.W) + self.hbias, atype='tanh') y = T.nnet.sigmoid(T.dot(H3, self.W_y)) return y
def build_shared_network(self): """ This part contains trhe sharred params (conv layer) for both policy and value networks Returns the shared output """ #from lasagne.layers import Conv2DLayer l_in = lasagne.layers.InputLayer( shape=(self.batch_size, self.history_length, self.img_height, self.img_width) ) #l_in = lasagne.layers.ReshapeLayer(l_in, (1, self.history_length, self.img_height, self.img_width)) #l_conv1 = dnn.Conv2DDNNLayer( l_conv1 = lasagne.layers.Conv2DLayer( incoming=l_in, num_filters=16, filter_size=(8, 8), stride=(4, 4), nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), # Defaults to Glorot b=lasagne.init.Constant(.1) #dimshuffle=True ) #l1_out=l_conv1.get_output_shape_for((self.history_length, self.img_height , self.img_width)) #print "L1:", l1_out l_conv2 = lasagne.layers.Conv2DLayer( incoming=l_conv1, num_filters=32, filter_size=(4, 4), stride=(2, 2), nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), b=lasagne.init.Constant(.1) #dimshuffle=True ) #l2_out=l_conv2.get_output_shape_for(l1_out) #print "L2:", l2_out l_hidden1 = lasagne.layers.DenseLayer( incoming=l_conv2, num_units=256, nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), b=lasagne.init.Constant(.1) ) return l_hidden1
def build_shared_network(self): """ This part contains trhe sharred params (conv layer) for both policy and value networks Returns the shared output """ #from lasagne.layers import Conv2DLayer l_in = lasagne.layers.InputLayer( shape=(self.history_length, self.img_height, self.img_width) ) l_in = lasagne.layers.ReshapeLayer(l_in, (1, self.history_length, self.img_height, self.img_width)) #l_conv1 = dnn.Conv2DDNNLayer( l_conv1 = lasagne.layers.Conv2DLayer( incoming=l_in, num_filters=16, filter_size=(8, 8), stride=(4, 4), nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), # Defaults to Glorot b=lasagne.init.Constant(.1) #dimshuffle=True ) #l1_out=l_conv1.get_output_shape_for((self.history_length, self.img_height , self.img_width)) #print "L1:", l1_out l_conv2 = lasagne.layers.Conv2DLayer( incoming=l_conv1, num_filters=32, filter_size=(4, 4), stride=(2, 2), nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), b=lasagne.init.Constant(.1) #dimshuffle=True ) #l2_out=l_conv2.get_output_shape_for(l1_out) #print "L2:", l2_out l_hidden1 = lasagne.layers.DenseLayer( incoming=l_conv2, num_units=256, nonlinearity=lasagne.nonlinearities.rectify, W=lasagne.init.HeUniform(), b=lasagne.init.Constant(.1) ) return l_hidden1
