def get_model(config, args, train_dataset, device):
# Pass vocabulary to construct Embedding layer.
encoder = Encoder(config["model"], train_dataset.vocabulary)
decoder = Decoder(config["model"], train_dataset.vocabulary)
print("Encoder: {}".format(config["model"]["encoder"]))
print("Decoder: {}".format(config["model"]["decoder"]))
# New: Initializing word_embed using GloVe
if "glove_npy" in config["dataset"]:
encoder.word_embed.weight.data = torch.from_numpy(
np.load(config["dataset"]["glove_npy"]))
print("Loaded glove vectors from {}".format(
config["dataset"]["glove_npy"]))
# Share word embedding between encoder and decoder.
if encoder.word_embed and decoder.word_embed:
decoder.word_embed = encoder.word_embed
# Wrap encoder and decoder in a model.
model = EncoderDecoderModel(encoder, decoder).to(device)
if -1 not in args.gpu_ids:
model = nn.DataParallel(model, args.gpu_ids)
return model
args.val_dense_json,
overfit=args.overfit,
in_memory=args.in_memory,
return_options=True,
add_boundary_toks=False,
sample_flag=False
)
val_dataloader = DataLoader(
val_dataset,
batch_size=config["solver"]["batch_size"],
num_workers=args.cpu_workers,
shuffle=True,
)
# Pass vocabulary to construct Embedding layer.
encoder = Encoder(config["model"], val_dataset.vocabulary)
decoder = Decoder(config["model"], val_dataset.vocabulary)
print("Encoder: {}".format(config["model"]["encoder"]))
print("Decoder: {}".format(config["model"]["decoder"]))
# Share word embedding between encoder and decoder.
if args.load_pthpath == "":
print('load glove')
decoder.word_embed = encoder.word_embed
glove = np.load('data/glove.npy')
encoder.word_embed.weight.data = torch.tensor(glove)
# Wrap encoder and decoder in a model.
model = EncoderDecoderModel(encoder, decoder).to(device)
if -1 not in args.gpu_ids:
model = nn.DataParallel(model, args.gpu_ids)
train_dataset = VisDialDataset(
config["dataset"], args.train_json, overfit=args.overfit, in_memory=args.in_memory
)
train_dataloader = DataLoader(
train_dataset, batch_size=config["solver"]["batch_size"], num_workers=args.cpu_workers
)
val_dataset = VisDialDataset(
config["dataset"], args.val_json, args.val_dense_json, overfit=args.overfit, in_memory=args.in_memory
)
val_dataloader = DataLoader(
val_dataset, batch_size=config["solver"]["batch_size"], num_workers=args.cpu_workers
)
# pass vocabulary to construct nn.Embedding
encoder = Encoder(config["model"], train_dataset.vocabulary)
decoder = Decoder(config["model"], train_dataset.vocabulary)
print("Encoder: {}".format(config["model"]["encoder"]))
print("Decoder: {}".format(config["model"]["decoder"]))
# share word embedding between encoder and decoder
decoder.word_embed = encoder.word_embed
# wrap encoder and decoder in a model
model = EncoderDecoderModel(encoder, decoder).to(device)
if -1 not in args.gpu_ids:
model = nn.DataParallel(model, args.gpu_ids)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=config["solver"]["initial_lr"])
scheduler = lr_scheduler.StepLR(optimizer, step_size=1, gamma=config["solver"]["lr_gamma"])
KAT.append(key_and_token)
elmo_token = {}
for item in KAT:
elmo_token[item[1]] = elmo[item[0]]
elmo_list = []
for i in range(len(glovevocabulary)):
if i in elmo_token.keys():
elmo_list.append(elmo_token[i])
else:
randArray = random.random(size=(1, 1024)).tolist()
elmo_list.append(randArray[0])
elmo_token = torch.Tensor(elmo_list).view(len(glovevocabulary), -1)
# Pass vocabulary to construct Embedding layer.
encoder = Encoder(config["model"], val_dataset.vocabulary, glove_token,
elmo_token)
decoder = Decoder(config["model"], val_dataset.vocabulary, glove_token,
elmo_token)
print("Encoder: {}".format(config["model"]["encoder"]))
print("Decoder: {}".format(config["model"]["decoder"]))
# Share word embedding between encoder and decoder.
decoder.glove_embed = encoder.glove_embed
decoder.elmo_embed = encoder.elmo_embed
decoder.embed_change = encoder.embed_change
# Wrap encoder and decoder in a model.
model = EncoderDecoderModel(encoder, decoder).to(device)
if -1 not in args.gpu_ids:
model = nn.DataParallel(model, args.gpu_ids)