seed = 2233235
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
# CONFIG SECTION
lang = sys.argv[1]
expname = "exp1_long"
batch_size = 32
# how many overall iterations.
iterations = 10
# A bunch of parameters from the data.
data_dict = get_data(lang)
train_dataset = data_dict["train"]
validation_dataset = data_dict["dev"]
test_dataset = data_dict["test"]
reader = data_dict["reader"]
WORD_EMB_DIM = data_dict["WORD_EMB_DIM"]
pretrained_file = data_dict["pretrained_file"]
vocab = data_dict["vocab"]
print("Train data stats:")
stats = get_stats(train_dataset)
print("total toks: ", stats["total_toks"])
print("total tags: ", stats["total_tags"])
# don't put a slash after this.
serialization_dir = "/scratch/models/{}/{}".format(lang, expname)
seed = 2233235 random.seed(seed) np.random.seed(seed) torch.manual_seed(seed) # CONFIG SECTION lang = sys.argv[1] expname = "oracle" batch_size = 32 # how many overall iterations. iterations = 5 # A bunch of parameters from the data. data_dict = get_data(lang) train_dataset = data_dict["train"] validation_dataset = data_dict["dev"] test_dataset = data_dict["test"] reader = data_dict["reader"] WORD_EMB_DIM = data_dict["WORD_EMB_DIM"] pretrained_file = data_dict["pretrained_file"] vocab = data_dict["vocab"] # this 0.999 thing is a hack. data_dict = get_data(lang, recall=0.999) gold_train_dataset = data_dict["train"] num_tags = len(reader.alltags) # This block sets all false negative values in the training data to have a uniform tag matrix.