# set up the model dnn_shared = None shared_layers = [] hidden_layers = shared_layers_sizes + task_specific_sizes[n] # use the first networks shared layers dawg if n > 0: dnn_shared = dnn_array[0] shared_layers = [m for m in xrange(shared_layers_num)] # create the network for the task # you can change the input dropout factor and the general dropout factor # look at the DNNDropout class if shareLayers: dnn = DNNDropout(np_rng=np_rng, theano_rng=theano_rng, hidden_layers_sizes=hidden_layers, n_ins=input_size, n_outs=output_size, input_dropout_factor=0.0, dropout_factor=0.0, dnn_shared=dnn_shared, shared_layers=shared_layers) else: dnn = DNNDropout(np_rng=np_rng, theano_rng=theano_rng, hidden_layers_sizes=hidden_layers, n_ins=input_size, n_outs=output_size, input_dropout_factor=0.1, dropout_factor=0.5) # add dnn and the functions to the list dnn_array.append(dnn) # # consider the tasks which have nonzero learning rate # active_tasks = [n for n in xrange(num_tasks)] log('> ... bootstrapping all tasks datasets and building the functions') # keep track of the training error in order to create the train/validation # curve
# set up the model dnn_shared = None shared_layers = [] hidden_layers = shared_layers_sizes + task_specific_sizes[n] # use the first networks shared layers dawg if n > 0: dnn_shared = dnn_array[0] shared_layers = [m for m in xrange(shared_layers_num)] # create the network for the task # you can change the input dropout factor and the general dropout factor # look at the DNNDropout class if shareLayers: dnn = DNNDropout(np_rng=np_rng, theano_rng=theano_rng, hidden_layers_sizes=hidden_layers, n_ins=input_size, n_outs=output_size, input_dropout_factor=0.0, dropout_factor=0.0, dnn_shared=dnn_shared, shared_layers=shared_layers) else: dnn = DNNDropout(np_rng=np_rng, theano_rng=theano_rng, hidden_layers_sizes=hidden_layers, n_ins=input_size, n_outs=output_size, input_dropout_factor=0.1, dropout_factor=0.5) # add dnn and the functions to the list dnn_array.append(dnn) # # consider the tasks which have nonzero learning rate # active_tasks = [n for n in xrange(num_tasks)] test_in, test_out, test_tasks = get_test_data() complete = np.hstack((test_tasks.reshape((-1,1)),test_in,test_out.reshape((-1,1)) )) total = 0.0 testin = []