#if set to true, then load model from file instead of start a new model load_model = False #if set to true, use adam optimizer instead of sgd adam_opt = True #batch size for training batch_size = 128 #data params #bucket_option = [i for i in range(1, 20+1)] bucket_option = [5,10,15,20,25,31] buckets = s2s_reader.create_bucket(bucket_option) # load the data set into s2s_reader # the total bucket numbers = bucket options number ^ 2 # if clean mode is true, the leftover data in the bucket will be used before the epoch is over reader = s2s_reader.reader(file_name = file_name, batch_size = batch_size, buckets = buckets, bucket_option = bucket_option, clean_mode=True) vocab_size = len(reader.dict) # if load_model = true, then we need to define the same parameter in the saved_model inorder to load it hidden_size = 512 projection_size = 300 embedding_size = 300 num_layers = 1 # ouput_size for softmax layer output_size = hidden_size if projection_size!=None: output_size = projection_size #training params, truncated_norm will resample x > 2std; so when std = 0.1, the range of x is [-0.2, 0.2] truncated_std = 0.1
file_name = "bbt_data"
# interactive mode allow user to talk to the model directly, if set to false, it will test on the training data instead
interactive = True
# regular expression for parsing user input
expression = r"[0-9]+|[']*[\w]+"
# signal mode allow user to insert signal token before the decoder generate sentence
signal = False
# batch size for testing
batch_size = 1
# data params
# bucket_option = [i for i in xrange(1, 20+1)]
bucket_option = [5, 10, 15, 20, 25, 31]
buckets = s2s_reader.create_bucket(bucket_option)
reader = s2s_reader.reader(file_name=file_name, batch_size=batch_size, buckets=buckets, bucket_option=bucket_option,
signal=signal)
vocab_size = len(reader.dict)
# if load_model = true, then we need to define the same parameter in the saved_model inorder to load it
hidden_size = 512
projection_size = 300
embedding_size = 300
num_layers = 1
# ouput_size for softmax layer
output_size = hidden_size
if projection_size != None:
output_size = projection_size
# model name & save path
model_name = "p" + str(projection_size) + "_h" + str(hidden_size) + "_x" + str(num_layers)
