diff --git a/datasets/sst.py b/datasets/sst.py index f1c9ed2..842e5a9 100644 --- a/datasets/sst.py +++ b/datasets/sst.py @@ -28,7 +28,8 @@ class SST1(TabularDataset): return len(ex.text) @classmethod - def splits(cls, path, train='stsa.fine.phrases.train', validation='stsa.fine.dev', test='stsa.fine.test', **kwargs): + def splits(cls, path, train=os.path.join('SST', 'stsa.fine.phrases.train'), + validation=os.path.join('SST', 'stsa.fine.dev'), test= os.path.join('SST', 'stsa.fine.test'), **kwargs): return super(SST1, cls).splits( path, train=train, validation=validation, test=test, format='tsv', fields=[('label', cls.LABEL_FIELD), ('text', cls.TEXT_FIELD)] @@ -57,6 +58,7 @@ class SST1(TabularDataset): return BucketIterator.splits((train, val, test), batch_size=batch_size, repeat=False, shuffle=shuffle, sort_within_batch=True, device=device) + class SST2(TabularDataset): NAME = 'SST-2' NUM_CLASSES = 5 diff --git a/kim_cnn/__main__.py b/kim_cnn/__main__.py index 7b001d8..fafad7b 100644 --- a/kim_cnn/__main__.py +++ b/kim_cnn/__main__.py @@ -12,11 +12,11 @@ from common.evaluation import EvaluatorFactory from common.train import TrainerFactory from datasets.sst import SST1 from datasets.sst import SST2 +from datasets.aapd import AAPD from datasets.reuters import Reuters from kim_cnn.args import get_args from kim_cnn.model import KimCNN - class UnknownWordVecCache(object): """ Caches the first randomly generated word vector for a certain size to make it is reused. @@ -82,6 +82,8 @@ if __name__ == '__main__': train_iter, dev_iter, test_iter = SST2.iters(args.data_dir, args.word_vectors_file, args.word_vectors_dir, batch_size=args.batch_size, device=args.gpu, unk_init=UnknownWordVecCache.unk) elif args.dataset == 'Reuters': train_iter, dev_iter, test_iter = Reuters.iters(args.data_dir, args.word_vectors_file, args.word_vectors_dir, batch_size=args.batch_size, device=args.gpu, unk_init=UnknownWordVecCache.unk) + elif args.dataset == 'AAPD': + train_iter, dev_iter, test_iter = AAPD.iters(args.data_dir, args.word_vectors_file, args.word_vectors_dir, batch_size=args.batch_size, device=args.gpu, unk_init=UnknownWordVecCache.unk) else: raise ValueError('Unrecognized dataset') @@ -123,6 +125,10 @@ if __name__ == '__main__': train_evaluator = EvaluatorFactory.get_evaluator(Reuters, model, None, train_iter, args.batch_size, args.gpu) test_evaluator = EvaluatorFactory.get_evaluator(Reuters, model, None, test_iter, args.batch_size, args.gpu) dev_evaluator = EvaluatorFactory.get_evaluator(Reuters, model, None, dev_iter, args.batch_size, args.gpu) + elif args.dataset == 'AAPD': + train_evaluator = EvaluatorFactory.get_evaluator(AAPD, model, None, train_iter, args.batch_size, args.gpu) + test_evaluator = EvaluatorFactory.get_evaluator(AAPD, model, None, test_iter, args.batch_size, args.gpu) + dev_evaluator = EvaluatorFactory.get_evaluator(AAPD, model, None, dev_iter, args.batch_size, args.gpu) else: raise ValueError('Unrecognized dataset') @@ -154,29 +160,12 @@ if __name__ == '__main__': elif args.dataset == 'Reuters': evaluate_dataset('dev', Reuters, model, None, dev_iter, args.batch_size, args.gpu) evaluate_dataset('test', Reuters, model, None, test_iter, args.batch_size, args.gpu) + elif args.dataset == 'AAPD': + evaluate_dataset('dev', AAPD, model, None, dev_iter, args.batch_size, args.gpu) + evaluate_dataset('test', AAPD, model, None, test_iter, args.batch_size, args.gpu) else: raise ValueError('Unrecognized dataset') - # Calculate dev and test metrics - for data_loader in [dev_iter, test_iter]: - predicted_labels = list() - target_labels = list() - for batch_idx, batch in enumerate(data_loader): - scores_rounded = F.sigmoid(model(batch.text)).round().long() - predicted_labels.extend(scores_rounded.cpu().detach().numpy()) - target_labels.extend(batch.label.cpu().detach().numpy()) - predicted_labels = np.array(predicted_labels) - target_labels = np.array(target_labels) - accuracy = metrics.accuracy_score(target_labels, predicted_labels) - precision = metrics.precision_score(target_labels, predicted_labels, average='micro') - recall = metrics.recall_score(target_labels, predicted_labels, average='micro') - f1 = metrics.f1_score(target_labels, predicted_labels, average='micro') - if data_loader == dev_iter: - print("Dev metrics:") - else: - print("Test metrics:") - print(accuracy, precision, recall, f1) - if args.onnx: device = torch.device('cuda') if torch.cuda.is_available() and args.cuda else torch.device('cpu') dummy_input = torch.zeros(args.onnx_batch_size, args.onnx_sent_len, dtype=torch.long, device=device) diff --git a/kim_cnn/args.py b/kim_cnn/args.py index 5eeb55d..b0bf4d3 100644 --- a/kim_cnn/args.py +++ b/kim_cnn/args.py @@ -12,7 +12,7 @@ def get_args(): parser.add_argument('--mode', type=str, default='multichannel', choices=['rand', 'static', 'non-static', 'multichannel']) parser.add_argument('--lr', type=float, default=1.0) parser.add_argument('--seed', type=int, default=3435) - parser.add_argument('--dataset', type=str, default='SST-1', choices=['SST-1', 'SST-2', 'Reuters']) + parser.add_argument('--dataset', type=str, default='SST-1', choices=['SST-1', 'SST-2', 'Reuters', 'AAPD']) parser.add_argument('--resume_snapshot', type=str, default=None) parser.add_argument('--dev_every', type=int, default=30) parser.add_argument('--log_every', type=int, default=10) diff --git a/kim_cnn/model.py b/kim_cnn/model.py index 1e54520..0af3169 100644 --- a/kim_cnn/model.py +++ b/kim_cnn/model.py @@ -38,7 +38,7 @@ class KimCNN(nn.Module): self.dropout = nn.Dropout(config.dropout) self.fc1 = nn.Linear(Ks * output_channel, target_class) - def forward(self, x): + def forward(self, x, **kwargs): if self.mode == 'rand': word_input = self.embed(x) # (batch, sent_len, embed_dim) x = word_input.unsqueeze(1) # (batch, channel_input, sent_len, embed_dim)