def __init__(self):
self.glove_path = os.path.join(
config.data_dir, "glove.6B.{}d.txt".format(config.embedding_size))
self.emb_matrix, self.word2id, self.id2word = get_glove(
self.glove_path, config.embedding_size)
self.train_context_path = os.path.join(config.data_dir,
"train.context")
self.train_qn_path = os.path.join(config.data_dir, "train.question")
self.train_ans_path = os.path.join(config.data_dir, "train.span")
self.dev_context_path = os.path.join(config.data_dir, "dev.context")
self.dev_qn_path = os.path.join(config.data_dir, "dev.question")
self.dev_ans_path = os.path.join(config.data_dir, "dev.span")
from config import Config
from data_utils import get_data
config = Config()
# Embeddings and word2id and id2word
glove_path = os.path.join(config.vectors_cache,
"glove.6B.{}d.txt".format(config.embedding_dim))
if not os.path.exists(glove_path):
print("\nDownloading wordvecs to {}".format(config.vectors_cache))
if not os.path.exists(config.vectors_cache):
os.makedirs(config.vectors_cache)
maybe_download(config.glove_base_url, config.glove_filename,
config.vectors_cache, 862182613)
emb_matrix, word2index, index2word = get_glove(glove_path,
config.embedding_dim)
train_context_path = os.path.join(config.data_dir, "train.context")
train_qn_path = os.path.join(config.data_dir, "train.question")
train_ans_path = os.path.join(config.data_dir, "train.span")
dev_context_path = os.path.join(config.data_dir, "dev.context")
dev_qn_path = os.path.join(config.data_dir, "dev.question")
dev_ans_path = os.path.join(config.data_dir, "dev.span")
def step(model, optimizer, batch):
"""
One batch of training
:return: loss
"""
# Here goes one batch of training