__author__ = 'iankuoli' import math import _pickle import numpy import theano import theano.tensor as T dataset='/Volumes/My Book/Downloads/MLDS_HW1_RELEASE_v1' partiotnguide = 'sample1' batch_size = 20 (train_set_x, train_set_y), (test_set_x, test_set_y), label_48to39, label2index, index2label = LoadTIMIT.load_data(dataset, partiotnguide) n_test_batches = math.floor(test_set_x.get_value(borrow=True).shape[0] / batch_size) f = open('model_8layer_lr0.0005_0.376.pkl', 'rb') load_dnn = _pickle.load(f, encoding='latin1') f.close() index = T.lscalar() # index to a [mini]batch x = T.matrix('x') # the feature vectors of training data y = T.ivector('y') # the labels rng = numpy.random.RandomState(1234) dnn = DNN.DNN(rng=rng, inputdata=x, num_in=train_set_x.container.data.shape[1], num_hidden=500, num_out=len(label2index), num_layer=len(load_dnn.layers)) print(len(load_dnn.layers))
def training(lr=0.0005, L1_reg=0.00, L2_reg=0.0000, n_epochs=200, dataset='/Volumes/My Book/Downloads/MLDS_HW1_RELEASE_v1', partiotnguide='sample1', batch_size=20, n_hidden=500, n_layer=8): (train_set_x, train_set_y), (test_set_x, test_set_y), label_48to39, label2index, index2label = LoadTIMIT.load_data(dataset, partiotnguide) # compute number of minibatches for training, validation and testing n_train_batches = math.floor(train_set_x.get_value(borrow=True).shape[0] / batch_size) n_test_batches = math.floor(test_set_x.get_value(borrow=True).shape[0] / batch_size) index = T.lscalar() # index to a [mini]batch x = T.matrix('x') # the feature vectors of training data y = T.ivector('y') # the labels rng = numpy.random.RandomState(1234) dnn = DNN.DNN(rng=rng, inputdata=x, num_in=train_set_x.container.data.shape[1], num_hidden=n_hidden, num_out=len(label2index), num_layer=n_layer) cost = (dnn.negative_log_likelihood(y) + L1_reg * dnn.L1 + L2_reg * dnn.L2_sqr) #cost = (dnn.l2_norm(y) + L1_reg * dnn.L1 + L2_reg * dnn.L2_sqr) # compiling a Theano function that computes the mistakes that are made # by the model on a minibatch test_model = theano.function( inputs=[index], outputs=dnn.errors(y), givens={ x: test_set_x[index * batch_size:(index + 1) * batch_size], y: test_set_y[index * batch_size:(index + 1) * batch_size] } ) gparams = [T.grad(cost, param) for param in dnn.params] updates = [(param, param - lr * gparam) for param, gparam in zip(dnn.params, gparams)] ll = 0.9 train_model = theano.function( inputs=[index, ll], outputs=cost, updates=updates, givens={ x: train_set_x[index * batch_size: (index + 1) * batch_size], y: train_set_y[index * batch_size: (index + 1) * batch_size] } ) ############### # TRAIN MODEL # ############### print('Start training ...') # # --- early-stopping parameters --- # # look as this many examples regardless patience = 500000 # wait this much longer when a new best is found patience_increase = 2 # a relative improvement of this much is considered significant improvement_threshold = 0.995 # go through this many minibatche before checking the network on the validation set; # in this case we check every epoch test_frequency = min(n_train_batches, patience / 100) best_test_loss = numpy.inf best_iter = 0 test_score = 0. start_time = time.clock() epoch = 0 done_looping = False while (epoch < n_epochs) and (not done_looping): epoch = epoch + 1 for minibatch_index in range(n_train_batches): minibatch_avg_cost = train_model(minibatch_index) # iteration number iter = (epoch - 1) * n_train_batches + minibatch_index if (iter + 1) % test_frequency == 0: # compute zero-one loss on validation set test_losses = [test_model(i) for i in range(n_test_batches)] this_test_loss = numpy.mean(test_losses) print('Epoch: %i; Batch: %i/%i, ValidationError: %f %%' % ( epoch, minibatch_index + 1, n_train_batches, this_test_loss * 100. ) ) # if we got the best validation score until now if this_test_loss < best_test_loss: #improve patience if loss improvement is good enough if this_test_loss < best_test_loss * improvement_threshold: patience = max(patience, iter * patience_increase) fh = open('model.pkl', 'wb') _pickle.dump(dnn, fh) best_test_loss = this_test_loss best_iter = iter if patience <= iter: done_looping = False break end_time = time.clock() print(('Optimization complete. Best validation score of %f %% ' 'Total iteration: %i. Best performance: %f %%') % ((1 - best_test_loss) * 100., best_iter + 1, (1 - test_score) * 100.)) test_num = 0 f_test = open(dataset + '/fbank/test.ark', 'r') list_test = list() testIDs = list() for l in f_test: line = l.strip('\n').split(' ') testIDs.append(line[0]) list_test.append(numpy.asarray(line[1:], dtype=float)) test_num += 1 f_test.close() test2_set_x = numpy.asarray(list_test) ''' col_sums = test2_set_x.sum(axis=0) tmp = test2_set_x / col_sums[numpy.newaxis, :] test2_set_x = tmp ''' shared_x = theano.shared(numpy.asarray(test2_set_x, dtype=theano.config.floatX), borrow=True) ############### # TEST MODEL # ############### test_model2 = theano.function( inputs=[index], on_unused_input='ignore', outputs=dnn.predict_labels(), givens={ x: shared_x, } ) list_pred_y = list() print(len(test2_set_x)) list_pred_y = test_model2(0) list_pred_labels = [index2label[i] for i in list_pred_y] f_lables = open('label_pred.csv', 'w') for i in range(len(list_pred_labels)): strii = testIDs[i] + ',' + list_pred_labels[i] + '\n' f_lables.write(strii) f_lables.close()