def get_net(net_path,
input_dim,
hidden_dims,
use_cuda=torch.cuda.is_available()):
net = extract_sdae_model(input_dim=input_dim, hidden_dims=hidden_dims)
checkpoint = torch.load(net_path)
net.load_state_dict(checkpoint['state_dict'])
if use_cuda:
net.cuda()
return net
def load_pretrained_fd_autoencoder(net_filename,
use_cuda=torch.cuda.is_available()):
"""
load pretrained stack denoise autoencoder
"""
checkpoint = torch.load(net_filename)
fd_ae = extract_sdae_model(cfg.INPUT_DIM, cfg.HIDDEN_DIMS)
# there some problems when loading cuda pretrained models
fd_ae.load_state_dict(checkpoint['state_dict'])
if use_cuda:
fd_ae.cuda()
return fd_ae
def __init__(self, root_dir, batch_size=256, n_clusters=4, fd_hidden_dim=10, layer_norm=True, lr=0.001,
direct_update=False, maxiter=2e4, update_interval=140, tol=0.001, gamma=0.1,
fine_tune_infersent=False, use_vat=False, use_tensorboard=False, semi_supervised=False, split_sents=False, id=0, verbose=True):
# model's settings
self.root_dir = root_dir
self.batch_size = batch_size
self.fd_hidden_dim = fd_hidden_dim
self.n_clusters = n_clusters
self.layer_norm = layer_norm
self.use_vat = use_vat
self.semi_supervised = semi_supervised
self.lr = lr
self.direct_update = direct_update
self.maxiter = maxiter
self.update_interval = update_interval
self.tol = tol
self.gamma = gamma
self.fine_tune_infersent = fine_tune_infersent
self.verbose = verbose
self.use_tensorboard = use_tensorboard
self.id = id
self.use_cuda = torch.cuda.is_available()
self.split_sents = split_sents
# data loader
self.corpus_loader = Corpus_Loader(self.root_dir,
layer_norm=self.layer_norm,
verbose=self.verbose,
use_cuda=self.use_cuda,
semi_supervised=self.semi_supervised,
split_sents=self.split_sents)
# model's components
self.kmeans = None
# self.fd_ae = extract_sdae_text(dim=fd_hidden_dim)
self.fd_ae = extract_sdae_model(input_dim=cfg.INPUT_DIM, hidden_dims=cfg.HIDDEN_DIMS)
self.cluster_layer = None
self.ae_criteron = nn.MSELoss()
self.cluster_criteron = F.binary_cross_entropy
self.optimizer = None
# model's state
self.current_p = None
self.current_q = None
self.current_pred_labels = None
self.past_pred_labels = None
self.current_cluster_acc = None
self.current_cluster_nmi = None
self.current_cluster_ari = None
# model's logger
self.logger_tensorboard = None
# initialize model's parameters and update current state
self.initialize_model()
self.initialize_tensorboard()
use_cuda = torch.cuda.is_available()
random_seed = args.seed
recons_lam = args.recons_lam
cluster_lam = args.cluster_lam
batch_size = args.batch_size
tol = args.tol
lr = args.lr
initialize_environment(random_seed=random_seed, use_cuda=use_cuda)
feat_path = os.path.join(data_dir, cfg.TRAIN_TEXT_FEAT_FILE_NAME)
feat, labels, ids = load_feat(feat_path)
outputdir = get_output_dir(data_dir)
net_filename = os.path.join(outputdir, cfg.PRETRAINED_FAE_FILENAME)
checkpoint = torch.load(net_filename)
net = extract_sdae_model(input_dim=cfg.INPUT_DIM,
hidden_dims=cfg.HIDDEN_DIMS)
net.load_state_dict(checkpoint['state_dict'])
if use_cuda:
net.cuda()
dcn = DCN(n_clusters,
net,
cfg.HIDDEN_DIMS[-1],
lr=lr,
tol=tol,
batch_size=batch_size,
recons_lam=recons_lam,
cluster_lam=cluster_lam,
use_cuda=use_cuda,
verbose=True)
dcn.fit(feat, labels=labels)