def run_op(self, argmap):
modname = argmap.get_str(('modname', 'original'))
seed = argmap.get_int(('seed', 10000))
relmod = importlib.import_module(modname)
print("Going to do Basic Model with module {}".format(modname))
# Load the sample data, and peel off the initial layer of <start_token> data.
origdata = self.load_data(seed)
origdata = origdata[:, 1:]
assert origdata.shape[1] == utility.SAMPLE_LENGTH
enc = utility.get_encoder()
hparams = utility.get_hparams()
with tf.Session(graph=tf.Graph()) as sess:
np.random.seed(seed)
tf.set_random_seed(seed)
tfop = relmod.model_or_sample(origdata.shape[0], origdata)
ckpt = tf.train.latest_checkpoint(
os.path.join('models', utility.MODEL_NAME))
tf.train.Saver().restore(sess, ckpt)
alpha = time.time()
result = sess.run(tfop)
print("Basic Model successful, took {:.03f} seconds".format(
time.time() - alpha))
utility.PickleData(relmod, 'modsample', seed, result=result).save()
def run_op(self, argmap):
modname = argmap.get_str(('modname', 'original'))
seed = argmap.get_int(('seed', 10000))
batch_size = argmap.get_int(('batchsize', 100))
enc = utility.get_encoder()
hparams = utility.get_hparams()
relmod = importlib.import_module(modname)
with tf.Session(graph=tf.Graph()) as sess:
np.random.seed(seed)
tf.set_random_seed(seed)
tfop = relmod.model_or_sample(batch_size)
ckpt = tf.train.latest_checkpoint(
os.path.join('models', utility.MODEL_NAME))
tf.train.Saver().restore(sess, ckpt)
alpha = time.time()
result = sess.run(tfop)
print(
"Sample successful, took {:.03f} seconds".format(time.time() -
alpha))
utility.PickleData(relmod, 'sample', seed, result=result).save()
def run_op(self, argmap):
modname = argmap.get_str(('modname', 'original'))
relmod = importlib.import_module(modname)
print("Going model hard-coded data with module {}".format(modname))
# Load the hard-coded sample data
origdata = utility.get_encoded_sents()
#enc = utility.get_encoder()
hparams = utility.get_hparams()
with tf.Session(graph=tf.Graph()) as sess:
# Notice!!! You don't need these set-seed operations here!!!
# np.random.seed(seed)
# tf.set_random_seed(seed)
tfop = relmod.model_or_sample(origdata.shape[0], origdata)
ckpt = tf.train.latest_checkpoint(
os.path.join('models', utility.MODEL_NAME))
tf.train.Saver().restore(sess, ckpt)
alpha = time.time()
result = sess.run(tfop)
print("Basic Model successful, took {:.03f} seconds".format(
time.time() - alpha))
utility.PickleData(relmod, 'modhcode', 0, result=result).save()
def model_or_sample(batch_size, modeldata=None):
length = utility.SAMPLE_LENGTH if modeldata is None else modeldata.shape[1]
return explicit_loop_bimodel(hparams=utility.get_hparams(),
length=length,
modeldata=modeldata,
start_token=utility.get_start_token()[0],
batch_size=batch_size)
def model_or_sample(batch_size):
return sample_sequence(hparams=utility.get_hparams(),
length=utility.SAMPLE_LENGTH,
start_token=utility.get_start_token()[0],
batch_size=batch_size)