def main(): train_x, train_y, val_x, val_y, test_x, test_y = load_data() train_x, train_y = preprocess_data(train_x, train_y) val_x, val_y = preprocess_data(val_x, val_y) test_x, test_y = preprocess_data(test_x, test_y) train_a_network(train_x, train_y, val_x, val_y, test_x, test_y)
def end2end(mcat_ratio: float, epochs: int, dim: int, nlayers: int, fname: str, gpu: bool, name: str, lr: float, batch_size: int, split_ratio: float, vectors: str, data_dir: str, pbar_width: int, wandb: bool, boom_dim: int, dropout: float, ema: float, overwrite_model: bool) -> str: ''' Loads the data, preprocesses it if needed, builds the SHARNN model, trains it and evaluates it. ''' from data import load_data from modules import SHARNN pp_path = get_path(data_dir) / f'{fname}_pp.tsv' if not pp_path.is_file(): from data import preprocess_data raw_path = get_path(data_dir) / f'{fname}.tsv' cats_path = get_path(data_dir) / 'cats.json' mcat_dict_path = get_path(data_dir) / 'mcat_dict.json' if not (raw_path.is_file() and cats_path.is_file() and mcat_dict_path.is_file()): from db import ArXivDatabase db = ArXivDatabase(data_dir=data_dir) db.get_mcat_dict() db.get_cats() if not raw_path.is_file(): db.get_training_df() preprocess_data(data_dir=data_dir) train_dl, val_dl, vocab = load_data(tsv_fname=f'{fname}_pp', batch_size=batch_size, split_ratio=split_ratio, vectors=vectors, data_dir=data_dir) model = SHARNN(dim=dim, nlayers=nlayers, data_dir=data_dir, pbar_width=pbar_width, vocab=vocab, boom_dim=boom_dim, dropout=dropout) if gpu: model.cuda() model = model.fit(train_dl, val_dl, epochs=epochs, lr=lr, mcat_ratio=mcat_ratio, name=name, use_wandb=wandb, ema=ema, overwrite_model=overwrite_model) return model.evaluate(val_dl)
def run(): model = GRU(FLAGS.input_dim, FLAGS.hidden_dim) print(model) optimizer = optim.SGD(model.parameters(), lr=0.001) data_paths = [FLAGS.data_path, FLAGS.eval_data_path] datasets, embeddings = preprocess_data(data_paths, FLAGS.embedding_path) training_data = Dataset(datasets[0]) eval_data = Dataset(datasets[1], shuffle=False, truncate_final_batch=True) if FLAGS.extract: raise NotImplementedError def eval_fn(): return evaluate(model, eval_data, embeddings) def ckpt_fn(ckpt_args): step = ckpt_args['step'] best_dev_error = ckpt_args['best_dev_error'] best = ckpt_args.get('best', False) save_dict = pack_checkpoint(step, best_dev_error, model, optimizer) filename = ckpt_path(FLAGS.save_path, filename=FLAGS.experiment_name, best=best) print("Checkpointing...") save(save_dict, filename) train_args = dict(step=0, best_dev_error=1.0) train(model, optimizer, training_data, embeddings, eval_fn, ckpt_fn, train_args)
def main(): print('getting data...') df = get_data() X_train, X_test, y_train, y_test, enc = preprocess_data(df) print('training...') model = Net() target = to_tensor(y_train) loss_fn = torch.nn.MSELoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.01) for i in range(100): y_pred = model(to_tensor(X_train)) loss = loss_fn(y_pred, target) print(f'loss on pass {i}: {loss}') optimizer.zero_grad() loss.backward() optimizer.step() print('trained!') y_pred = enc.inverse_transform(model.predict(to_tensor(X_test))) # print('y_test') y_test_deenc = enc.inverse_transform((y_test)) # print(y_test_deenc) # print('y_pred') # print((y_pred)) print(f'accuracy score: {balanced_accuracy_score(y_test_deenc,y_pred)}') return X_train, X_test, y_train, y_test, model
def _init(): # Read data and shuffle it dataframe = pandas.read_csv('data/log.csv', sep=',', header=0) dataframe = dataframe.reindex(numpy.random.permutation(dataframe.index)) # Generate features and targets processed_features, processed_targets = preprocess_data(dataframe) print('==========Feature Summary==========') print(processed_features.describe()) print() print('==========Target Summary===========') print(processed_targets.describe()) print() # ratio = training data / all data training_data_ratio = 0.8 # Separate data to training and validation sets training_data_count = math.ceil(training_data_ratio * len(dataframe)) training_examples = processed_features.head(training_data_count) training_targets = processed_targets.head(training_data_count) validation_data_count = len(dataframe) - training_data_count validation_examples = processed_features.tail(validation_data_count) validation_targets = processed_targets.tail(validation_data_count) # Train neural network model = NeuralNetworkModel(hidden_units=[10, 10]) model.train( learning_rate=0.005, steps=750, batch_size=training_data_count, training_examples=training_examples, training_targets=training_targets, validation_examples=validation_examples, validation_targets=validation_targets ) # Predict unlabeled data and display result unlabeled_dataframe = pandas.read_csv('data/log_unlabeled.csv', sep=',', header=0) unlabeled_features, unlabeled_targets = preprocess_data(unlabeled_dataframe) unlabeled_predictions = model.predict(unlabeled_features, unlabeled_targets) result_table = restore_category(unlabeled_features).assign(prediction=unlabeled_predictions) print('\n==========Unlabeled Data Predictions==========') print(result_table.to_string())
def extract_data(size=256): print("Extracting data..") X, y = data.extract_data(size=256) print("Preprocessing data..") X, y, nb_samples, num_categories = data.preprocess_data(X, y, save=True, subtract_mean=True) return X, y, nb_samples, num_categories
def build_validation_data_loader(self) -> keras.InputData: if not self.data_downloaded: self.download_directory = download_cifar10_tf_sequence( download_directory=self.download_directory, url=self.context.get_data_config()["url"], ) self.data_downloaded = True (_, _), (test_data, test_labels) = get_data(self.download_directory) return preprocess_data(test_data), preprocess_labels(test_labels)
def main(): # Use the loading function from Assignment 1 train_x, train_y, val_x, val_y, test_x, test_y = load_data(use_all=True, val_size=5000) # Use the preprocessing function from Assignment 1 train_x, train_y = preprocess_data(train_x, train_y) val_x, val_y = preprocess_data(val_x, val_y) test_x, test_y = preprocess_data(test_x, test_y) # Process the data so they have a zero mean mean, std = np.mean(train_x), np.std( train_x) # Find mean and std of training data train_x = process_zero_mean(train_x, mean, std) val_x = process_zero_mean(val_x, mean, std) test_x = process_zero_mean(test_x, mean, std) # Testing gradients # test_grad_computations(train_x, train_y) # Testing model capabilities # overfit_test(train_x, train_y) # Test different model parameters # train_a_network(train_x, train_y, val_x, val_y, test_x, test_y) # Find a good regularisation value # lambda_search(train_x, train_y, val_x, val_y) # Train a network with optimal hyper-parameter values using almost all of the data # use_all=True, val_size=1000 | cycles = 5 | n_s = 1000 | lambda = 0.00087 # train_a_network(train_x, train_y, val_x, val_y, test_x, test_y, n_cycles=5) # Report the effect of adding noise to the training data # test_noise(train_x, train_y, val_x, val_y, test_x, test_y) # Report the effect of different number of nodes # test_nodes(train_x, train_y, val_x, val_y, test_x, test_y) # Report the effect of ensemble method test_ensemble(train_x, train_y, val_x, val_y, test_x, test_y)
def main(): # Use the loading function from Assignment 1 train_x, train_y, val_x, val_y, test_x, test_y = load_data(use_all=True, val_size=5000) # Use the preprocessing function from Assignment 1 train_x, train_y = preprocess_data(train_x, train_y) val_x, val_y = preprocess_data(val_x, val_y) test_x, test_y = preprocess_data(test_x, test_y) # Process the data so they have a zero mean mean, std = np.mean(train_x), np.std( train_x) # Find mean and std of training data train_x = process_zero_mean(train_x, mean, std) val_x = process_zero_mean(val_x, mean, std) test_x = process_zero_mean(test_x, mean, std) # Testing gradients # test_grad_computations(train_x, train_y) # Training simple k-layer networks train_simple(train_x, train_y, val_x, val_y, test_x, test_y)
# os.chdir(wpath) # print(os.getcwd()) with open("train2012_MAX.txt", 'r', encoding="utf8", errors="ignore" ) as f: # encoding='cp1252', encoding='utf8', errors="replace" data_train = f.readlines() # data_train = load_file('data/test2011') data_valid = load_file('data/dev2012') data_test = load_file('data/test2011') tokenizer = BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=True) print('PREPROCESSING DATA...') X_train, y_train = preprocess_data(data_train, tokenizer, punctuation_enc, segment_size) X_valid, y_valid = preprocess_data(data_valid, tokenizer, punctuation_enc, segment_size) print('INITIALIZING MODEL...') output_size = len(punctuation_enc) bert_punc = nn.DataParallel( BertPunc(segment_size, output_size, dropout).cuda()) # print('TRAINING TOP LAYER...') # data_loader_train = create_data_loader(X_train, y_train, True, batch_size_top) # data_loader_valid = create_data_loader(X_valid, y_valid, False, batch_size_top) # freeze all layers of bert and just train task based classifier so that the gradients are not computed in backward(). Parameters of newly constructed modules (below requires_grad=False statment) have requires_grad=True by default # for p in bert_punc.module.bert.parameters(): # p.requires_grad = False
def run(epochs=500, training_percentage=0.4, validation_percentage=0.1, extract=True, cont=True, size=256, top_k=5): '''Does the routine required to get the data, put them in needed format and start training the model saves weights whenever the model produces a better test result and keeps track of the best loss''' if extract: print("Extracting data..") X, y = data.extract_data(size=size) print("Preprocessing data..") X, y, nb_samples, num_categories = data.preprocess_data(X, y, save=True, subtract_mean=True) else: print("Loading data..") h5f = h5py.File('data.hdf5', 'r') nb_samples = h5f['nb_samples'].value num_categories = h5f['n_categories'].value h5f.close() print("Number of categories: {}".format(num_categories)) print("Number of samples {}".format(nb_samples)) data_ids = np.arange(start=0, stop=nb_samples) val_ids = data.produce_validation_indices(data_ids, nb_samples * validation_percentage) train_ids = data.produce_train_indices(dataset_indx=data_ids, number_of_samples=nb_samples * training_percentage, val_indx=val_ids) # X_train, y_train, X_test, y_test = data.split_data(X, y, split_ratio=split) X_train, y_train, X_val, y_val = data.load_dataset_bit_from_hdf5(train_ids, val_ids, only_train=False) X_val = X_val / 255 print("Building and Compiling model..") model = m.get_model(n_outputs=num_categories, input_size=size) if cont: # model.load_weights_until_layer("pre_trained_weights/latest_model_weights.hdf5", 26) model.load_weights("pre_trained_weights/latest_model_weights.hdf5") model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=["accuracy"]) print("Training..") best_performance = np.inf for i in range(epochs): train_ids = data.produce_train_indices(dataset_indx=data_ids, number_of_samples=15000, val_indx=val_ids) X_train, y_train = data.load_dataset_bit_from_hdf5(train_ids, val_ids, only_train=True) X_train = X_train / 255 X_train = data.augment_data(X_train) # fit the model on the batches generated by datagen.flow() metadata = model.fit(X_train, y_train, validation_data=[X_val, y_val], batch_size=64, nb_epoch=1, verbose=1, shuffle=True, class_weight=None, sample_weight=None) current_loss = metadata.history['loss'][-1] current_val_loss = metadata.history['val_loss'][-1] preds = model.predict_proba(X_val, batch_size=64) print("Loss: {}".format(current_loss)) print("Val_loss: {}".format(current_val_loss)) top_3_error = get_top_n_error(preds, y_val, top_k) print("Top 3 error: {}".format(top_3_error)) if current_val_loss < best_performance: model.save_weights("pre_trained_weights/model_weights.hdf5", overwrite=True) best_performance = current_val_loss print("Saving weights..") model.save_weights("pre_trained_weights/latest_model_weights.hdf5", overwrite=True)
axes[(0, 0)].set_title('Truth: 1', fontsize=15) axes[(0, 1)].set_title('Truth: 0', fontsize=15) axes[(0, 0)].set_ylabel('Pred: 1', fontsize=15) axes[(1, 0)].set_ylabel('Pred: 0', fontsize=15) # axes[(1,1)].y plt.savefig('uncertainty') if __name__ == "__main__": parser = argparse.ArgumentParser(description='Uncertainty estimation and evaluation') parser.add_argument("checkpoint", type=str) parser.add_argument("--datadir", type=str, default='dataset'), params, _ = parser.parse_known_args() print("preprocess") train_set, val_set, test_set, tay_set = data.preprocess_data( data_folder=params.datadir) model = torch.load(params.checkpoint, map_location=device) train_it, val_it, test_it, tay_it = data.get_batch_iterators( 64, train_set, val_set, test_set, tay_set) ground_truth, predictions, dropout_preds, ids = predict(model,test_it, num_samples=100) np.savetxt('predictions.csv', predictions) np.savetxt('dropout_preds.csv', dropout_preds) np.savetxt('ground_truth.csv', ground_truth) sorted_variances = dropout_preds.var(axis=1).argsort(axis=0) max_indices, min_indices = sorted_variances[:5], sorted_variances[-5:] # print(dropout_preds.var(axis=1).max(axis=0)) # print(dropout_preds.var(axis=1).min(axis=0))
# network training parser.add_argument("--batch_size", type=int, default=16, help="batch size") parser.add_argument("--unit_epoch", type=int, default=50, help="number of transition epochs") parser.add_argument("--lambda_gp", type=float, default=10., help="weight of gradient penalty in WGAN") parser.add_argument("--lambda_drift", type=float, default=0.001, help="weight of drift term in D_loss") parser.add_argument("--num_aug", type=int, default=5, help="times of data augmentation; \ x5 when training unconditional gan (using only mel data); x1 when training conditional gan (using the whole training set)") parser.add_argument("--lr", type=float, default=0.001, help="initial learning rate") parser.add_argument("--outf", type=str, default="logs", help='path of log files') # inference parser.add_argument("--num", type=int, default=1000, help="number of images to generate") arg = parser.parse_args() if arg.preprocess: mel_only = False if arg.cond else True preprocess_data(root_dir='../data_2018', mel_only=mel_only) assert arg.mode == "train" or arg.mode == "test", "invalid argument!" if arg.mode == "train": if arg.cond: print("\ntrain conditional GAN ...\n") gan_trainer = CondTrainer(arg, device, device_ids) else: gan_trainer = Trainer(arg, device, device_ids) gan_trainer.train() if arg.mode == "test": gan_generator = ImageGenerator(arg, device) gan_generator.generate(arg.num)
target_words_emb) word_sim = fluid.layers.reduce_sum(word_sim, dim=-1) word_sim = fluid.layers.reshape(word_sim, shape=[-1]) pred = fluid.layers.sigmoid(word_sim) # 通过估计的输出概率定义损失函数,注意我们使用的是sigmoid_cross_entropy_with_logits函数 # 将sigmoid计算和cross entropy合并成一步计算可以更好的优化,所以输入的是word_sim,而不是pred loss = fluid.layers.sigmoid_cross_entropy_with_logits(word_sim, label) loss = fluid.layers.reduce_mean(loss) # 返回前向计算的结果,飞桨会通过backward函数自动计算出反向结果。 return pred, loss corpus = preprocess_data() word2id_dict, word2id_freq, id2word_dict = build_dict(corpus) corpus = [word2id_dict[word] for word in corpus] vocab_size = len(word2id_dict) print(f"vocab size: {vocab_size}") corpus = subsampling(corpus, word2id_freq) print("after subsampling %d tokens in the corpus" % len(corpus)) # print(f"finish create dateset: {len(dataset)} sample") step = 0 learning_rate = 0.001 # 定义一个使用word-embedding查询同义词的函数 # 这个函数query_token是要查询的词,k表示要返回多少个最相似的词,embed是我们学习到的word-embedding参数 # 我们通过计算不同词之间的cosine距离,来衡量词和词的相似度
I_val = utils.make_grid(fixed_batch, nrow=4, normalize=True, scale_each=True) writer.add_image('val/image', I_val, epoch) if opt.log_images and (not opt.no_attention): print('\nlog attention maps ...\n') # training data __, a1, a2 = model(inputs[0:16,:,:,:]) if a1 is not None: attn1 = visualize_attn(I_train, a1, up_factor=opt.base_up_factor, nrow=4) writer.add_image('train/attention_map_1', attn1, epoch) if a2 is not None: attn2 = visualize_attn(I_train, a2, up_factor=2*opt.base_up_factor, nrow=4) writer.add_image('train/attention_map_2', attn2, epoch) # val data __, a1, a2 = model(fixed_batch) if a1 is not None: attn1 = visualize_attn(I_val, a1, up_factor=opt.base_up_factor, nrow=4) writer.add_image('val/attention_map_1', attn1, epoch) if a2 is not None: attn2 = visualize_attn(I_val, a2, up_factor=2*opt.base_up_factor, nrow=4) writer.add_image('val/attention_map_2', attn2, epoch) # Append the row to the dataframes train_results = train_results.append(train_row, ignore_index=True) val_results = val_results.append(val_row, ignore_index=True) # write to csvs train_results.to_csv('train_scores.csv') val_results.to_csv('val_scores.csv') if __name__ == "__main__": preprocess_data(root_dir='../ImageDataSet/Xray') main()
def run(epochs=500, training_percentage=0.4, validation_percentage=0.1, extract=True, cont=True, size=256, top_k=5): '''Does the routine required to get the data, put them in needed format and start training the model saves weights whenever the model produces a better test result and keeps track of the best loss''' if extract: print("Extracting data..") X, y = data.extract_data(size=size) print("Preprocessing data..") X, y, nb_samples, num_categories = data.preprocess_data( X, y, save=True, subtract_mean=True) else: print("Loading data..") h5f = h5py.File('data.hdf5', 'r') nb_samples = h5f['nb_samples'].value num_categories = h5f['n_categories'].value h5f.close() print("Number of categories: {}".format(num_categories)) print("Number of samples {}".format(nb_samples)) data_ids = np.arange(start=0, stop=nb_samples) val_ids = data.produce_validation_indices( data_ids, nb_samples * validation_percentage) train_ids = data.produce_train_indices(dataset_indx=data_ids, number_of_samples=nb_samples * training_percentage, val_indx=val_ids) # X_train, y_train, X_test, y_test = data.split_data(X, y, split_ratio=split) X_train, y_train, X_val, y_val = data.load_dataset_bit_from_hdf5( train_ids, val_ids, only_train=False) X_val = X_val / 255 print("Building and Compiling model..") model = m.get_model(n_outputs=num_categories, input_size=size) if cont: # model.load_weights_until_layer("pre_trained_weights/latest_model_weights.hdf5", 26) model.load_weights("pre_trained_weights/latest_model_weights.hdf5") model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=["accuracy"]) print("Training..") best_performance = np.inf for i in range(epochs): train_ids = data.produce_train_indices(dataset_indx=data_ids, number_of_samples=15000, val_indx=val_ids) X_train, y_train = data.load_dataset_bit_from_hdf5(train_ids, val_ids, only_train=True) X_train = X_train / 255 X_train = data.augment_data(X_train) # fit the model on the batches generated by datagen.flow() metadata = model.fit(X_train, y_train, validation_data=[X_val, y_val], batch_size=64, nb_epoch=1, verbose=1, shuffle=True, class_weight=None, sample_weight=None) current_loss = metadata.history['loss'][-1] current_val_loss = metadata.history['val_loss'][-1] preds = model.predict_proba(X_val, batch_size=64) print("Loss: {}".format(current_loss)) print("Val_loss: {}".format(current_val_loss)) top_3_error = get_top_n_error(preds, y_val, top_k) print("Top 3 error: {}".format(top_3_error)) if current_val_loss < best_performance: model.save_weights("pre_trained_weights/model_weights.hdf5", overwrite=True) best_performance = current_val_loss print("Saving weights..") model.save_weights("pre_trained_weights/latest_model_weights.hdf5", overwrite=True)
from utils import read_imagenet_classnames, display_results, run_inference, parse_base64 from torchvision import models parser = argparse.ArgumentParser(description='Inference Trained Model') parser.add_argument('--data', metavar='DIR', default='./data', help='default data path') parser.add_argument('-bs', '--batch-size', metavar='BS', default=2, help='maximum batchsize') parser.add_argument('-tp', '--top-predictions', metavar='NUMPRED',\ default=5, help='number of top predictions per sample') parser.add_argument('-exp', '--export', action="store_true",help='export model to onnx') def export_model(model): model_path = 'checkpoints/model.pt' sample_input = torch.randn((1, 3, 256, 256)) model = model.cpu() model.eval() model = torch.jit.trace(model, sample_input) torch.jit.save(model, model_path) if __name__ == "__main__": args = parser.parse_args() model = models.resnet18(pretrained=True) if args.export: export_model(model) else: imagenet_classes = read_imagenet_classnames("cache/imagenet_classnames.txt") data = preprocess_data("cache") predictions = run_inference(model, data[0], args.top_predictions) display_results(data, predictions, imagenet_classes)
def train(): # get batch iterator print("preprocess") train_set, val_set, test_set, _ = data.preprocess_data( data_folder=params.datadir) print(model_config) print("Training model...") model_config['num_embeddings'] = train_set.fields[ 'text'].vocab.vectors.shape[0] model_config['embedding_dim'] = train_set.fields[ 'text'].vocab.vectors.shape[1] model = Main(model_config, train_set.fields['text'].vocab) pytorch_total_params = sum(p.numel() for p in model.parameters()) print(pytorch_total_params) model.to(device) grad_params = [p for p in model.parameters() if p.requires_grad] # weight = torch.FloatTensor(N_CLASSES).fill_(1) weight = torch.FloatTensor([0.3, 0.7]) ce_loss = nn.CrossEntropyLoss(weight=weight).to(device) lr = model_config['learning_rate'] epoch = 1 prev_dev_accuracy = 0 optimizer = optim.Adam(grad_params, lr, weight_decay=model_config['weight_decay']) best_epoch = 0 while epoch < params.epochs: # writer.add_scalar( # 'Learning rate', optimizer.param_groups[0]['lr'], epoch) optimizer.param_groups[0]['lr'] = optimizer.param_groups[0]['lr'] * \ LR_DECAY if epoch > 1 else optimizer.param_groups[0]['lr'] train_it, val_it, test_it, _ = data.get_batch_iterators( MINIBATCH_SIZE, train_set, val_set, test_set) train_loss = 0 train_loss_epoch = 0 batches = 0 for batch in train_it: optimizer.zero_grad() output = model.forward(batch.text) loss = ce_loss(output, batch.label_a) loss.backward() train_loss += loss train_loss_epoch += loss optimizer.step() batches += 1 if batches % 50 == 0: train_loss = 0 # writer.add_scalar('Loss', train_loss, batches / 50) print("Training Loss at epoch: {} is: {}".format(epoch, train_loss)) # writer.add_scalar('Loss (Epoch)', train_loss_epoch, epoch) n_correct = 0 n_tested = 0 true_positive = [0, 0] true_negative = [0, 0] false_positive = [0, 0] false_negative = [0, 0] true_labels = [0, 0] for batch in val_it: output = model.forward(batch.text) scores, predictions = torch.max(output, dim=1) for i in range(2): true_labels = (batch.label_a == i).nonzero() false_labels = (batch.label_a != i).nonzero() true_positive[i] += ( predictions[true_labels] == i).sum().item() false_negative[i] += (predictions[true_labels] != i).sum().item() false_positive[i] += ( predictions[false_labels] == i).sum().item() true_negative[i] += (predictions[false_labels] != i).sum().item() n_correct += (batch.label_a == predictions).sum() n_tested += batch.label_a.shape[0] # writer.add_scalar('Validation accuracy', accuracy, epoch) macro_precision = 0.0 macro_recall = 0.0 macro_f1 = 0.0 non_zero_devision = 1e-6 for i in range(2): precision = ( true_positive[i] / (true_positive[i] + false_positive[i] + non_zero_devision)) macro_precision += precision recall = ( true_positive[i] / (true_positive[i] + false_negative[i] + non_zero_devision)) macro_recall += recall macro_f1 += 2 * (precision*recall) / \ (precision+precision+ non_zero_devision) macro_precision /= 2 macro_recall /= 2 macro_f1 /= 2 print( f'Precision: {macro_precision}\nRecall: {macro_recall}\nF1: {macro_f1}' ) accuracy = n_correct.item() / n_tested if macro_f1 <= prev_dev_accuracy: # optimizer.param_groups[0]['lr'] /= 2 lr = optimizer.param_groups[0]['lr'] print(f"lr: {lr} and threshold: {THRESHOLD}") else: prev_dev_accuracy = macro_f1 best_epoch = epoch if (params.save_model): torch.save( model, os.path.join(params.outputdir, params.model + "_epoch_" + str(epoch) + ".pt")) print("Validation accuracy at epoch: {} is: {}, f1 {}".format( epoch, accuracy, macro_f1)) # Store model epoch += 1 validate(test_it, best_epoch, train_set.fields['text'].vocab, model_config)
# Dataset parameters img_rows, img_cols = 28, 28 #input image dimensions for the MNIST dataset num_classes = 10 #number of classes in the MNIST dataset # Load the data, and split between train and test sets (x_train, y_train), (x_test, y_test) = mnist.load_data() # Specify channel order K.set_image_data_format('channels_last') # Preprocess the data input_shape, x_train, y_train, x_test, y_test = preprocess_data( x_train, y_train, x_test, y_test, img_rows, img_cols, num_classes, K.image_data_format(), print_shape=False) model = ThreeLayersCNN(input_shape=input_shape, nb_filters=num_filters, nb_classes=num_classes, print_model_summary=False) model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy']) model.fit(x_train,
#### #### #### Responsible for loading data as well as training and evaluating the model #### #### import constants as C from data import load_raw_data, preprocess_data, split_data_to_training_and_test_sets from model import create_logistic_regression_model, create_SVM, calculate_accuracy if __name__ == '__main__': # Loads the raw data dataset = load_raw_data() # Splits the data to training and test sets X, y = preprocess_data(dataset, C.FEATURES) X_train, X_test, y_train, y_test = split_data_to_training_and_test_sets( X, y) # Generate the LR model, train, and get the accuracy score lr_classifier = create_logistic_regression_model(X_train, y_train) accuracy = calculate_accuracy(lr_classifier, X_test, y_test) print() print('--------------') print('Accuracy of logistic regression model using all features:', accuracy) print('--------------') print() # Generate the SVC model, train, and get the accuracy score