def build_model(fields, args, checkpoint=None):
"""Build model."""
model = LMModel(fields, args)
if checkpoint is not None:
logger.info("Set model using saved checkpoint")
model.load_state_dict(checkpoint["model"])
return model.to(args.device)
def build_test_model(fields, params):
"""Build test model."""
model = LMModel(fields, params["args"])
model.load_state_dict(params["model"])
model = model.to(params["args"].device)
model.eval()
return model
def main(_):
if not FLAGS.data_path:
raise ValueError("Must set --data_path to PTB data directory")
raw_data = reader.ptb_raw_data(FLAGS.data_path)
test_data= raw_data
eval_config = get_config()
eval_config.batch_size = 1
eval_config.num_steps = 1
config_gpu = tf.ConfigProto(allow_soft_placement=True)
with tf.Graph().as_default(),tf.Session(config=config_gpu) as sess:
initializer = tf.random_uniform_initializer(-eval_config.init_scale,eval_config.init_scale)
with tf.name_scope("Test"):
with tf.variable_scope("model", reuse=None,initializer=initializer):
mtest = LMModel(is_training=False, config=eval_config)
tf.global_variables_initializer().run()
saver = tf.train.Saver()
checkpoint=tf.train.latest_checkpoint(FLAGS.save_path)
saver.restore(sess,checkpoint)
test_perplexity = do_eval(sess, mtest, test_data)
print("Test Perplexity: %.3f" % test_perplexity)
# load data
train_batch_size = args.train_batch_size
eval_batch_size = args.eval_batch_size
batch_size = {'train': train_batch_size, 'valid': eval_batch_size}
data_loader = data.Corpus("../data/ptb", batch_size, args.max_sql)
# WRITE CODE HERE within two '#' bar
########################################
# Build LMModel model (bulid your language model here)
nvoc = len(data_loader.vocabulary)
print("nvoc: " + str(nvoc))
ninput = 150
nhid = 150
nlayer = 4
model = LMModel(nvoc, ninput, nhid, nlayer, device, args.attention).to(device)
########################################
criterion = nn.CrossEntropyLoss().to(device)
# WRITE CODE HERE within two '#' bar
########################################
# Evaluation Function
# Calculate the average cross-entropy loss between the prediction and the ground truth word.
# And then exp(average cross-entropy loss) is perplexity.
def bleu_metric(candidates, references, max_n):
def get_max_n_grams(s):
ans = []
else:
device = torch.device("cpu")
# load data
train_batch_size = args.train_batch_size
eval_batch_size = args.eval_batch_size
batch_size = {'train': train_batch_size, 'valid': eval_batch_size}
data_loader = data.Corpus("../data/ptb", batch_size, args.max_sql)
# WRITE CODE HERE within two '#' bar
########################################
# Build LMModel model (build your language model here)
n_input = 256
n_hid = 512
n_layers = 2
model = LMModel(data_loader.nvoc, n_input, n_hid, n_layers)
if args.cuda:
model.cuda()
opt = optim.Adam(model.parameters(),
lr=args.learning_rate,
weight_decay=0.00001)
criterion = nn.CrossEntropyLoss()
########################################
# WRITE CODE HERE within two '#' bar
########################################
# Evaluation Function
# Calculate the average cross-entropy loss between the prediction and the ground truth word.