print("-- Loading index")
with open(args.model_filename + ".index", mode="rb") as in_file:
index = pickle.load(in_file)
token2id = index["token2id"]
id2token = index["id2token"]
label2id = index["label2id"]
id2label = index["id2label"]
num_tokens = len(token2id)
num_labels = len(label2id)
print("-- Loading test set")
test_labels, test_padded_premises, test_padded_hypotheses, test_original_premises, test_original_hypotheses =\
load_te_dataset(
args.test_filename,
token2id,
label2id
)
print("-- Restoring model")
premise_input = tf.placeholder(tf.int32, (None, None),
name="premise_input")
hypothesis_input = tf.placeholder(tf.int32, (None, None),
name="hypothesis_input")
label_input = tf.placeholder(tf.int32, (None, ), name="label_input")
dropout_input = tf.placeholder(tf.float32, name="dropout_input")
logits = build_simple_te_model_h(premise_input, hypothesis_input,
dropout_input, num_tokens, num_labels,
None, params["embeddings_size"],
params["train_embeddings"],
params["rnn_hidden_size"],
json.dump(vars(args), out_file)
print("Params saved to: {}".format(args.model_save_filename +
".params"))
with open(args.model_save_filename + ".index", mode="wb") as out_file:
pickle.dump(
{
"token2id": token2id,
"id2token": id2token,
"label2id": label2id,
"id2label": id2label
}, out_file)
print("Index saved to: {}".format(args.model_save_filename + ".index"))
print("-- Loading training set")
train_labels, train_premises, train_hypotheses, _, _ = load_te_dataset(
args.train_filename, token2id, label2id)
print("-- Loading development set")
dev_labels, dev_premises, dev_hypotheses, _, _ = load_te_dataset(
args.dev_filename, token2id, label2id)
print("-- Building model")
premise_input = tf.placeholder(tf.int32, (None, None),
name="premise_input")
hypothesis_input = tf.placeholder(tf.int32, (None, None),
name="hypothesis_input")
label_input = tf.placeholder(tf.int32, (None, ), name="label_input")
dropout_input = tf.placeholder(tf.float32, name="dropout_input")
logits = build_simple_te_model(premise_input, hypothesis_input,
dropout_input, num_tokens, num_labels,
embeddings, args.embeddings_size,