) # the data is presented as a vector of inputs with many exchangeable examples of this vector is_train = T.iscalar( 'is_train') # pseudo boolean for switching between training and prediction rng = numpy.random.RandomState(1234) # Architecture: input --> LSTM --> predict one-ahead lstm_1 = LSTM(rng, x, n_in=data_set_x.get_value(borrow=True).shape[1], n_out=n_hidden) output = LogisticRegression(input=lstm_1.output, n_in=n_hidden, n_out=data_set_x.get_value(borrow=True).shape[1]) ################################ # Objective function and GD ################################ print 'defining cost, parameters, and learning function...' # the cost we minimize during training is the negative log likelihood of # the model cost = T.mean(output.cross_entropy_binary(y)) #Defining params params = lstm_1.params + output.params
ahead = T.matrix('ahead') sent = T.matrix('sentence') phonemes = T.imatrix('phonemes') rng = numpy.random.RandomState(1234) init_reg = LinearRegression(x, 60, 30,True) lstm_1 = LSTM(rng,init_reg.E_y_given_x,30,lstm_1_hidden) lstm_2 = LSTM(rng,lstm_1.output,lstm_1_hidden,lstm_2_hidden) reg_input = lstm_2.output #need log_reg and cross covariate layers log_reg = LogisticRegression(reg_input,lstm_2_hidden, 41) #lin_reg = LinearRegression(reg_input,lstm_2_hidden,1,True) log_reg.reconstruct(log_reg.p_y_given_x) #lin_reg.reconstruct(lin_reg.E_y_given_x) #reconstructed_regressions = T.concatenate([log_reg.reconstructed_x,lin_reg.reconstructed_x],axis=1) # #reverse_layer = LinearRegression(reconstructed_regressions, 2*lstm_2_hidden, lstm_2_hidden,False) lstm_3 = LSTM(rng,log_reg.reconstructed_x,lstm_2_hidden,lstm_1_hidden) lstm_4 = LSTM(rng,lstm_3.output,lstm_1_hidden,30) init_reg.reconstruct(lstm_4.output)
ahead = T.matrix('ahead') sent = T.matrix('sentence') phonemes = T.imatrix('phonemes') rng = numpy.random.RandomState(1234) init_reg = LinearRegression(x, 1, 30, True) lstm_1 = LSTM(rng, init_reg.E_y_given_x, 30, lstm_1_hidden) lstm_2 = LSTM(rng, lstm_1.output, lstm_1_hidden, lstm_2_hidden) reg_input = lstm_2.output #need log_reg and cross covariate layers log_reg = LogisticRegression(reg_input, lstm_2_hidden, 41) #lin_reg = LinearRegression(reg_input,lstm_2_hidden,1,True) log_reg.reconstruct(log_reg.p_y_given_x) #lin_reg.reconstruct(lin_reg.E_y_given_x) #reconstructed_regressions = T.concatenate([log_reg.reconstructed_x,lin_reg.reconstructed_x],axis=1) # #reverse_layer = LinearRegression(reconstructed_regressions, 2*lstm_2_hidden, lstm_2_hidden,False) lstm_3 = LSTM(rng, log_reg.reconstructed_x, lstm_2_hidden, lstm_1_hidden) lstm_4 = LSTM(rng, lstm_3.output, lstm_1_hidden, 30) init_reg.reconstruct(lstm_4.output)
poolsize=(1, 1), dim2 = 1 ) layer3 = LeNetConvPoolLayer( rng, input=layer2.output, image_shape=(minibatch_size, layer2_filters, 15, 15), filter_shape=( layer3_filters, layer2_filters, 2, 2), poolsize=(1, 1), dim2 = 1 ) reg_input = layer3.output.flatten(2) log_reg = LogisticRegression(reg_input,15*15*layer3_filters, 41) lin_reg = LinearRegressionRandom(reg_input,15*15*layer3_filters,2,True) log_input = log_reg.p_y_given_x lin_input = lin_reg.E_y_given_x log_reg.reconstruct(log_input) lin_reg.reconstruct(lin_input) reconstructed_regressions = T.concatenate([log_reg.reconstructed_x,lin_reg.reconstructed_x],axis=1) reverse_layer = LinearRegression(reconstructed_regressions, 2*15*15*layer3_filters, 15*15*layer3_filters,False) reconstruct = reverse_layer.E_y_given_x.reshape((minibatch_size,layer3_filters,15,15))
2), poolsize=(1, 1), dim2=1) layer3 = LeNetConvPoolLayer(rng, input=layer2.output, image_shape=(minibatch_size, layer2_filters, 15, 15), filter_shape=(layer3_filters, layer2_filters, 2, 2), poolsize=(1, 1), dim2=1) reg_input = layer3.output.flatten(2) log_reg = LogisticRegression(reg_input, 15 * 15 * layer3_filters, 41) lin_reg = LinearRegressionRandom(reg_input, 15 * 15 * layer3_filters, 2, True) log_input = log_reg.p_y_given_x lin_input = lin_reg.E_y_given_x log_reg.reconstruct(log_input) lin_reg.reconstruct(lin_input) reconstructed_regressions = T.concatenate( [log_reg.reconstructed_x, lin_reg.reconstructed_x], axis=1) reverse_layer = LinearRegression(reconstructed_regressions, 2 * 15 * 15 * layer3_filters, 15 * 15 * layer3_filters, False)
# allocate symbolic variables for the data index = T.lscalar() # index to a [mini]batch x = T.matrix('x') # the data is presented as a vector of inputs with many exchangeable examples of this vector x = clip_gradient(x,1.0) y = T.matrix('y') # the data is presented as a vector of inputs with many exchangeable examples of this vector is_train = T.iscalar('is_train') # pseudo boolean for switching between training and prediction rng = numpy.random.RandomState(1234) # Architecture: input --> LSTM --> predict one-ahead lstm_1 = LSTM(rng, x, n_in=data_set_x.get_value(borrow=True).shape[1], n_out=n_hidden) output = LogisticRegression(input=lstm_1.output, n_in=n_hidden, n_out=data_set_x.get_value(borrow=True).shape[1]) ################################ # Objective function and GD ################################ print 'defining cost, parameters, and learning function...' # the cost we minimize during training is the negative log likelihood of # the model cost = T.mean(output.cross_entropy_binary(y)) #Defining params params = lstm_1.params + output.params