default=1,
type=int,
help="Maximum number of threads to use.")
args = parser.parse_args()
# Create logdir name
args.logdir = "logs/{}-{}-{}".format(
os.path.basename(__file__),
datetime.datetime.now().strftime("%Y-%m-%d_%H%M%S"), ",".join(
("{}={}".format(re.sub("(.)[^_]*_?", r"\1", key), value)
for key, value in sorted(vars(args).items()))))
if not os.path.exists("logs"):
os.mkdir("logs") # TF 1.6 will do this by itself
# Load the data
train = nli_dataset.NLIDataset("nli-train.txt")
dev = nli_dataset.NLIDataset("nli-dev.txt",
train=train,
shuffle_batches=False)
test = nli_dataset.NLIDataset("nli-test.txt",
train=train,
shuffle_batches=False)
# Construct the network
network = Network(threads=args.threads)
network.construct(args, len(train.vocabulary("words")),
len(train.vocabulary("chars")),
len(train.vocabulary("languages")))
# Train
for i in range(args.epochs):
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--batch_size", default=16, type=int, help="Batch size.")
parser.add_argument("--data_train", default="nli-train.txt", type=str, help="Training data file.")
parser.add_argument("--data_dev", default="nli-dev.txt", type=str, help="Development data file.")
parser.add_argument("--data_test", default="nli-test.txt", type=str, help="Testing data file.")
parser.add_argument("--epochs", default=10, type=int, help="Number of epochs.")
parser.add_argument("--logdir", default="logs", type=str, help="Logdir name.")
parser.add_argument("--rnn_cell", default="GRU", type=str, help="RNN cell type.")
parser.add_argument("--rnn_cell_dim", default=100, type=int, help="RNN cell dimension.")
parser.add_argument("--threads", default=1, type=int, help="Maximum number of threads to use.")
args = parser.parse_args()
# Load the data
print("Loading the data.", file=sys.stderr)
data_train = nli_dataset.NLIDataset(args.data_train)
data_dev = nli_dataset.NLIDataset(args.data_dev, train=data_train)
data_test = nli_dataset.NLIDataset(args.data_test, train=data_train, no_languages=True)
# Construct the network
print("Constructing the network.", file=sys.stderr)
expname = "nli-{}{}-bs{}-epochs{}".format(args.rnn_cell, args.rnn_cell_dim, args.batch_size, args.epochs)
network = Network(rnn_cell=args.rnn_cell, rnn_cell_dim=args.rnn_cell_dim,
num_words=len(data_train.vocabulary('words')), num_chars=len(data_train.vocabulary('chars')),
logdir=args.logdir, expname=expname, threads=args.threads)
# Train
best_dev_accuracy = 0
test_predictions = None
for epoch in range(args.epochs):