# Train LSTM to predict entailment from SNLI data
import random
import sys
import time
import tensorflow as tf
import numpy as np
np.set_printoptions(threshold=np.nan)
from util.Layer import Layers
from util.Train_entail_lstm import Training
from util.DataLoader import DataLoader
Data = DataLoader()
Layer = Layers()
sys.path.append(".")
def run(**args):
tf.reset_default_graph()
random.seed(20160408)
tf.set_random_seed(20160408)
start_time = time.time()
graph = tf.get_default_graph()
# Read Training/Dev/Test data
data_dir = 'data/' + args['data_dir'] + '/'
np_matrix, index = Data.read_glove_vectors('data/' + args['data_dir'] +
'/' + args['vector_file'])
num_classes = 3
# entailment data
# train model to predict conditional probabilities and p(x) individual phrase probabilities
import random
import sys
from tensorflow.contrib.rnn.python.ops import lstm_ops
import tensorflow as tf
import numpy as np
np.set_printoptions(threshold=np.nan)
from util.Layer import Layers
from util.Train_probability import Training
from util.DataLoader import DataLoader
Sparse = DataLoader()
Layer = Layers()
from util import cube_exp_prob
from util import Probability
from util import Bilinear
from util import corr_prob
import time
sys.path.append(".")
def get_lstm_input(hidden_dim, embeddings, inputs1, inputs2, lengths1,
lengths2, dropout, lstm):
# lstm produce sentence representation, used in all models
Wemb1 = tf.nn.embedding_lookup(embeddings, inputs1)
Wemb2 = tf.nn.embedding_lookup(embeddings, inputs2)
# Wemb1 = tf.Print(Wemb1, [Wemb1], 'term1 embedding')
lstm_output, fstate1 = tf.nn.dynamic_rnn(lstm,
Wemb1,
sequence_length=lengths1,
work_path_name = str(sys.argv[11])
from pathlib import *
work_path = Path('/home/ouyangzhihao/Backup/Exp/ZYY/RandomLR')
total_depth = int(epochs/mini_epoch) #20
# mini_epoch = int(epochs / total_depth) #10
exp_name = '%s_%d_%d_%s_%d_%.4f_%d_%d_%d_%d_%d_MCTS' % (
dataset_name, epochs, batch_size, optimizer, random_range, init_lr, random_num, power,mini_epoch,sub_tree_depth,0.1
)
if ((work_path / work_path_name / 'TB_Logs' / exp_name).exists()):
print('Already Finished!')
exit()
data_loader = DataLoader(dataset_name=dataset_name)
x_train, y_train, x_test, y_test,num_classes = data_loader.load_data()
print('Using real-time data augmentation.')
datagen = ImageDataGenerator(
# set input mean to 0 over the dataset
featurewise_center=False,
# set each sample mean to 0
samplewise_center=False,
# divide inputs by std of dataset
featurewise_std_normalization=False,
# divide each input by its std
samplewise_std_normalization=False,
# apply ZCA whitening
zca_whitening=False,
# epsilon for ZCA whitening
zca_epsilon=1e-06,