def decode(self):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely, loading latest checkpoint at regular intervals"""
t0 = time.time()
counter = FLAGS.decode_after
while True:
tf.reset_default_graph()
batch = self._batcher.next_batch() # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info("Decoder has finished reading dataset for single_pass.")
tf.logging.info("Output has been saved in %s and %s. Now starting ROUGE eval...", self._rouge_ref_dir, self._rouge_dec_dir)
results_dict = rouge_eval(self._rouge_ref_dir, self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[0] # list of strings
article_withunks = data.show_art_oovs(original_article, self._vocab) # string
abstract_withunks = data.show_abs_oovs(original_abstract, self._vocab, (batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
# Run beam search to get best Hypothesis
if FLAGS.ac_training:
best_hyp = beam_search.run_beam_search(self._sess, self._model, self._vocab, batch, self._dqn, self._dqn_sess, self._dqn_graph)
else:
best_hyp = beam_search.run_beam_search(self._sess, self._model, self._vocab, batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(output_ids, self._vocab, (batch.art_oovs[0] if FLAGS.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(data.STOP_DECODING) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
if FLAGS.single_pass:
self.write_for_rouge(original_abstract_sents, decoded_words, counter) # write ref summary and decoded summary to file, to eval with pyrouge later
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks, decoded_output) # log output to screen
self.write_for_attnvis(article_withunks, abstract_withunks, decoded_words, best_hyp.attn_dists, best_hyp.p_gens) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1-t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info('We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint', t1-t0)
_ = util.load_ckpt(self._saver, self._sess, FLAGS.decode_from)
t0 = time.time()
def run_eval(model, batcher, vocab):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(eval_dir, 'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
# run eval on the batch
t0=time.time()
results = model.run_eval_step(sess, batch)
t1=time.time()
tf.logging.info('seconds for batch: %.2f', t1-t0)
# print the loss and coverage loss to screen
loss = results['loss']
tf.logging.info('loss: %f', loss)
if FLAGS.coverage:
coverage_loss = results['coverage_loss']
tf.logging.info("coverage_loss: %f", coverage_loss)
# add summaries
summaries = results['summaries']
train_step = results['global_step']
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
running_avg_loss = calc_running_avg_loss(np.asscalar(loss), running_avg_loss, summary_writer, train_step)
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
tf.logging.info('Found new best model with %.3f running_avg_loss. Saving to %s', running_avg_loss, bestmodel_save_path)
saver.save(sess, bestmodel_save_path, global_step=train_step, latest_filename='checkpoint_best')
best_loss = running_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def __init__(self, model, batcher, vocab, dqn = None):
"""Initialize decoder.
Args:
model: a Seq2SeqAttentionModel object.
batcher: a Batcher object.
vocab: Vocabulary object
"""
self._model = model
self._model.build_graph()
self._batcher = batcher
self._vocab = vocab
self._saver = tf.train.Saver() # we use this to load checkpoints for decoding
self._sess = tf.Session(config=util.get_config())
if FLAGS.ac_training:
self._dqn = dqn
self._dqn_graph = tf.Graph()
with self._dqn_graph.as_default():
self._dqn.build_graph()
self._dqn_saver = tf.train.Saver() # we use this to load checkpoints for decoding
self._dqn_sess = tf.Session(config=util.get_config())
_ = util.load_dqn_ckpt(self._dqn_saver, self._dqn_sess)
# Load an initial checkpoint to use for decoding
ckpt_path = util.load_ckpt(self._saver, self._sess, FLAGS.decode_from)
if FLAGS.single_pass:
# Make a descriptive decode directory name
ckpt_name = "{}-ckpt-".format(FLAGS.decode_from) + ckpt_path.split('-')[
-1] # this is something of the form "ckpt-123456"
self._decode_dir = os.path.join(FLAGS.log_root, get_decode_dir_name(ckpt_name))
else: # Generic decode dir name
self._decode_dir = os.path.join(FLAGS.log_root, "decode")
# Make the decode dir if necessary
if not os.path.exists(self._decode_dir): os.mkdir(self._decode_dir)
if FLAGS.single_pass:
# Make the dirs to contain output written in the correct format for pyrouge
self._rouge_ref_dir = os.path.join(self._decode_dir, "reference")
if not os.path.exists(self._rouge_ref_dir): os.mkdir(self._rouge_ref_dir)
self._rouge_dec_dir = os.path.join(self._decode_dir, "decoded")
if not os.path.exists(self._rouge_dec_dir): os.mkdir(self._rouge_dec_dir)
def convert_to_coverage_model():
"""Load non-coverage checkpoint, add initialized extra variables for coverage, and save as new checkpoint"""
tf.logging.info("converting non-coverage model to coverage model..")
# initialize an entire coverage model from scratch
sess = tf.Session(config=util.get_config())
print "initializing everything..."
sess.run(tf.global_variables_initializer())
# load all non-coverage weights from checkpoint
saver = tf.train.Saver([v for v in tf.global_variables() if "coverage" not in v.name and "Adagrad" not in v.name])
print "restoring non-coverage variables..."
curr_ckpt = util.load_ckpt(saver, sess)
print "restored."
# save this model and quit
new_fname = curr_ckpt + '_cov_init'
print "saving model to %s..." % (new_fname)
new_saver = tf.train.Saver() # this one will save all variables that now exist
new_saver.save(sess, new_fname)
print "saved."
exit()
def restore_best_model():
"""Load bestmodel file from eval directory, add variables for adagrad, and save to train directory"""
tf.logging.info("Restoring bestmodel for training...")
# Initialize all vars in the model
sess = tf.Session(config=util.get_config())
print "Initializing all variables..."
sess.run(tf.initialize_all_variables())
# Restore the best model from eval dir
saver = tf.train.Saver([v for v in tf.all_variables() if "Adagrad" not in v.name])
print "Restoring all non-adagrad variables from best model in eval dir..."
curr_ckpt = util.load_ckpt(saver, sess, "eval")
print "Restored %s." % curr_ckpt
# Save this model to train dir and quit
new_model_name = curr_ckpt.split("/")[-1].replace("bestmodel", "model")
new_fname = os.path.join(FLAGS.log_root, "train", new_model_name)
print "Saving model to %s..." % (new_fname)
new_saver = tf.train.Saver() # this saver saves all variables that now exist, including Adagrad variables
new_saver.save(sess, new_fname)
print "Saved."
exit()
def __init__(self, model, batcher, vocab):
"""Initialize decoder.
Args:
model: a Seq2SeqAttentionModel object.
batcher: a Batcher object.
vocab: Vocabulary object
"""
self._model = model
self._model.build_graph()
self._batcher = batcher
self._vocab = vocab
self._saver = tf.train.Saver() # we use this to load checkpoints for decoding
self._sess = tf.Session(config=util.get_config())
# Load an initial checkpoint to use for decoding
ckpt_path = util.load_ckpt(self._saver, self._sess)
if FLAGS.single_pass:
# Make a descriptive decode directory name
ckpt_name = "ckpt-" + ckpt_path.split('-')[-1] # this is something of the form "ckpt-123456"
self._decode_dir = os.path.join(FLAGS.log_root, get_decode_dir_name(ckpt_name))
if os.path.exists(self._decode_dir):
raise Exception("single_pass decode directory %s should not already exist" % self._decode_dir)
else: # Generic decode dir name
self._decode_dir = os.path.join(FLAGS.log_root, "decode")
# Make the decode dir if necessary
if not os.path.exists(self._decode_dir): os.mkdir(self._decode_dir)
if FLAGS.single_pass:
# Make the dirs to contain output written in the correct format for pyrouge
self._rouge_ref_dir = os.path.join(self._decode_dir, "reference")
if not os.path.exists(self._rouge_ref_dir): os.mkdir(self._rouge_ref_dir)
self._rouge_dec_dir = os.path.join(self._decode_dir, "decoded")
if not os.path.exists(self._rouge_dec_dir): os.mkdir(self._rouge_dec_dir)
def __init__(self, model, vocab,single_pass,hps,pointer_gen,log_root):
"""Initialize decoder.
Args:
model: a Seq2SeqAttentionModel object.
batcher: a Batcher object.
vocab: Vocabulary object
"""
self._model = model
self._model.build_graph()
# self._batcher = batcher
self._vocab = vocab
self._saver = tf.train.Saver() # we use this to load checkpoints for decoding
self._sess = tf.Session(config=util.get_config())
self.single_pass=single_pass
self. max_dec_steps=hps.max_dec_steps
self.min_dec_steps=hps.min_dec_steps
self.max_dec_steps=hps.max_dec_steps
self.beam_size=hps.beam_size
self.pointer_gen=pointer_gen
# Load an initial checkpoint to use for decoding
ckpt_path = util.load_ckpt(self._saver, self._sess,log_root)
print ckpt_path
def restore_best_model():
"""Load bestmodel file from eval directory, add variables for adagrad, and save to train directory"""
tf.logging.info("Restoring bestmodel for training...")
# Initialize all vars in the model
sess = tf.Session(config=util.get_config())
print("Initializing all variables...")
sess.run(tf.initialize_all_variables())
# Restore the best model from eval dir
saver = tf.train.Saver(
[v for v in tf.all_variables() if "Adagrad" not in v.name])
print("Restoring all non-adagrad variables from best model in eval dir...")
curr_ckpt = util.load_ckpt(saver, sess, "eval")
print("Restored %s." % curr_ckpt)
# Save this model to train dir and quit
new_model_name = curr_ckpt.split("/")[-1].replace("bestmodel", "model")
new_fname = os.path.join(FLAGS.log_root, "train", new_model_name)
print("Saving model to %s..." % (new_fname))
new_saver = tf.train.Saver(
) # this saver saves all variables that now exist, including Adagrad variables
new_saver.save(sess, new_fname)
print("Saved.")
exit()
def run_decoding(model, batcher):
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
decode_dir = os.path.join(
FLAGS.exp_name,
"decode") # make a subdir of the root dir for eval data
batches = batcher.getBatches()
_ = util.load_ckpt(saver, sess) # load a new checkpoint
epoch_avg_loss = 0.
epoch_decode_steps = 0
for batch in batches:
results = model.run_decode_step(sess, batch)
loss = results['loss']
epoch_decode_steps += 1
epoch_avg_loss = (epoch_avg_loss * (epoch_decode_steps - 1.) +
loss) / epoch_decode_steps
print("Average loss %f" % (epoch_avg_loss))
def main(unused_argv):
if len(unused_argv
) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(
tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if "train" in FLAGS.mode:
os.makedirs(FLAGS.log_root)
else:
raise Exception(
"Logdir %s doesn't exist. Run in train mode to create it." %
(FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# print('FLAGS.flag_values_dict() ->', FLAGS.flag_values_dict())
flags_dict = FLAGS.flag_values_dict()
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = [
'mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag',
'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim',
'batch_size', 'max_dec_sen_num', 'max_dec_steps', 'max_enc_steps'
]
hps_dict = {}
for key, val in flags_dict.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
print('hps_dict ->', json.dumps(hps_dict, ensure_ascii=False))
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size',
'max_enc_sen_num', 'max_enc_seq_len'
]
hps_dict = {}
for key, val in flags_dict.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
# # test
# model_dis = Discriminator(hps_discriminator, vocab)
# model_dis.build_graph()
# sys.exit(0)
# # test
print('before load batcher...')
# Create a batcher object that will create minibatches of data
batcher = GenBatcher(vocab, hps_generator)
print('after load batcher...')
tf.set_random_seed(111) # a seed value for randomness
if hps_generator.mode == 'adversarial_train':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab,
"discriminator_train/positive/*",
"discriminator_train/negative/*",
"discriminator_test/positive/*",
"discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(
model_dis)
util.load_ckpt(saver_dis, sess_dis, ckpt_dir="train-discriminator")
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
if not os.path.exists("MLE"): os.mkdir("MLE")
print("evaluate the diversity of MLE (decode based on sampling)")
generated.generator_test_sample_example("MLE/" + "MLE_sample_positive",
"MLE/" + "MLE_sample_negative",
200)
print(
"evaluate the diversity of MLE (decode based on max probability)")
generated.generator_test_max_example("MLE/" + "MLE_max_temp_positive",
"MLE/" + "MLE_max_temp_negative",
200)
print("Start adversarial training......")
if not os.path.exists("train_sample_generated"):
os.mkdir("train_sample_generated")
if not os.path.exists("test_max_generated"):
os.mkdir("test_max_generated")
if not os.path.exists("test_sample_generated"):
os.mkdir("test_sample_generated")
whole_decay = False
for epoch in range(10):
batches = batcher.get_batches(mode='train')
for step in range(int(len(batches) / 1000)):
run_train_generator(
model, model_dis, sess_dis, batcher, dis_batcher,
batches[step * 1000:(step + 1) * 1000], sess_ge, saver_ge,
train_dir_ge, generated
) # (model, discirminator_model, discriminator_sess, batcher, dis_batcher, batches, sess, saver, train_dir, generated):
generated.generator_sample_example(
"train_sample_generated/" + str(epoch) + "epoch_step" +
str(step) + "_temp_positive", "train_sample_generated/" +
str(epoch) + "epoch_step" + str(step) + "_temp_negative",
1000)
# generated.generator_max_example("max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_positive", "max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_negetive", 200)
tf.logging.info("test performance: ")
tf.logging.info("epoch: " + str(epoch) + " step: " + str(step))
print(
"evaluate the diversity of DP-GAN (decode based on max probability)"
)
generated.generator_test_sample_example(
"test_sample_generated/" + str(epoch) + "epoch_step" +
str(step) + "_temp_positive", "test_sample_generated/" +
str(epoch) + "epoch_step" + str(step) + "_temp_negative",
200)
print(
"evaluate the diversity of DP-GAN (decode based on sampling)"
)
generated.generator_test_max_example(
"test_max_generated/" + str(epoch) + "epoch_step" +
str(step) + "_temp_positive", "test_max_generated/" +
str(epoch) + "epoch_step" + str(step) + "_temp_negative",
200)
dis_batcher.train_queue = []
dis_batcher.train_queue = []
for i in range(epoch + 1):
for j in range(step + 1):
dis_batcher.train_queue += dis_batcher.fill_example_queue(
"train_sample_generated/" + str(i) + "epoch_step" +
str(j) + "_temp_positive/*")
dis_batcher.train_queue += dis_batcher.fill_example_queue(
"train_sample_generated/" + str(i) + "epoch_step" +
str(j) + "_temp_negative/*")
dis_batcher.train_batch = dis_batcher.create_batches(
mode="train", shuffleis=True)
# dis_batcher.valid_batch = dis_batcher.train_batch
whole_decay = run_train_discriminator(
model_dis, 5, dis_batcher,
dis_batcher.get_batches(mode="train"), sess_dis, saver_dis,
train_dir_dis, whole_decay)
elif hps_generator.mode == 'train_generator':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
print("Start pre-training generator......")
# this is an infinite loop until
run_pre_train_generator(model, batcher, 10, sess_ge, saver_ge,
train_dir_ge, generated)
print("Generating negative examples......")
generated.generator_train_negative_example()
generated.generator_test_negative_example()
elif hps_generator.mode == 'train_discriminator':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
# util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab,
"discriminator_train/positive/*",
"discriminator_train/negative/*",
"discriminator_test/positive/*",
"discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(
model_dis)
print("Start pre-training discriminator......")
# run_test_discriminator(model_dis, dis_batcher, sess_dis, saver_dis, "test")
if not os.path.exists("discriminator_result"):
os.mkdir("discriminator_result")
run_pre_train_discriminator(model_dis, dis_batcher, 25, sess_dis,
saver_dis, train_dir_dis)
def run_eval_pred(model, batcher, vocab):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(
FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
while np.sum(batch.rst_batch) == FLAGS.batch_size or np.sum(
batch.rst_batch) == 0:
batch = batcher.next_batch()
# run eval on the batch
t0 = time.time()
results = model.run_predict_eval_step(sess, batch)
t1 = time.time()
tf.logging.info('seconds for batch: %.2f', t1 - t0)
# print the loss and coverage loss to screen
loss = results['pred_loss']
tf.logging.info('loss: %f', loss)
# add summaries
summaries = results['summaries']
train_step = results['global_step']
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
running_avg_loss = calc_running_avg_loss(np.asscalar(loss),
running_avg_loss,
summary_writer, train_step)
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
tf.logging.info(
'Found new best model with %.3f running_avg_loss. Saving to %s',
running_avg_loss, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = running_avg_loss
# y_true = np.array(results['rst'])
# y_pred = np.array(results['pred'])
#
# print(sklearn.metrics.classification_report(y_true, y_pred))
# print(confusion_matrix(y_true, y_pred))
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def run_eval(model, batcher):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(
FLAGS.log_root,
"eval_loss") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
train_dir = os.path.join(FLAGS.log_root, "train")
while True:
ckpt_state = tf.train.get_checkpoint_state(train_dir)
tf.logging.info('max_enc_steps: %d, max_dec_steps: %d',
FLAGS.max_enc_steps, FLAGS.max_dec_steps)
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
# run eval on the batch
t0 = time.time()
results = model.run_eval_step(sess, batch)
t1 = time.time()
tf.logging.info('seconds for batch: %.2f', t1 - t0)
# print the loss and coverage loss to screen
loss = results['loss']
tf.logging.info('loss: %f', loss)
train_step = results['global_step']
tf.logging.info("pgen_avg: %f", results['p_gen_avg'])
if FLAGS.coverage:
tf.logging.info("coverage_loss: %f", results['coverage_loss'])
# add summaries
summaries = results['summaries']
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
running_avg_loss = util.calc_running_avg_loss(np.asscalar(loss),
running_avg_loss,
summary_writer,
train_step,
'running_avg_loss')
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
tf.logging.info(
'Found new best model with %.3f running_avg_loss. Saving to %s',
running_avg_loss, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = running_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def run_training(model, batcher, sess_context_manager, sv, summary_writer, word_vector, \
selector_saver=None, rewriter_saver=None, all_saver=None):
"""Repeatedly runs training iterations, logging loss to screen and writing summaries"""
tf.logging.info("starting run_training")
train_step = 0
ckpt_path = os.path.join(FLAGS.log_root, "train", "model.ckpt_cov")
with sess_context_manager as sess:
if FLAGS.pretrained_selector_path: # Load the pre-trained model
tf.logging.info('Loading selector model')
_ = util.load_ckpt(selector_saver,
sess,
ckpt_path=FLAGS.pretrained_selector_path)
if FLAGS.pretrained_rewriter_path:
tf.logging.info('Loading rewriter model')
_ = util.load_ckpt(rewriter_saver,
sess,
ckpt_path=FLAGS.pretrained_rewriter_path)
for _ in range(FLAGS.max_train_iter): # repeats until interrupted
batch = batcher.next_batch()
tf.logging.info('running training step...')
t0 = time.time()
results = model.run_train_step(sess, batch, word_vector,
train_step)
t1 = time.time()
tf.logging.info('seconds for training step: %.3f', t1 - t0)
loss = results['loss']
tf.logging.info('rl_loss: %f', loss) # print the loss to screen
train_step = results['global_step']
if not np.isfinite(loss):
raise Exception("Loss is not finite. Stopping.")
tf.logging.info("reinforce_avg_logprobs: %f",
results['reinforce_avg_logprobs'])
if FLAGS.coverage:
tf.logging.info("coverage_loss: %f", results['coverage_loss']
) # print the coverage loss to screen
tf.logging.info("reinforce_coverage_loss: %f",
results['reinforce_coverage_loss'])
if FLAGS.inconsistent_loss:
tf.logging.info('inconsistent_loss: %f',
results['inconsist_loss'])
tf.logging.info("selector_loss: %f", results['selector_loss'])
recall, ratio, _ = util.get_batch_ratio(
batch.original_articles_sents, batch.original_extracts_ids,
results['probs'])
write_to_summary(ratio, 'SentSelector/select_ratio/recall=0.9',
train_step, summary_writer)
# get the summaries and iteration number so we can write summaries to tensorboard
summaries = results[
'summaries'] # we will write these summaries to tensorboard using summary_writer
summary_writer.add_summary(summaries,
train_step) # write the summaries
if train_step % 100 == 0: # flush the summary writer every so often
summary_writer.flush()
if train_step % FLAGS.save_model_every == 0:
if FLAGS.pretrained_selector_path and FLAGS.pretrained_rewriter_path:
all_saver.save(sess, ckpt_path, global_step=train_step)
else:
sv.saver.save(sess, ckpt_path, global_step=train_step)
print('Step: ', train_step)
def run_decode(model, batcher, vocab):
print "build graph..."
model.build_graph()
saver = tf.train.Saver(max_to_keep=3)
sess = tf.Session(config=util.get_config())
saver = tf.train.Saver()
ckpt_path = util.load_ckpt(saver, sess)
if FLAGS.single_pass:
ckpt_name = "ckpt-" + ckpt_path.split('-')[-1]
dirname = "decode_maxenc_%ibeam_%imindec_%imaxdec_%i" % (FLAGS.max_enc_steps, FLAGS.beam_size, FLAGS.min_dec_steps, FLAGS.max_dec_steps)
decode_dir = os.path.join(FLAGS.log_root, dirname + ckpt_name)
if os.path.exists(decode_dir):
raise Exception('single_pass decode directory %s should not exist', decode_dir)
else:
decode_dir = os.path.join(FLAGS.log_root, 'decode')
if not os.path.exists(decode_dir): os.mkdir(decode_dir)
if FLAGS.single_pass:
rouge_ref_dir = os.path.join(decode_dir, "reference")
if not os.path.exists(rouge_ref_dir): os.mkdir(rouge_ref_dir)
rouge_dec_dir = os.path.join(decode_dir, "decoded")
if not os.path.exists(rouge_dec_dir): os.mkdir(rouge_dec_dir)
counter = 0
t0 = time.time()
while True:
batch = batcher.next_batch() # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
print("Decoder has finished reading dataset for single_pass.")
print("Output has been saved in %s and %s. Now starting ROUGE eval...", rouge_ref_dir,
rouge_dec_dir)
results_dict = rouge_eval(rouge_ref_dir, rouge_dec_dir)
rouge_log(results_dict, decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[0] # list of strings
article_withunks = data.show_art_oovs(original_article, vocab) # string
abstract_withunks = data.show_abs_oovs(original_abstract, vocab, None) # string
# Run beam search to get best Hypothesis
output = model.run_beam_decode_step(sess, batch, vocab)
output_ids = [int(t) for t in output]
decoded_words = data.outputids2words(output_ids, vocab, None)
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(data.STOP_DECODING) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
if FLAGS.single_pass:
write_for_rouge(original_abstract_sents, decoded_words, counter, rouge_ref_dir, rouge_dec_dir) # write ref summary and decoded summary to file, to eval with pyrouge later
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks, decoded_output) # log output to screen
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info('We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(saver, sess)
t0 = time.time()
def run_eval(model, batcher, word_vector):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
if FLAGS.embedding:
sess.run(tf.global_variables_initializer()
) #, feed_dict={model.embedding_place: word_vector}
eval_dir = os.path.join(
FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir) # 指定一个文件用来保存图
running_avg_ratio = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_ratio = None # will hold the best loss achieved so far
train_dir = os.path.join(FLAGS.log_root, "train")
while True:
ckpt_state = tf.train.get_checkpoint_state(
train_dir) # get the info of checkpoint file
#tf.logging.info('max_enc_steps: %d, max_dec_steps: %d', FLAGS.max_enc_steps, FLAGS.max_dec_steps)
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
# run eval on the batch
t0 = time.time()
results = model.run_eval_step(sess, batch)
t1 = time.time()
tf.logging.info('seconds for batch: %.2f', t1 - t0)
# print the loss and coverage loss to screen
loss = results['loss']
tf.logging.info('loss: %f', loss)
train_step = results['global_step']
recall, ratio, _ = util.get_batch_ratio(batch.original_articles_sents, \
batch.original_extracts_ids, results['probs'])
write_to_summary(ratio, 'SentSelector/select_ratio/recall=0.9',
train_step, summary_writer)
# add summaries
summaries = results['summaries']
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
running_avg_ratio = util.calc_running_avg_loss(ratio,
running_avg_ratio,
summary_writer,
train_step,
'running_avg_ratio')
print("run_avg_ratio: ", running_avg_ratio)
tf.log("run_avg_ratio: ", running_avg_ratio)
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_ratio is None or running_avg_ratio < best_ratio:
tf.logging.info(
'Found new best model with %.3f running_avg_ratio. Saving to %s',
running_avg_ratio, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_ratio = running_avg_ratio
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def main(unused_argv):
if len(unused_argv
) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(
tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if FLAGS.mode == "train":
os.makedirs(FLAGS.log_root)
else:
raise Exception(
"Logdir %s doesn't exist. Run in train mode to create it." %
(FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = [
'mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag',
'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim',
'batch_size', 'max_dec_steps', 'max_enc_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size', 'max_dec_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
tf.set_random_seed(
111
) # a seed value for randomness # train-classification train-sentiment train-cnn-classificatin train-generator
if FLAGS.mode == "train-classifier":
#print("Start pre-training......")
model_class = Classification(hps_discriminator, vocab)
cla_batcher = ClaBatcher(hps_discriminator, vocab)
sess_cls, saver_cls, train_dir_cls = setup_training_classification(
model_class)
print("Start pre-training classification......")
run_pre_train_classification(model_class, cla_batcher, 1, sess_cls,
saver_cls, train_dir_cls) #10
generated = Generate_training_sample(model_class, vocab, cla_batcher,
sess_cls)
print("Generating training examples......")
generated.generate_training_example("train")
generated.generate_test_example("test")
elif FLAGS.mode == "train-sentimentor":
model_class = Classification(hps_discriminator, vocab)
cla_batcher = ClaBatcher(hps_discriminator, vocab)
sess_cls, saver_cls, train_dir_cls = setup_training_classification(
model_class)
print("Start pre_train_sentimentor......")
model_sentiment = Sentimentor(hps_generator, vocab)
sentiment_batcher = SenBatcher(hps_generator, vocab)
sess_sen, saver_sen, train_dir_sen = setup_training_sentimentor(
model_sentiment)
util.load_ckpt(saver_cls, sess_cls, ckpt_dir="train-classification")
run_pre_train_sentimentor(model_sentiment, sentiment_batcher, 1,
sess_sen, saver_sen, train_dir_sen) #1
elif FLAGS.mode == "test":
config = {
'n_epochs': 5,
'kernel_sizes': [3, 4, 5],
'dropout_rate': 0.5,
'val_split': 0.4,
'edim': 300,
'n_words': None, # Leave as none
'std_dev': 0.05,
'sentence_len': 50,
'n_filters': 100,
'batch_size': 50
}
config['n_words'] = 50000
cla_cnn_batcher = CNN_ClaBatcher(hps_discriminator, vocab)
cnn_classifier = CNN(config)
sess_cnn_cls, saver_cnn_cls, train_dir_cnn_cls = setup_training_cnnclassifier(
cnn_classifier)
#util.load_ckpt(saver_cnn_cls, sess_cnn_cls, ckpt_dir="train-cnnclassification")
run_train_cnn_classifier(cnn_classifier, cla_cnn_batcher, 1,
sess_cnn_cls, saver_cnn_cls,
train_dir_cnn_cls) #1
files = os.listdir("test-generate-transfer/")
for file_ in files:
run_test_our_method(cla_cnn_batcher, cnn_classifier, sess_cnn_cls,
"test-generate-transfer/" + file_ + "/*")
#elif FLAGS.mode == "test":
elif FLAGS.mode == "train-generator":
model_class = Classification(hps_discriminator, vocab)
cla_batcher = ClaBatcher(hps_discriminator, vocab)
sess_cls, saver_cls, train_dir_cls = setup_training_classification(
model_class)
model_sentiment = Sentimentor(hps_generator, vocab)
sentiment_batcher = SenBatcher(hps_generator, vocab)
sess_sen, saver_sen, train_dir_sen = setup_training_sentimentor(
model_sentiment)
config = {
'n_epochs': 5,
'kernel_sizes': [3, 4, 5],
'dropout_rate': 0.5,
'val_split': 0.4,
'edim': 300,
'n_words': None, # Leave as none
'std_dev': 0.05,
'sentence_len': 50,
'n_filters': 100,
'batch_size': 50
}
config['n_words'] = 50000
cla_cnn_batcher = CNN_ClaBatcher(hps_discriminator, vocab)
cnn_classifier = CNN(config)
sess_cnn_cls, saver_cnn_cls, train_dir_cnn_cls = setup_training_cnnclassifier(
cnn_classifier)
model = Generator(hps_generator, vocab)
batcher = GenBatcher(vocab, hps_generator)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
util.load_ckpt(saver_cnn_cls,
sess_cnn_cls,
ckpt_dir="train-cnnclassification")
util.load_ckpt(saver_sen, sess_sen, ckpt_dir="train-sentimentor")
generated = Generated_sample(model, vocab, batcher, sess_ge)
print("Start pre-training generator......")
run_pre_train_generator(model, batcher, 1, sess_ge, saver_ge,
train_dir_ge, generated, cla_cnn_batcher,
cnn_classifier, sess_cnn_cls) # 4
generated.generate_test_negetive_example(
"temp_negetive",
batcher) # batcher, model_class, sess_cls, cla_batcher
generated.generate_test_positive_example("temp_positive", batcher)
#run_test_our_method(cla_cnn_batcher, cnn_classifier, sess_cnn_cls,
# "temp_negetive" + "/*")
loss_window = 0
t0 = time.time()
print("begin reinforcement learning:")
for epoch in range(30):
batches = batcher.get_batches(mode='train')
for i in range(len(batches)):
current_batch = copy.deepcopy(batches[i])
sentiment_batch = batch_sentiment_batch(
current_batch, sentiment_batcher)
result = model_sentiment.max_generator(sess_sen,
sentiment_batch)
weight = result['generated']
current_batch.weight = weight
sentiment_batch.weight = weight
cla_batch = batch_classification_batch(current_batch, batcher,
cla_batcher)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
cc = SmoothingFunction()
reward_sentiment = 1 - np.abs(0.5 - result['y_pred_auc'])
reward_BLEU = []
for k in range(FLAGS.batch_size):
reward_BLEU.append(
sentence_bleu(
[current_batch.original_reviews[k].split()],
cla_batch.original_reviews[k].split(),
smoothing_function=cc.method1))
reward_BLEU = np.array(reward_BLEU)
reward_de = (2 / (1.0 / (1e-6 + reward_sentiment) + 1.0 /
(1e-6 + reward_BLEU)))
result = model.run_train_step(sess_ge, current_batch)
train_step = result[
'global_step'] # we need this to update our running average loss
loss = result['loss']
loss_window += loss
if train_step % 100 == 0:
t1 = time.time()
tf.logging.info(
'seconds for %d training generator step: %.3f ',
train_step, (t1 - t0) / 100)
t0 = time.time()
tf.logging.info('loss: %f', loss_window /
100) # print the loss to screen
loss_window = 0.0
if train_step % 10000 == 0:
generated.generate_test_negetive_example(
"test-generate-transfer/" + str(epoch) + "epoch_step" +
str(train_step) + "_temp_positive", batcher)
generated.generate_test_positive_example(
"test-generate/" + str(epoch) + "epoch_step" +
str(train_step) + "_temp_positive", batcher)
#saver_ge.save(sess, train_dir + "/model", global_step=train_step)
#run_test_our_method(cla_cnn_batcher, cnn_classifier, sess_cnn_cls,
# "test-generate-transfer/" + str(epoch) + "epoch_step" + str(
# train_step) + "_temp_positive" + "/*")
cla_batch, bleu = output_to_classification_batch(
result['generated'], current_batch, batcher, cla_batcher,
cc)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
reward_result_sentiment = result['y_pred_auc']
reward_result_bleu = np.array(bleu)
reward_result = (2 / (1.0 /
(1e-6 + reward_result_sentiment) + 1.0 /
(1e-6 + reward_result_bleu)))
current_batch.score = 1 - current_batch.score
result = model.max_generator(sess_ge, current_batch)
cla_batch, bleu = output_to_classification_batch(
result['generated'], current_batch, batcher, cla_batcher,
cc)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
reward_result_transfer_sentiment = result['y_pred_auc']
reward_result_transfer_bleu = np.array(bleu)
reward_result_transfer = (
2 / (1.0 /
(1e-6 + reward_result_transfer_sentiment) + 1.0 /
(1e-6 + reward_result_transfer_bleu)))
#tf.logging.info("reward_nonsentiment: "+str(reward_sentiment) +" output_original_sentiment: "+str(reward_result_sentiment)+" output_original_bleu: "+str(reward_result_bleu))
reward = reward_result_transfer #reward_de + reward_result_sentiment +
#tf.logging.info("reward_de: "+str(reward_de))
model_sentiment.run_train_step(sess_sen, sentiment_batch,
reward)
def decode(self):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely,
loading latest checkpoint at regular intervals"""
t0 = time.time()
counter = 0
if FLAGS.decode_bleu:
ref_file = os.path.join(self._bleu_dec_dir, "reference.txt")
decoded_file = os.path.join(self._bleu_dec_dir, "decoded.txt")
if os.path.exists(decoded_file):
tf.logging.info('正在删除 %s', decoded_file)
os.remove(decoded_file)
if os.path.exists(ref_file):
tf.logging.info('正在删除 %s', ref_file)
os.remove(ref_file)
while True:
batch = self._batcher.next_batch(
) # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info(
"Decoder has finished reading dataset for single_pass.")
if FLAGS.decode_rouge:
tf.logging.info(
"Output has been saved in %s and %s. Now starting ROUGE eval...",
self._rouge_ref_dir, self._rouge_dec_dir)
try:
t0 = time.time()
results_dict = rouge_eval(self._rouge_ref_dir,
self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
t1 = time.time()
tf.logging.info(
'calculate Rouge score cost %d seconds', t1 - t0)
except Exception as e:
tf.logging.error('计算ROUGE出错 %s', e)
if FLAGS.decode_bleu:
ref_file = os.path.join(self._bleu_dec_dir,
"reference.txt")
decoded_file = os.path.join(self._bleu_dec_dir,
"decoded.txt")
t0 = time.time()
bleu, bleu1, bleu2, bleu3, bleu4 = calcu_bleu(
decoded_file, ref_file)
sys_bleu = sys_bleu_file(decoded_file, ref_file)
sys_bleu_perl = sys_bleu_perl_file(decoded_file, ref_file)
t1 = time.time()
tf.logging.info(bcolors.HEADER +
'-----------BLEU SCORE-----------' +
bcolors.ENDC)
tf.logging.info(
bcolors.OKGREEN + '%f \t %f \t %f \t %f \t %f' +
bcolors.ENDC, bleu, bleu1, bleu2, bleu3, bleu4)
tf.logging.info(
bcolors.OKGREEN + 'sys_bleu %f' + bcolors.ENDC,
sys_bleu)
tf.logging.info(
bcolors.OKGREEN + 'sys_bleu_perl %s' + bcolors.ENDC,
sys_bleu_perl)
tf.logging.info(bcolors.HEADER +
'-----------BLEU SCORE-----------' +
bcolors.ENDC)
tf.logging.info('calculate BLEU score cost %d seconds',
t1 - t0)
break
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[
0] # list of strings
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
abstract_withunks = data.show_abs_oovs(
original_abstract, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
# Run beam search to get best Hypothesis
best_hyp = beam_search.run_beam_search(self._sess, self._model,
self._vocab, batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(
data.STOP_DECODING) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ''.join(decoded_words) # single string
if FLAGS.single_pass:
print_results(article_withunks, abstract_withunks,
decoded_output, counter) # log output to screen
if FLAGS.decode_rouge:
self.write_for_rouge(
original_abstract_sents, decoded_words, counter
) # write ref summary and decoded summary to file, to eval with pyrouge later
if FLAGS.decode_bleu:
self.write_for_bleu(original_abstract_sents, decoded_words)
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks,
decoded_output) # log output to screen
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens
) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info(
'We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(self._saver, self._sess)
t0 = time.time()
def decode(self, output_dir=None):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely, loading latest checkpoint at regular intervals"""
t0 = time.time()
counter = 0
idx = 0
# used to store values during decoding in a list each
outut_str = ""
beam_search_str = ""
metadata = []
# evaluate over a fixed number of test set
while True: #idx <=100 :
print("[%d]" % idx)
batch = self._batcher.next_batch(
) # 1 example repeated across batch
# if idx < 11000:
# idx += 1
# continue
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info(
"Decoder has finished reading dataset for single_pass.")
tf.logging.info(
"Output has been saved in %s and %s. Now starting ROUGE eval...",
self._rouge_ref_dir, self._rouge_dec_dir)
results_dict = rouge_eval(self._rouge_ref_dir,
self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[
0] # list of strings
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
abstract_withunks = data.show_abs_oovs(
original_abstract, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
# Run beam search to get all the Hypothesis
all_hyp = beam_search.run_beam_search(
self._sess, self._model, self._vocab, batch, counter,
self._lm_model, self._lm_word2idx, self._lm_idx2word
) #TODO changed the method signature just to look at the outputs of beam search
if FLAGS.save_values:
for h in all_hyp:
output_ids = [int(t) for t in h.tokens[1:]]
search_str = str(
data.outputids2words(output_ids, self._vocab,
(batch.art_oovs[0]
if FLAGS.pointer_gen else None)))
beam_search_str += search_str
beam_search_str += "\n"
# Extract the get best Hypothesis
best_hyp = all_hyp[0]
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None))
metadata.append(decoded_words)
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(
data.STOP_DECODING) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
###########################
#print best hyp statistics
hyp_stat = ""
# log prob
hyp_stat += "\navg log prob: %s.\n" % best_hyp.avg_log_prob
# words overlap with article: this is buggy
# tri, bi, uni = word_overlap.gram_search(ngrams(nltk.pos_tag(article_withunks.strip().split()), 3), ngrams(nltk.pos_tag(decoded_output.strip().split()), 3))
# hyp_stat += "trigram overlap: %s. bigram overlap: %s. unigram overlap: %s.\n"%(uni, bi, tri)
print_statistics.get_overlap(article_withunks.strip(),
decoded_output.strip(),
match_count=self.overlap_dict)
hyp_stat += "word overlap: "
for key, value in self.overlap_dict.iteritems():
hyp_stat += "\n%d-gram avg overlap: %d" % (key, value /
(counter + 1))
# num sentences and avg length
self.total_nsentence += len(decoded_output.strip().split(
".")) #sentences are seperated by "."
self.total_length += len(decoded_output.strip().split())
avg_nsentence, avg_length = self.total_nsentence / (
counter + 1), self.total_length / (counter + 1)
hyp_stat += "\nnum sentences: %s. avg len: %s.\n" % (avg_nsentence,
avg_length)
# entropy??
if FLAGS.print_info:
print(hyp_stat)
###########################
# saves data into numpy files for analysis
if FLAGS.save_values:
save_decode_data.save_data_iteration(self._decode_dir, counter,
best_hyp)
if FLAGS.single_pass: #change to counter later
self.write_for_rouge(
original_abstract_sents, decoded_words, counter
) # write ref summary and decoded summary to file, to eval with pyrouge later
# writing all the output combined to a file
if FLAGS.print_info:
output = '\nARTICLE: %s\n REFERENCE SUMMARY: %s\n' 'GENERATED SUMMARY: %s\n' % (
article_withunks, abstract_withunks, decoded_output)
print(output)
outut_str += output
# Leena: modifying this to save more stuff
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens, counter,
best_hyp.log_prob, best_hyp.avg_log_prob,
best_hyp.average_pgen) #change to counter later
counter += 1 # this is how many examples we've decoded
else: #Leena: I use the above condition so might have neglected making change to the below condition
print_results(article_withunks, abstract_withunks,
decoded_output) # log output to screen
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens,
counter) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info(
'We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(self._saver, self._sess)
t0 = time.time()
idx += 1
#Leena: saving entire output and beam output as a string to write to a file
if FLAGS.save_values:
save_decode_data.save_data_once(self._decode_dir,
FLAGS.result_path, outut_str,
beam_search_str, metadata)
def decode(self):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely, loading latest checkpoint at regular intervals"""
if not FLAGS.generate:
t0 = time.time()
counter = 0
while True:
batch = self._batcher.next_batch(
) # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info(
"Decoder has finished reading dataset for single_pass."
)
tf.logging.info(
"Output has been saved in %s and %s. Now starting ROUGE eval...",
self._rouge_ref_dir, self._rouge_dec_dir)
results_dict = rouge_eval(self._rouge_ref_dir,
self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[
0] # list of strings
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
abstract_withunks = data.show_abs_oovs(
original_abstract, self._vocab,
(batch.art_oovs[0]
if FLAGS.pointer_gen else None)) # string
# Run beam search to get best Hypothesis
best_hyp = beam_search.run_beam_search(self._sess, self._model,
self._vocab, batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(
data.STOP_DECODING
) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
if FLAGS.single_pass:
# write ref summary and decoded summary to file, to eval with pyrouge later
self.write_for_rouge(original_abstract_sents,
decoded_words, counter)
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks,
decoded_output) # log output to screen
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens
) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info(
'We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(self._saver, self._sess)
t0 = time.time()
# when generate=True
else:
counter = 0
while True:
batch = self._batcher.next_batch(
) # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info(
"Decoder has finished reading dataset for single_pass."
)
return
original_article = batch.original_articles[0] # string
# original_abstract = batch.original_abstracts[0] # string
# original_abstract_sents = batch.original_abstracts_sents[0] # list of strings
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
# abstract_withunks = data.show_abs_oovs(original_abstract, self._vocab, (batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
# Run beam search to get best Hypothesis
best_hyp = beam_search.run_beam_search(self._sess, self._model,
self._vocab, batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(
data.STOP_DECODING
) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
counter += 1
# log output to screen
print(
"---------------------------------------------------------------------------"
)
tf.logging.info('ARTICLE: %s', article_withunks)
tf.logging.info('GENERATED SUMMARY: %s', decoded_output)
print(
"---------------------------------------------------------------------------"
)
# self.write_for_rouge(original_abstract_sents, decoded_words, counter)
# Write to file
decoded_sents = []
while len(decoded_words) > 0:
try:
fst_period_idx = decoded_words.index(".")
except ValueError: # there is text remaining that doesn't end in "."
fst_period_idx = len(decoded_words)
sent = decoded_words[:fst_period_idx +
1] # sentence up to and including the period
decoded_words = decoded_words[fst_period_idx +
1:] # everything else
decoded_sents.append(' '.join(sent))
# pyrouge calls a perl script that puts the data into HTML files.
# Therefore we need to make our output HTML safe.
decoded_sents = [make_html_safe(w) for w in decoded_sents]
# Write to file
result_file = os.path.join(self._result_dir,
"%06d_summary.txt" % counter)
with open(result_file, "w") as f:
for idx, sent in enumerate(decoded_sents):
f.write(sent) if idx == len(
decoded_sents) - 1 else f.write(sent + "\n")
def main(config, resume, phase):
# Dataset
fine_dataset = fine_clustering_dataset(config)
# Dataloder
train_loader = DataLoader(fine_dataset,
shuffle=True,
batch_size=config['batch_size'],
num_workers=32)
val_loader = DataLoader(fine_dataset,
shuffle=False,
batch_size=config['batch_size'],
num_workers=32)
test_loader = DataLoader(fine_dataset,
shuffle=False,
batch_size=config['batch_size'],
num_workers=32)
# Model
start_epoch = 0
if config['model_name'].startswith('resnet'):
model = ResNet(config)
elif config['model_name'].startswith('densenet'):
model = DenseNet(config)
elif config['model_name'].startswith('deeplab'):
cluster_vector_dim = config['cluster_vector_dim']
model = DeepLabv3_plus(nInputChannels=3,
n_classes=3,
os=16,
cluster_vector_dim=cluster_vector_dim,
pretrained=True,
_print=True)
elif config['model_name'].startswith('bagnet'):
model = BagNet(config=config)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if resume:
filepath = config['pretrain_path']
start_epoch, learning_rate, optimizer, M, s = load_ckpt(
model, filepath)
start_epoch += 1
if torch.cuda.device_count() > 1:
print("Let's use", torch.cuda.device_count(), "GPUs!")
model = nn.DataParallel(model)
model.to(device)
#Optimizer
optimizer = torch.optim.Adam(model.parameters(),
lr=config['learning_rate'],
weight_decay=1e-5)
#resume or not
if start_epoch == 0:
print("Grand New Training")
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
optimizer, patience=config['switch_learning_rate_interval'])
# log_dir = config['log_dir']+"/{}_{}_".format(config['date'],config['model_name'])+"ep_{}-{}_lr_{}".format(start_epoch,start_epoch+config['num_epoch'],config['learning_rate'])
# best loss
if not resume:
learning_rate = config['learning_rate']
M, s = cluster_initialization(train_loader, model, config, phase)
print(start_epoch)
if config['if_train']:
for epoch in range(start_epoch + 1,
start_epoch + config['num_epoch'] + 1):
loss_tr = train(
train_loader, model, optimizer, epoch, config, M,
s) #if training, delete learning rate and add optimizer
if config['if_valid'] and epoch % config[
'valid_epoch_interval'] == 0:
with torch.no_grad():
loss_val, M, s = valid(val_loader, model, epoch, config,
learning_rate, M, s, phase)
scheduler.step(loss_val)
save_ckpt(model, optimizer, epoch, loss_tr, loss_val, config,
M, s)
else:
val_log = open("../log/val_" + config['date'] + ".txt", "a")
val_log.write('epoch ' + str(epoch) + '\n')
val_log.close()
test(test_loader, model, config, M, phase)
store_config(config, phase)
print("Training finished ...")
def run_eval(self):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
self.model.build_graph() # build the graph
saver = tf.train.Saver(max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
if FLAGS.embedding:
sess.run(tf.global_variables_initializer(),feed_dict={self.model.embedding_place:self.word_vector})
eval_dir = os.path.join(FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(eval_dir, 'bestmodel') # this is where checkpoints of best models are saved
self.summary_writer = tf.summary.FileWriter(eval_dir)
if FLAGS.ac_training:
tf.logging.info('DDQN building graph')
t1 = time.time()
dqn_graph = tf.Graph()
with dqn_graph.as_default():
self.dqn.build_graph() # build dqn graph
tf.logging.info('building current network took {} seconds'.format(time.time()-t1))
self.dqn_target.build_graph() # build dqn target graph
tf.logging.info('building target network took {} seconds'.format(time.time()-t1))
dqn_saver = tf.train.Saver(max_to_keep=3) # keep 3 checkpoints at a time
dqn_sess = tf.Session(config=util.get_config())
dqn_train_step = 0
replay_buffer = ReplayBuffer(self.dqn_hps)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = self.restore_best_eval_model() # will hold the best loss achieved so far
train_step = 0
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
if FLAGS.ac_training:
_ = util.load_dqn_ckpt(dqn_saver, dqn_sess) # load a new checkpoint
processed_batch = 0
avg_losses = []
# evaluate for 100 * batch_size before comparing the loss
# we do this due to memory constraint, best to run eval on different machines with large batch size
while processed_batch < 100*FLAGS.batch_size:
processed_batch += FLAGS.batch_size
batch = self.batcher.next_batch() # get the next batch
if FLAGS.ac_training:
t0 = time.time()
transitions = self.model.collect_dqn_transitions(sess, batch, train_step, batch.max_art_oovs) # len(batch_size * k * max_dec_steps)
tf.logging.info('Q values collection time: {}'.format(time.time()-t0))
with dqn_graph.as_default():
# if using true Q-value to train DQN network,
# we do this as the pre-training for the DQN network to get better estimates
batch_len = len(transitions)
b = ReplayBuffer.create_batch(self.dqn_hps, transitions,len(transitions), use_state_prime = True, max_art_oovs = batch.max_art_oovs)
b_prime = ReplayBuffer.create_batch(self.dqn_hps, transitions,len(transitions), use_state_prime = True, max_art_oovs = batch.max_art_oovs)
dqn_results = self.dqn.run_test_steps(sess=dqn_sess, x= b._x, return_best_action=True)
q_estimates = dqn_results['estimates'] # shape (len(transitions), vocab_size)
dqn_best_action = dqn_results['best_action']
tf.logging.info('running test step on dqn_target')
dqn_target_results = self.dqn_target.run_test_steps(dqn_sess, x= b_prime._x)
q_vals_new_t = dqn_target_results['estimates'] # shape (len(transitions), vocab_size)
# we need to expand the q_estimates to match the input batch max_art_oov
q_estimates = np.concatenate([q_estimates,np.zeros((len(transitions),batch.max_art_oovs))],axis=-1)
tf.logging.info('fixing the action q-estimates')
for i, tr in enumerate(transitions):
if tr.done:
q_estimates[i][tr.action] = tr.reward
else:
q_estimates[i][tr.action] = tr.reward + FLAGS.gamma * q_vals_new_t[i][dqn_best_action[i]]
if FLAGS.dqn_scheduled_sampling:
tf.logging.info('scheduled sampling on q-estimates')
q_estimates = self.scheduled_sampling(batch_len, FLAGS.sampling_probability, b._y_extended, q_estimates)
if not FLAGS.calculate_true_q:
# when we are not training DQN based on true Q-values
# we need to update Q-values in our transitions based on this q_estimates we collected from DQN current network.
for trans, q_val in zip(transitions,q_estimates):
trans.q_values = q_val # each have the size vocab_extended
q_estimates = np.reshape(q_estimates, [FLAGS.batch_size, FLAGS.k, FLAGS.max_dec_steps, -1]) # shape (batch_size, k, max_dec_steps, vocab_size_extended)
tf.logging.info('run eval step on seq2seq model.')
t0=time.time()
results = self.model.run_eval_step(sess, batch, train_step, q_estimates)
t1=time.time()
else:
tf.logging.info('run eval step on seq2seq model.')
t0=time.time()
results = self.model.run_eval_step(sess, batch, train_step)
t1=time.time()
tf.logging.info('experiment: {}'.format(FLAGS.exp_name))
tf.logging.info('processed_batch: {}, seconds for batch: {}'.format(processed_batch, t1-t0))
printer_helper = {}
loss = printer_helper['pgen_loss']= results['pgen_loss']
if FLAGS.coverage:
printer_helper['coverage_loss'] = results['coverage_loss']
if FLAGS.rl_training or FLAGS.ac_training:
printer_helper['rl_cov_total_loss']= results['reinforce_cov_total_loss']
loss = printer_helper['pointer_cov_total_loss'] = results['pointer_cov_total_loss']
if FLAGS.rl_training or FLAGS.ac_training:
printer_helper['shared_loss'] = results['shared_loss']
printer_helper['rl_loss'] = results['rl_loss']
printer_helper['rl_avg_logprobs'] = results['rl_avg_logprobs']
if FLAGS.rl_training:
printer_helper['sampled_r'] = np.mean(results['sampled_sentence_r_values'])
printer_helper['greedy_r'] = np.mean(results['greedy_sentence_r_values'])
printer_helper['r_diff'] = printer_helper['greedy_r'] - printer_helper['sampled_r']
if FLAGS.ac_training:
printer_helper['dqn_loss'] = np.mean(self.avg_dqn_loss) if len(self.avg_dqn_loss) > 0 else 0
for (k,v) in printer_helper.items():
if not np.isfinite(v):
raise Exception("{} is not finite. Stopping.".format(k))
tf.logging.info('{}: {}\t'.format(k,v))
# add summaries
summaries = results['summaries']
train_step = results['global_step']
self.summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
avg_losses.append(self.calc_running_avg_loss(np.asscalar(loss), running_avg_loss, train_step))
tf.logging.info('-------------------------------------------')
running_avg_loss = np.mean(avg_losses)
tf.logging.info('==========================================')
tf.logging.info('best_loss: {}\trunning_avg_loss: {}\t'.format(best_loss, running_avg_loss))
tf.logging.info('==========================================')
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
tf.logging.info('Found new best model with %.3f running_avg_loss. Saving to %s', running_avg_loss, bestmodel_save_path)
saver.save(sess, bestmodel_save_path, global_step=train_step, latest_filename='checkpoint_best')
best_loss = running_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
self.summary_writer.flush()
def main(unused_argv):
if len(unused_argv
) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(
tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if FLAGS.mode == "train":
os.makedirs(FLAGS.log_root)
else:
raise Exception(
"Logdir %s doesn't exist. Run in train mode to create it." %
(FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = [
'mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag',
'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim',
'batch_size', 'max_dec_steps', 'max_enc_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size', 'max_dec_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
tf.set_random_seed(111) # a seed value for randomness
if hps_generator.mode == 'train':
print("Start pre-training......")
model_class = Classification(hps_discriminator, vocab)
cla_batcher = ClaBatcher(hps_discriminator, vocab)
sess_cls, saver_cls, train_dir_cls = setup_training_classification(
model_class)
print("Start pre-training classification......")
#run_pre_train_classification(model_class, cla_batcher, 10, sess_cls, saver_cls, train_dir_cls)
#generated = Generate_training_sample(model_class, vocab, cla_batcher, sess_cls)
#print("Generating training examples......")
#generated.generate_training_example("train")
#generated.generator_validation_example("valid")
model_sentiment = Sentimentor(hps_generator, vocab)
sentiment_batcher = SenBatcher(hps_generator, vocab)
sess_sen, saver_sen, train_dir_sen = setup_training_sentimentor(
model_sentiment)
#run_pre_train_sentimentor(model_sentiment,sentiment_batcher,1,sess_sen,saver_sen,train_dir_sen)
sentiment_generated = Generate_non_sentiment_weight(
model_sentiment, vocab, sentiment_batcher, sess_sen)
#sentiment_generated.generate_training_example("train_sentiment")
#sentiment_generated.generator_validation_example("valid_sentiment")
model = Generator(hps_generator, vocab)
# Create a batcher object that will create minibatches of data
batcher = GenBatcher(vocab, hps_generator)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
util.load_ckpt(saver_sen, sess_sen, ckpt_dir="train-sentimentor")
util.load_ckpt(saver_cls, sess_cls, ckpt_dir="train-classification")
generated = Generated_sample(model, vocab, batcher, sess_ge)
#print("Start pre-training generator......")
run_pre_train_generator(
model, batcher, 4, sess_ge, saver_ge, train_dir_ge, generated,
model_class, sess_cls,
cla_batcher) # this is an infinite loop until interrupted
#generated.generator_validation_negetive_example("temp_negetive", batcher, model_class,sess_cls,cla_batcher) # batcher, model_class, sess_cls, cla_batcher
#generated.generator_validation_positive_example(
# "temp_positive", batcher, model_class,sess_cls,cla_batcher)
loss_window = 0
t0 = time.time()
print("begin dual learning:")
for epoch in range(30):
batches = batcher.get_batches(mode='train')
for i in range(len(batches)):
current_batch = copy.deepcopy(batches[i])
sentiment_batch = batch_sentiment_batch(
current_batch, sentiment_batcher)
result = model_sentiment.max_generator(sess_sen,
sentiment_batch)
weight = result['generated']
current_batch.weight = weight
sentiment_batch.weight = weight
cla_batch = batch_classification_batch(current_batch, batcher,
cla_batcher)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
cc = SmoothingFunction()
reward_sentiment = 1 - np.abs(0.5 - result['y_pred_auc'])
reward_BLEU = []
for k in range(FLAGS.batch_size):
reward_BLEU.append(
sentence_bleu(
[current_batch.original_reviews[k].split()],
cla_batch.original_reviews[k].split(),
smoothing_function=cc.method1))
reward_BLEU = np.array(reward_BLEU)
reward_de = (2 / (1.0 / (1e-6 + reward_sentiment) + 1.0 /
(1e-6 + reward_BLEU)))
result = model.run_train_step(sess_ge, current_batch)
train_step = result[
'global_step'] # we need this to update our running average loss
loss = result['loss']
loss_window += loss
if train_step % 100 == 0:
t1 = time.time()
tf.logging.info(
'seconds for %d training generator step: %.3f ',
train_step, (t1 - t0) / 100)
t0 = time.time()
tf.logging.info('loss: %f', loss_window /
100) # print the loss to screen
loss_window = 0.0
if train_step % 10000 == 0:
#bleu_score = generatored.compute_BLEU(str(train_step))
#tf.logging.info('bleu: %f', bleu_score) # print the loss to screen
generated.generator_validation_negetive_example(
"valid-generated-transfer/" + str(epoch) +
"epoch_step" + str(train_step) + "_temp_positive",
batcher, model_class, sess_cls, cla_batcher)
generated.generator_validation_positive_example(
"valid-generated/" + str(epoch) + "epoch_step" +
str(train_step) + "_temp_positive", batcher,
model_class, sess_cls, cla_batcher)
#saver_ge.save(sess, train_dir + "/model", global_step=train_step)
cla_batch, bleu = output_to_classification_batch(
result['generated'], current_batch, batcher, cla_batcher,
cc)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
reward_result_sentiment = result['y_pred_auc']
reward_result_bleu = np.array(bleu)
reward_result = (2 / (1.0 /
(1e-6 + reward_result_sentiment) + 1.0 /
(1e-6 + reward_result_bleu)))
current_batch.score = 1 - current_batch.score
result = model.max_generator(sess_ge, current_batch)
cla_batch, bleu = output_to_classification_batch(
result['generated'], current_batch, batcher, cla_batcher,
cc)
result = model_class.run_ypred_auc(sess_cls, cla_batch)
reward_result_transfer_sentiment = result['y_pred_auc']
reward_result_transfer_bleu = np.array(bleu)
reward_result_transfer = (
2 / (1.0 /
(1e-6 + reward_result_transfer_sentiment) + 1.0 /
(1e-6 + reward_result_transfer_bleu)))
#tf.logging.info("reward_nonsentiment: "+str(reward_sentiment) +" output_original_sentiment: "+str(reward_result_sentiment)+" output_original_bleu: "+str(reward_result_bleu))
reward = reward_result_transfer #reward_de + reward_result_sentiment +
#tf.logging.info("reward_de: "+str(reward_de))
model_sentiment.run_train_step(sess_sen, sentiment_batch,
reward)
elif hps_generator.mode == 'decode':
decode_model_hps = hps_generator # This will be the hyperparameters for the decoder model
#model = Generator(decode_model_hps, vocab)
#generated = Generated_sample(model, vocab, batcher)
#bleu_score = generated.compute_BLEU()
#tf.logging.info('bleu: %f', bleu_score) # print the loss to screen
else:
raise ValueError("The 'mode' flag must be one of train/eval/decode")
def run_eval(model, batcher):
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(
FLAGS.exp_name, "val") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
batches = batcher.getBatches()
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
epoch_avg_loss = 0.
epoch_train_steps = 0
for batch in batches:
# run eval on the batch
t0 = time.time()
results = model.run_eval_step(sess, batch)
t1 = time.time()
#tf.logging.info('seconds for batch: %.2f', t1-t0)
# print the loss and coverage loss to screen
loss = results['loss']
epoch_train_steps += 1
#tf.logging.info('loss: %f', loss)
# add summodemaries
summaries = results['summaries']
train_step = results['global_step']
summary_writer.add_summary(summaries, train_step)
epoch_avg_loss = (epoch_avg_loss * (epoch_train_steps - 1.) +
loss) / epoch_train_steps
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
print("Average loss for Epoch %f" % (epoch_avg_loss))
if best_loss is None or epoch_avg_loss < best_loss:
#tf.logging.info('Found new best model with %.3f running_avg_loss. Saving to %s', running_avg_loss, bestmodel_save_path)
print('Found new best model with %f epoch_avg_loss. Saving to %s' %
(epoch_avg_loss, bestmodel_save_path))
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = epoch_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
if (FLAGS.dataset == 'timeseries'):
batcher.getData(updateData=True)
batches = batcher.getBatches()
def run_eval(self):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
self.model.build_graph() # build the graph
saver = tf.train.Saver(max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
if FLAGS.embedding:
sess.run(tf.global_variables_initializer(),feed_dict={self.model.embedding_place:self.word_vector})
#eval_dir = os.path.join(FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
eval_dir = os.path.join(FLAGS.log_root, "eval_{}".format(
"rouge" if FLAGS.rouge_based_eval else "loss")) # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(eval_dir, 'bestmodel') # this is where checkpoints of best models are saved
self.summary_writer = tf.summary.FileWriter(eval_dir)
if FLAGS.ac_training:
tf.logging.info('DDQN building graph')
t1 = time.time()
dqn_graph = tf.Graph()
with dqn_graph.as_default():
self.dqn.build_graph() # build dqn graph
tf.logging.info('building current network took {} seconds'.format(time.time()-t1))
self.dqn_target.build_graph() # build dqn target graph
tf.logging.info('building target network took {} seconds'.format(time.time()-t1))
dqn_saver = tf.train.Saver(max_to_keep=3) # keep 3 checkpoints at a time
dqn_sess = tf.Session(config=util.get_config())
dqn_train_step = 0
replay_buffer = ReplayBuffer(self.dqn_hps)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = self.restore_best_eval_model() # will hold the best loss achieved so far
train_step = 0
decay = 0.99
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
if FLAGS.ac_training:
_ = util.load_dqn_ckpt(dqn_saver, dqn_sess) # load a new checkpoint
processed_batch = 0
avg_losses = []
greedy_rouges = []
sampled_rouges = []
# evaluate for 100 * batch_size before comparing the loss
# we do this due to memory constraint, best to run eval on different machines with large batch size
while processed_batch < FLAGS.eval_interval*FLAGS.batch_size:
processed_batch += FLAGS.batch_size
batch = full_batch = self.full_batcher.next_batch()
if FLAGS.rl_training:
partial_batch = self.partial_batcher.next_batch()
batch = Batcher.merge_batches(full_batch, partial_batch)
if batch.is_any_null():
print(partial_batch.original_abstracts_sents)
print(full_batch.original_abstracts_sents)
import ipdb
ipdb.set_trace()
print(np.concatenate((full_batch.original_abstracts_sents, partial_batch.original_abstracts_sents), axis=0))
raise Exception
else:
partial_batch = None
if FLAGS.ac_training:
t0 = time.time()
transitions = self.model.collect_dqn_transitions(sess, batch, train_step, batch.max_art_oovs) # len(batch_size * k * max_dec_steps)
tf.logging.info('Q values collection time: {}'.format(time.time()-t0))
with dqn_graph.as_default():
# if using true Q-value to train DQN network,
# we do this as the pre-training for the DQN network to get better estimates
batch_len = len(transitions)
b = ReplayBuffer.create_batch(self.dqn_hps, transitions,len(transitions), use_state_prime = True, max_art_oovs = batch.max_art_oovs)
b_prime = ReplayBuffer.create_batch(self.dqn_hps, transitions,len(transitions), use_state_prime = True, max_art_oovs = batch.max_art_oovs)
dqn_results = self.dqn.run_test_steps(sess=dqn_sess, x= b._x, return_best_action=True)
q_estimates = dqn_results['estimates'] # shape (len(transitions), vocab_size)
dqn_best_action = dqn_results['best_action']
tf.logging.info('running test step on dqn_target')
dqn_target_results = self.dqn_target.run_test_steps(dqn_sess, x= b_prime._x)
q_vals_new_t = dqn_target_results['estimates'] # shape (len(transitions), vocab_size)
# we need to expand the q_estimates to match the input batch max_art_oov
q_estimates = np.concatenate([q_estimates,np.zeros((len(transitions),batch.max_art_oovs))],axis=-1)
tf.logging.info('fixing the action q-estimates')
for i, tr in enumerate(transitions):
if tr.done:
q_estimates[i][tr.action] = tr.reward
else:
q_estimates[i][tr.action] = tr.reward + FLAGS.gamma * q_vals_new_t[i][dqn_best_action[i]]
if FLAGS.dqn_scheduled_sampling:
tf.logging.info('scheduled sampling on q-estimates')
q_estimates = self.scheduled_sampling(batch_len, FLAGS.sampling_probability, b._y_extended, q_estimates)
if not FLAGS.calculate_true_q:
# when we are not training DQN based on true Q-values
# we need to update Q-values in our transitions based on this q_estimates we collected from DQN current network.
for trans, q_val in zip(transitions,q_estimates):
trans.q_values = q_val # each have the size vocab_extended
q_estimates = np.reshape(q_estimates, [FLAGS.batch_size, FLAGS.k, FLAGS.max_dec_steps, -1]) # shape (batch_size, k, max_dec_steps, vocab_size_extended)
tf.logging.info('run eval step on seq2seq model.')
t0=time.time()
results = self.model.run_eval_steps(sess, batch, train_step, q_estimates)
t1=time.time()
else:
tf.logging.info('run eval step on seq2seq model.')
t0=time.time()
results = self.model.run_eval_steps(sess, batch, train_step)
t1=time.time()
tf.logging.info('experiment: {}'.format(FLAGS.exp_name))
tf.logging.info('processed_batch: {}, seconds for batch: {}'.format(processed_batch, t1-t0))
printer_helper = {}
loss = printer_helper['pgen_loss']= results['pgen_loss']
printer_helper['rl_full_reward_sampled'] = np.mean(results['full_ssr'])
printer_helper['rl_full_reward_greedy'] = np.mean(results['full_gsr'])
printer_helper['rl_full_reward_diff'] = results['full_reward_diff']
if FLAGS.rl_training:
loss = np.mean([results['full_rl_avg_logprobs'], results['partial_rl_avg_logprobs']])
printer_helper['shared_loss'] = results['shared_loss']
printer_helper['rl_full_loss'] = results['full_rl_loss']
printer_helper['rl_full_avg_logprobs'] = results['full_rl_avg_logprobs']
printer_helper['rl_partial_loss'] = results['partial_rl_loss']
printer_helper['rl_partial_avg_logprobs'] = results['partial_rl_avg_logprobs']
printer_helper['rl_partial_reward_sampled'] = results['partial_ssr']
printer_helper['rl_partial_reward_greedy'] = results['partial_gsr']
printer_helper['rl_partial_reward_diff'] = results['partial_reward_diff']
if FLAGS.coverage:
loss = printer_helper['coverage_loss'] = results['coverage_loss']
if FLAGS.rl_training or FLAGS.ac_training:
loss = printer_helper['rl_cov_total_loss'] = results['reinforce_cov_total_loss']
elif FLAGS.pointer_gen:
loss = printer_helper['pointer_cov_total_loss'] = results['pointer_cov_total_loss']
if FLAGS.rl_training:
greedy_rouges.append(np.mean([results['full_gsr'],results['partial_gsr']]))
sampled_rouges.append(np.mean([results['full_ssr'],results['partial_ssr']]))
else:
greedy_rouges.append(np.mean(results['full_gsr']))
sampled_rouges.append(np.mean(results['full_ssr']))
for (k,v) in sorted(printer_helper.items(), key=lambda x: x[0]):
if not np.isfinite(v):
raise Exception("{} is not finite. Stopping.".format(k))
tf.logging.info('{}: {}\t'.format(k,v))
tf.logging.info('-------------------------------------------')
time.sleep(2)
# add summaries
summaries = results['summaries']
train_step = results['global_step']
self.summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
avg_losses.append(self.calc_running_avg_loss(np.asscalar(loss), running_avg_loss))
tf.logging.info('-------------------------------------------')
time.sleep(2)
running_avg_loss = np.mean(avg_losses)
running_greedy_rouge = np.mean(greedy_rouges)
running_sampled_rouge = np.mean(sampled_rouges)
self.summary_writer.add_summary(
tf.Summary(value=[tf.Summary.Value(tag="running_greedy_rouge", simple_value=running_greedy_rouge), ]),
train_step)
self.summary_writer.add_summary(
tf.Summary(value=[tf.Summary.Value(tag="running_sampled_rouge", simple_value=running_sampled_rouge), ]),
train_step)
self.summary_writer.add_summary(tf.Summary(
value=[tf.Summary.Value(tag='running_avg_loss/decay=%f' % (decay), simple_value=running_avg_loss), ]),
train_step)
tf.logging.info('==========================================')
tf.logging.info('best_loss: {}\trunning_avg_loss: {}\t'.format(best_loss, running_avg_loss))
tf.logging.info('greedy rouges: {}\tsampled rouges: {}\t'.format(running_greedy_rouge, running_sampled_rouge))
tf.logging.info('==========================================')
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if (best_loss is None) or (FLAGS.rouge_based_eval and running_greedy_rouge > best_loss) or (
not FLAGS.rouge_based_eval and running_avg_loss < best_loss):
tf.logging.info('Found new best model with %.3f %s. Saving to %s',
running_greedy_rouge if FLAGS.rouge_based_eval else running_avg_loss,
"running_greedy_rouge" if FLAGS.rouge_based_eval else "running_avg_loss",
bestmodel_save_path)
saver.save(sess, bestmodel_save_path, global_step=train_step, latest_filename='checkpoint_best')
best_loss = running_greedy_rouge if FLAGS.rouge_based_eval else running_avg_loss
time.sleep(15)
# flush the summary writer every so often
if train_step % 100 == 0:
self.summary_writer.flush()
def __init__(self, model, batcher, vocab):
"""Initialize decoder.
Args:
model: a SentSelector object.
batcher: a Batcher object.
vocab: Vocabulary object
"""
# get the data split set
if "train" in FLAGS.data_path: self._dataset = "train"
elif "val" in FLAGS.data_path: self._dataset = "val"
elif "test" in FLAGS.data_path: self._dataset = "test"
else:
raise ValueError(
"FLAGS.data_path %s should contain one of train, val or test" %
(FLAGS.data_path))
# create the data loader
self._batcher = batcher
if FLAGS.eval_gt_rouge: # no need to load model, default is fasle
# Make a descriptive decode directory name
self._decode_dir = os.path.join(FLAGS.log_root,
'select_gt' + self._dataset)
tf.logging.info('Save evaluation results to ' + self._decode_dir)
if os.path.exists(self._decode_dir):
raise Exception(
"single_pass decode directory %s should not already exist"
% self._decode_dir)
# Make the decode dir
os.makedirs(self._decode_dir)
# Make the dirs to contain output written in the correct format for pyrouge
self._rouge_ref_dir = os.path.join(self._decode_dir, "reference")
if not os.path.exists(self._rouge_ref_dir):
os.mkdir(self._rouge_ref_dir)
self._rouge_gt_dir = os.path.join(self._decode_dir, "gt_selected")
if not os.path.exists(self._rouge_gt_dir):
os.mkdir(self._rouge_gt_dir)
else: # FALSE
self._model = model
self._model.build_graph()
self._vocab = vocab
self._saver = tf.train.Saver(
) # we use this to load checkpoints for decoding
self._sess = tf.Session(config=util.get_config())
# Load an initial checkpoint to use for decoding
print(" eval_ckpt_path ", FLAGS.eval_ckpt_path)
if FLAGS.load_best_eval_model:
tf.logging.info('Loading best eval checkpoint')
ckpt_path = util.load_ckpt(self._saver,
self._sess,
ckpt_dir='eval')
elif FLAGS.eval_ckpt_path:
ckpt_path = util.load_ckpt(self._saver,
self._sess,
ckpt_path=FLAGS.eval_ckpt_path,
ckpt_dir='eval')
else:
tf.logging.info('Loading best train checkpoint')
ckpt_path = util.load_ckpt(self._saver, self._sess)
if FLAGS.single_pass:
# Make a descriptive decode directory name
ckpt_name = "ckpt-" + ckpt_path.split('-')[
-1] # this is something of the form "ckpt-123456"
decode_root_dir, decode_dir = get_decode_dir_name(
ckpt_name, self._dataset)
self._decode_root_dir = os.path.join(FLAGS.log_root,
decode_root_dir)
self._decode_dir = os.path.join(FLAGS.log_root,
decode_root_dir, decode_dir)
tf.logging.info('Save evaluation results to ' +
self._decode_dir)
if os.path.exists(self._decode_dir):
raise Exception(
"single_pass decode directory %s should not already exist"
% self._decode_dir)
else: # Generic decode dir name
self._decode_dir = os.path.join(FLAGS.log_root, "select")
# Make the decode dir if necessary
if not os.path.exists(self._decode_dir):
os.makedirs(self._decode_dir)
if FLAGS.single_pass:
# Make the dirs to contain output written in the correct format for pyrouge
self._rouge_ref_dir = os.path.join(self._decode_dir,
"reference")
if not os.path.exists(self._rouge_ref_dir):
os.mkdir(self._rouge_ref_dir)
self._rouge_dec_dir = os.path.join(self._decode_dir,
"selected")
if not os.path.exists(self._rouge_dec_dir):
os.mkdir(self._rouge_dec_dir)
if FLAGS.save_pkl:
self._result_dir = os.path.join(self._decode_dir,
"select_result")
if not os.path.exists(self._result_dir):
os.mkdir(self._result_dir)
self._probs_pkl_path = os.path.join(self._decode_root_dir,
"probs.pkl")
if not os.path.exists(self._probs_pkl_path):
self._make_probs_pkl = True
else:
self._make_probs_pkl = False
self._precision = []
self._recall = []
self._accuracy = []
self._ratio = []
self._select_sent_num = []
def main(unused_argv):
if len(unused_argv) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if FLAGS.mode=="train":
os.makedirs(FLAGS.log_root)
else:
raise Exception("Logdir %s doesn't exist. Run in train mode to create it." % (FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = ['mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size', 'max_dec_sen_num','max_dec_steps', 'max_enc_steps']
hps_dict = {}
for key,val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = ['lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std', 'max_grad_norm',
'hidden_dim', 'emb_dim', 'batch_size', 'max_enc_sen_num', 'max_enc_seq_len']
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
# Create a batcher object that will create minibatches of data
batcher = GenBatcher(vocab, hps_generator)
tf.set_random_seed(111) # a seed value for randomness
if hps_generator.mode == 'train':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
print("Start pre-training generator......")
run_pre_train_generator(model, batcher, 10, sess_ge, saver_ge, train_dir_ge,generated) # this is an infinite loop until
print("Generating negetive examples......")
generated.generator_whole_negetive_example()
generated.generator_test_negetive_example()
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab, "train/generated_samples_positive/*", "train/generated_samples_negetive/*", "test/generated_samples_positive/*", "test/generated_samples_negetive/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(model_dis)
print("Start pre-training discriminator......")
#run_test_discriminator(model_dis, dis_batcher, sess_dis, saver_dis, "test")
run_pre_train_discriminator(model_dis, dis_batcher, 25, sess_dis, saver_dis, train_dir_dis)
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
generated.generator_sample_example("sample_temp_positive", "sample_temp_negetive", 1000)
generated.generator_test_sample_example("test_sample_temp_positive",
"test_sample_temp_negetive",
200)
generated.generator_test_max_example("test_max_temp_positive",
"test_max_temp_negetive",
200)
tf.logging.info("true data diversity: ")
eva = Evaluate()
eva.diversity_evaluate("test_sample_temp_positive" + "/*")
print("Start adversial training......")
whole_decay = False
for epoch in range(1):
batches = batcher.get_batches(mode='train')
for step in range(int(len(batches)/1000)):
run_train_generator(model,model_dis,sess_dis,batcher,dis_batcher,batches[step*1000:(step+1)*1000],sess_ge, saver_ge, train_dir_ge,generated) #(model, discirminator_model, discriminator_sess, batcher, dis_batcher, batches, sess, saver, train_dir, generated):
generated.generator_sample_example("sample_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_positive", "sample_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_negetive", 1000)
#generated.generator_max_example("max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_positive", "max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_negetive", 200)
tf.logging.info("test performance: ")
tf.logging.info("epoch: "+str(epoch)+" step: "+str(step))
generated.generator_test_sample_example(
"test_sample_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_positive",
"test_sample_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negetive", 200)
generated.generator_test_max_example("test_max_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_positive",
"test_max_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negetive",
200)
dis_batcher.train_queue = []
dis_batcher.train_queue = []
for i in range(epoch+1):
for j in range(step+1):
dis_batcher.train_queue += dis_batcher.fill_example_queue("sample_generated/"+str(i)+"epoch_step"+str(j)+"_temp_positive/*")
dis_batcher.train_queue += dis_batcher.fill_example_queue("sample_generated/"+str(i)+"epoch_step"+str(j)+"_temp_negetive/*")
dis_batcher.train_batch = dis_batcher.create_batches(mode="train", shuffleis=True)
#dis_batcher.valid_batch = dis_batcher.train_batch
whole_decay = run_train_discriminator(model_dis, 5, dis_batcher, dis_batcher.get_batches(mode="train"),
sess_dis, saver_dis, train_dir_dis, whole_decay)
'''elif hps_generator.mode == 'decode':
decode_model_hps = hps_generator # This will be the hyperparameters for the decoder model
model = Generator(decode_model_hps, vocab)
generated = Generated_sample(model, vocab, batcher)
bleu_score = generated.compute_BLEU()'=
tf.logging.info('bleu: %f', bleu_score) # print the loss to screen'''
else:
print('finetune')
lr = opt.lr_finetune
net.freeze_enc_bn = True
else:
lr = opt.lr
optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad,
net.parameters()),
lr=lr,
betas=(0.5, 0.999))
criterion = TotalLoss(VGG16FeatureExtractor()).to(device)
start_iter = 0
if opt.resume:
start_iter = load_ckpt(opt.resume, [('model', net)],
[('optimizer', optimizer)])
for param_group in optimizer.param_groups:
param_group['lr'] = lr
print('Starting from iter ', start_iter)
# --------
# Training
# --------
for epoch in range(start_iter, opt.n_epochs):
net.train()
for i, (image, mask, gt) in enumerate(iterator_train):
image, mask, gt = Variable(image).to(device), Variable(
mask, requires_grad=False).to(device), Variable(gt).to(device)
output, _ = net(image, mask)
def main(unused_argv):
# %%
# choose what level of logging you want
tf.logging.set_verbosity(tf.logging.INFO)
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# 創建字典
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size)
hparam_list = [
'mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag',
'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim',
'batch_size', 'max_dec_sen_num', 'max_dec_steps', 'max_enc_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items():
if key in hparam_list:
hps_dict[key] = val.value # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size',
'max_enc_sen_num', 'max_enc_seq_len'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list:
hps_dict[key] = val.value # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
# # 取出最小batch size 的資料量
batcher = GenBatcher(vocab, hps_generator)
# print(batcher.train_batch[0].original_review_inputs)
# print(len(batcher.train_batch[0].original_review_inputs))
tf.set_random_seed(123)
# %%
if FLAGS.mode == 'train_generator':
# print("Start pre-training ......")
ge_model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(ge_model)
generated = Generated_sample(ge_model, vocab, batcher, sess_ge)
print("Start pre-training generator......")
run_pre_train_generator(ge_model, batcher, 300, sess_ge, saver_ge,
train_dir_ge)
# util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
print("finish load train-generator")
print("Generating negative examples......")
generator_graph = tf.Graph()
with generator_graph.as_default():
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
print("finish load train-generator")
generated.generator_train_negative_example()
generated.generator_test_negative_example()
print("finish write")
elif FLAGS.mode == 'train_discriminator':
# print("Start pre-training ......")
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab,
"discriminator_train/positive/*",
"discriminator_train/negative/*",
"discriminator_test/positive/*",
"discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(
model_dis)
print("Start pre-training discriminator......")
if not os.path.exists("discriminator_result"):
os.mkdir("discriminator_result")
run_pre_train_discriminator(model_dis, dis_batcher, 1000, sess_dis,
saver_dis, train_dir_dis)
elif FLAGS.mode == "adversarial_train":
generator_graph = tf.Graph()
discriminatorr_graph = tf.Graph()
print("Start adversarial-training......")
# tf.reset_default_graph()
with generator_graph.as_default():
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
print("finish load train-generator")
with discriminatorr_graph.as_default():
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab,
"discriminator_train/positive/*",
"discriminator_train/negative/*",
"discriminator_test/positive/*",
"discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(
model_dis)
util.load_ckpt(saver_dis, sess_dis, ckpt_dir="train-discriminator")
print("finish load train-discriminator")
print("Start adversarial training......")
if not os.path.exists("train_sample_generated"):
os.mkdir("train_sample_generated")
if not os.path.exists("test_max_generated"):
os.mkdir("test_max_generated")
if not os.path.exists("test_sample_generated"):
os.mkdir("test_sample_generated")
whole_decay = False
for epoch in range(100):
print('開始訓練')
batches = batcher.get_batches(mode='train')
for step in range(int(len(batches) / 14)):
run_train_generator(model, model_dis, sess_dis, batcher,
dis_batcher,
batches[step * 14:(step + 1) * 14],
sess_ge, saver_ge, train_dir_ge)
generated.generator_sample_example(
"train_sample_generated/" + str(epoch) + "epoch_step" +
str(step) + "_temp_positive", "train_sample_generated/" +
str(epoch) + "epoch_step" + str(step) + "_temp_negative",
14)
tf.logging.info("test performance: ")
tf.logging.info("epoch: " + str(epoch) + " step: " + str(step))
# print("evaluate the diversity of DP-GAN (decode based on max probability)")
# generated.generator_test_sample_example(
# "test_sample_generated/" +
# str(epoch) + "epoch_step" + str(step) + "_temp_positive",
# "test_sample_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negative", 14)
#
# print("evaluate the diversity of DP-GAN (decode based on sampling)")
# generated.generator_test_max_example(
# "test_max_generated/" +
# str(epoch) + "epoch_step" + str(step) + "_temp_positive",
# "test_max_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negative", 14)
dis_batcher.train_queue = []
for i in range(epoch + 1):
for j in range(step + 1):
dis_batcher.train_queue += dis_batcher.fill_example_queue(
"train_sample_generated/" + str(i) + "epoch_step" +
str(j) + "_temp_positive/*")
dis_batcher.train_queue += dis_batcher.fill_example_queue(
"train_sample_generated/" + str(i) + "epoch_step" +
str(j) + "_temp_negative/*")
dis_batcher.train_batch = dis_batcher.create_batches(
mode="train", shuffleis=True)
whole_decay = run_train_discriminator(
model_dis, 5, dis_batcher,
dis_batcher.get_batches(mode="train"), sess_dis, saver_dis,
train_dir_dis, whole_decay)
elif FLAGS.mode == "test_language_model":
ge_model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(ge_model)
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
print("finish load train-generator")
# generator_graph = tf.Graph()
# with generator_graph.as_default():
# util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
# print("finish load train-generator")
# jieba.load_userdict('dir.txt')
inputs = ''
while inputs != "close":
inputs = input("Enter your ask: ")
sentence = segmentor.segment(t2s.convert(inputs))
# sentence = jieba.cut(inputs)
sentence = (" ".join(sentence))
sentence = s2t.convert(sentence)
print(sentence)
sentence = sentence.split()
enc_input = [vocab.word2id(w) for w in sentence]
enc_lens = np.array([len(enc_input)])
enc_input = np.array([enc_input])
out_sentence = ('[START]').split()
dec_batch = [vocab.word2id(w) for w in out_sentence]
#dec_batch = [2] + dec_batch
# dec_batch.append(3)
while len(dec_batch) < 40:
dec_batch.append(1)
dec_batch = np.array([dec_batch])
dec_batch = np.resize(dec_batch, (1, 1, 40))
dec_lens = np.array([len(dec_batch)])
if (FLAGS.beamsearch == 'beamsearch_train'):
result = ge_model.run_test_language_model(
sess_ge, enc_input, enc_lens, dec_batch, dec_lens)
# print(result['generated'])
# print(result['generated'].shape)
output_ids = result['generated'][0]
decoded_words = data.outputids2words(output_ids, vocab, None)
print("decoded_words :", decoded_words)
else:
results = ge_model.run_test_beamsearch_example(
sess_ge, enc_input, enc_lens, dec_batch, dec_lens)
beamsearch_outputs = results['beamsearch_outputs']
for i in range(5):
predict_list = np.ndarray.tolist(beamsearch_outputs[:, :,
i])
predict_list = predict_list[0]
predict_seq = [vocab.id2word(idx) for idx in predict_list]
decoded_words = " ".join(predict_seq).split()
# decoded_words = decoded_words
try:
if decoded_words[0] == '[STOPDOC]':
decoded_words = decoded_words[1:]
# index of the (first) [STOP] symbol
fst_stop_idx = decoded_words.index(data.STOP_DECODING)
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
if decoded_words[-1] != '.' and decoded_words[
-1] != '!' and decoded_words[-1] != '?':
decoded_words.append('.')
decoded_words_all = []
decoded_output = ' '.join(
decoded_words).strip() # single string
decoded_words_all.append(decoded_output)
decoded_words_all = ' '.join(decoded_words_all).strip()
decoded_words_all = decoded_words_all.replace("[UNK] ", "")
decoded_words_all = decoded_words_all.replace("[UNK]", "")
decoded_words_all = decoded_words_all.replace(" ", "")
decoded_words_all, _ = re.subn(r"(! ){2,}", "",
decoded_words_all)
decoded_words_all, _ = re.subn(r"(\. ){2,}", "",
decoded_words_all)
if decoded_words_all.startswith('，'):
decoded_words_all = decoded_words_all[1:]
print("The resonse : {}".format(decoded_words_all))
def main(unused_argv):
if len(unused_argv) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if "train" in FLAGS.mode:
os.makedirs(FLAGS.log_root)
else:
raise Exception("Logdir %s doesn't exist. Run in train mode to create it." % (FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = ['mode', 'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std', 'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size', 'max_dec_sen_num','max_dec_steps', 'max_enc_steps']
hps_dict = {}
for key,val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = ['lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std', 'max_grad_norm',
'hidden_dim', 'emb_dim', 'batch_size', 'max_enc_sen_num', 'max_enc_seq_len']
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_discriminator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
# Create a batcher object that will create minibatches of data
batcher = GenBatcher(vocab, hps_generator)
tf.set_random_seed(111) # a seed value for randomness
if hps_generator.mode == 'adversarial_train':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab, "discriminator_train/positive/*", "discriminator_train/negative/*", "discriminator_test/positive/*", "discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(model_dis)
util.load_ckpt(saver_dis, sess_dis, ckpt_dir="train-discriminator")
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
if not os.path.exists("MLE"): os.mkdir("MLE")
print("evaluate the diversity of MLE (decode based on sampling)")
generated.generator_test_sample_example("MLE/"+"MLE_sample_positive",
"MLE/"+"MLE_sample_negative",
200)
print("evaluate the diversity of MLE (decode based on max probability)")
generated.generator_test_max_example("MLE/"+"MLE_max_temp_positive",
"MLE/"+"MLE_max_temp_negative",
200)
print("Start adversarial training......")
if not os.path.exists("train_sample_generated"): os.mkdir("train_sample_generated")
if not os.path.exists("test_max_generated"): os.mkdir("test_max_generated")
if not os.path.exists("test_sample_generated"): os.mkdir("test_sample_generated")
whole_decay = False
for epoch in range(10):
batches = batcher.get_batches(mode='train')
for step in range(int(len(batches)/1000)):
run_train_generator(model,model_dis,sess_dis,batcher,dis_batcher,batches[step*1000:(step+1)*1000],sess_ge, saver_ge, train_dir_ge,generated) #(model, discirminator_model, discriminator_sess, batcher, dis_batcher, batches, sess, saver, train_dir, generated):
generated.generator_sample_example("train_sample_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_positive", "train_sample_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_negative", 1000)
#generated.generator_max_example("max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_positive", "max_generated/"+str(epoch)+"epoch_step"+str(step)+"_temp_negetive", 200)
tf.logging.info("test performance: ")
tf.logging.info("epoch: "+str(epoch)+" step: "+str(step))
print("evaluate the diversity of DP-GAN (decode based on max probability)")
generated.generator_test_sample_example(
"test_sample_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_positive",
"test_sample_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negative", 200)
print("evaluate the diversity of DP-GAN (decode based on sampling)")
generated.generator_test_max_example("test_max_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_positive",
"test_max_generated/" + str(epoch) + "epoch_step" + str(step) + "_temp_negative",
200)
dis_batcher.train_queue = []
dis_batcher.train_queue = []
for i in range(epoch+1):
for j in range(step+1):
dis_batcher.train_queue += dis_batcher.fill_example_queue("train_sample_generated/"+str(i)+"epoch_step"+str(j)+"_temp_positive/*")
dis_batcher.train_queue += dis_batcher.fill_example_queue("train_sample_generated/"+str(i)+"epoch_step"+str(j)+"_temp_negative/*")
dis_batcher.train_batch = dis_batcher.create_batches(mode="train", shuffleis=True)
#dis_batcher.valid_batch = dis_batcher.train_batch
whole_decay = run_train_discriminator(model_dis, 5, dis_batcher, dis_batcher.get_batches(mode="train"),
sess_dis, saver_dis, train_dir_dis, whole_decay)
elif hps_generator.mode == 'train_generator':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, batcher, sess_ge)
print("Start pre-training generator......")
run_pre_train_generator(model, batcher, 10, sess_ge, saver_ge, train_dir_ge,generated) # this is an infinite loop until
print("Generating negative examples......")
generated.generator_train_negative_example()
generated.generator_test_negative_example()
elif hps_generator.mode == 'train_discriminator':
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
#util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
model_dis = Discriminator(hps_discriminator, vocab)
dis_batcher = DisBatcher(hps_discriminator, vocab, "discriminator_train/positive/*", "discriminator_train/negative/*", "discriminator_test/positive/*", "discriminator_test/negative/*")
sess_dis, saver_dis, train_dir_dis = setup_training_discriminator(model_dis)
print("Start pre-training discriminator......")
#run_test_discriminator(model_dis, dis_batcher, sess_dis, saver_dis, "test")
if not os.path.exists("discriminator_result"): os.mkdir("discriminator_result")
run_pre_train_discriminator(model_dis, dis_batcher, 25, sess_dis, saver_dis, train_dir_dis)
def __init__(self,
model,
batcher,
vocab,
lm_model=None,
lm_word2idx=None,
lm_idx2word=None):
"""Initialize decoder.
Args:
model: a Seq2SeqAttentionModel object.
batcher: a Batcher object.
vocab: Vocabulary object
"""
self._model = model
self._model.build_graph()
self._batcher = batcher
self._vocab = vocab
self._saver = tf.train.Saver(
) # we use this to load checkpoints for decoding
self._sess = tf.Session(config=util.get_config())
# these keep running stats during decoding
self.total_nsentence = 0
self.total_length = 0
self.overlap_dict = dict()
# these are for external lm
self._lm_model = lm_model
self._lm_word2idx = lm_word2idx
self._lm_idx2word = lm_idx2word
# Load an initial checkpoint to use for decoding
ckpt_path = util.load_ckpt(self._saver, self._sess)
if FLAGS.single_pass:
# Make a descriptive decode directory name
if FLAGS.pointer_gen_only_vocab:
ckpt_name = "vocab-ckpt-" + ckpt_path.split('-')[
-1] # this is something of the form "ckpt-123456"
elif FLAGS.pointer_gen_only_attn:
ckpt_name = "attn-ckpt-" + ckpt_path.split('-')[-1]
else:
ckpt_name = "pre_cov_val_ckpt-" + ckpt_path.split('-')[-1]
self._decode_dir = os.path.join(FLAGS.log_root,
get_decode_dir_name(ckpt_name))
if os.path.exists(self._decode_dir):
raise Exception(
"single_pass decode directory %s should not already exist"
% self._decode_dir)
else: # Generic decode dir name
self._decode_dir = os.path.join(FLAGS.log_root, "decode")
# Make the decode dir if necessary
if not os.path.exists(self._decode_dir): os.mkdir(self._decode_dir)
if FLAGS.single_pass:
# Make the dirs to contain output written in the correct format for pyrouge
self._rouge_ref_dir = os.path.join(self._decode_dir, "reference")
if not os.path.exists(self._rouge_ref_dir):
os.mkdir(self._rouge_ref_dir)
self._rouge_dec_dir = os.path.join(self._decode_dir, "decoded")
if not os.path.exists(self._rouge_dec_dir):
os.mkdir(self._rouge_dec_dir)
def run_eval(model, batcher, vocab):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(
FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
extra_input = {}
if 'key_phrases' in model._extra_info:
# TODO: calculate the key phrase input part for this specific batch
raise NotImplementedError(
'Key phrases part has not been implemented here!')
# run eval on the batch
t0 = time.time()
results = model.run_eval_step(sess, batch, extra_input=extra_input)
t1 = time.time()
tf.logging.info('seconds for batch: %.2f', t1 - t0)
# print the loss and coverage loss to screen
loss = results['loss']
tf.logging.info('loss: %f', loss)
if FLAGS.coverage:
coverage_loss = results['coverage_loss']
tf.logging.info("coverage_loss: %f", coverage_loss)
# add summaries
summaries = results['summaries']
train_step = results['global_step']
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
running_avg_loss = calc_running_avg_loss(np.asscalar(loss),
running_avg_loss,
summary_writer, train_step)
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
tf.logging.info(
'Found new best model with %.3f running_avg_loss. Saving to %s',
running_avg_loss, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = running_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def main(unused_argv):
if len(unused_argv
) != 1: # prints a message if you've entered flags incorrectly
raise Exception("Problem with flags: %s" % unused_argv)
tf.logging.set_verbosity(
tf.logging.INFO) # choose what level of logging you want
tf.logging.info('Starting running in %s mode...', (FLAGS.mode))
# Change log_root to FLAGS.log_root/FLAGS.exp_name and create the dir if necessary
FLAGS.log_root = os.path.join(FLAGS.log_root, FLAGS.exp_name)
if not os.path.exists(FLAGS.log_root):
if FLAGS.mode == "train":
os.makedirs(FLAGS.log_root)
else:
raise Exception(
"Logdir %s doesn't exist. Run in train mode to create it." %
(FLAGS.log_root))
vocab = Vocab(FLAGS.vocab_path, FLAGS.vocab_size) # create a vocabulary
# Make a namedtuple hps, containing the values of the hyperparameters that the model needs
hparam_list = [
'vocab_size', 'dataset', 'mode', 'lr', 'adagrad_init_acc',
'rand_unif_init_mag', 'trunc_norm_init_std', 'max_grad_norm',
'hidden_dim', 'emb_dim', 'batch_size', 'max_dec_sen_num',
'max_enc_num', 'max_dec_steps', 'max_enc_steps'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size',
'srl_max_dec_seq_len', 'srl_max_dec_sen_num', 'srl_max_enc_seq_len',
'srl_max_enc_sen_num'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_srl_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
hparam_list = [
'lr', 'adagrad_init_acc', 'rand_unif_init_mag', 'trunc_norm_init_std',
'max_grad_norm', 'hidden_dim', 'emb_dim', 'batch_size',
'sc_max_dec_seq_len', 'sc_max_enc_seq_len'
]
hps_dict = {}
for key, val in FLAGS.__flags.items(): # for each flag
if key in hparam_list: # if it's in the list
hps_dict[key] = val # add it to the dict
hps_sc_generator = namedtuple("HParams", hps_dict.keys())(**hps_dict)
# Create a batcher object that will create minibatches of data
sc_batcher = Sc_GenBatcher(vocab, hps_sc_generator)
tf.set_random_seed(111) # a seed value for randomness
if hps_generator.mode == 'train':
print("Start pre-training......")
sc_model = Sc_Generator(hps_sc_generator, vocab)
sess_sc, saver_sc, train_dir_sc = setup_training_sc_generator(sc_model)
sc_generated = Generated_sc_sample(sc_model, vocab, sess_sc)
print("Start pre-training generator......")
run_pre_train_sc_generator(sc_model, sc_batcher, 40, sess_sc, saver_sc,
train_dir_sc, sc_generated)
if not os.path.exists("data/" + str(0) + "/"):
os.mkdir("data/" + str(0) + "/")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-train"),
"data/" + str(0) + "/train_skeleton.txt")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-valid"),
"data/" + str(0) + "/valid_skeleton.txt")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-test"),
"data/" + str(0) + "/test_skeleton.txt")
merge("data/story/train_process.txt", "data/0/train_skeleton.txt",
"data/0/train.txt")
merge("data/story/validation_process.txt", "data/0/valid_skeleton.txt",
"data/0/valid.txt")
merge("data/story/test_process.txt", "data/0/test_skeleton.txt",
"data/0/test.txt")
#################################################################################################
batcher = GenBatcher(vocab, hps_generator)
srl_batcher = Srl_GenBatcher(vocab, hps_srl_generator)
print("Start pre-training......")
model = Generator(hps_generator, vocab)
sess_ge, saver_ge, train_dir_ge = setup_training_generator(model)
generated = Generated_sample(model, vocab, sess_ge)
print("Start pre-training generator......")
run_pre_train_generator(model, batcher, 30, sess_ge, saver_ge,
train_dir_ge, generated)
##################################################################################################
srl_generator_model = Srl_Generator(hps_srl_generator, vocab)
sess_srl_ge, saver_srl_ge, train_dir_srl_ge = setup_training_srl_generator(
srl_generator_model)
util.load_ckpt(saver_ge, sess_ge, ckpt_dir="train-generator")
util.load_ckpt(saver_sc, sess_sc, ckpt_dir="train-sc-generator")
srl_generated = Generated_srl_sample(srl_generator_model, vocab,
sess_srl_ge)
whole_generated = Generated_whole_sample(model, srl_generator_model,
vocab, sess_ge, sess_srl_ge,
batcher, srl_batcher)
print("Start pre-training srl_generator......")
run_pre_train_srl_generator(srl_generator_model, batcher, srl_batcher,
20, sess_srl_ge, saver_srl_ge,
train_dir_srl_ge, srl_generated,
whole_generated)
loss_window = 0
t0 = time.time()
print("begin reinforcement learning:")
for epoch in range(10):
loss_window = 0.0
batcher = GenBatcher(vocab, hps_generator)
srl_batcher = Srl_GenBatcher(vocab, hps_srl_generator)
batches = batcher.get_batches(mode='train')
srl_batches = srl_batcher.get_batches(mode='train')
sc_batches = sc_batcher.get_batches(mode='train')
len_sc = len(sc_batches)
for i in range(min(len(batches), len(srl_batches))):
current_batch = batches[i]
current_srl_batch = srl_batches[i]
current_sc_batch = sc_batches[i % (len_sc - 1)]
results = model.run_pre_train_step(sess_ge, current_batch)
loss_list = results['without_average_loss']
example_skeleton_list = current_batch.original_review_outputs
example_text_list = current_batch.original_target_sentences
new_batch = sc_batcher.get_text_queue(example_skeleton_list,
example_text_list,
loss_list)
results_sc = sc_model.run_rl_train_step(sess_sc, new_batch)
loss = results_sc['loss']
loss_window += loss
results_srl = srl_generator_model.run_pre_train_step(
sess_srl_ge, current_srl_batch)
loss_list_srl = results_srl['without_average_loss']
example_srl_text_list = current_srl_batch.orig_outputs
example_skeleton_srl_list = current_srl_batch.orig_inputs
new_batch = sc_batcher.get_text_queue(
example_skeleton_srl_list, example_srl_text_list,
loss_list_srl)
results_sc = sc_model.run_rl_train_step(sess_sc, new_batch)
loss = results_sc['loss']
loss_window += loss
results_sc = sc_model.run_rl_train_step(
sess_sc, current_sc_batch)
loss = results_sc['loss']
loss_window += loss
train_step = results['global_step']
if train_step % 100 == 0:
t1 = time.time()
tf.logging.info(
'seconds for %d training generator step: %.3f ',
train_step, (t1 - t0) / 300)
t0 = time.time()
tf.logging.info('loss: %f', loss_window /
100) # print the loss to screen
loss_window = 0.0
train_srl_step = results_srl['global_step']
if train_srl_step % 10000 == 0:
saver_sc.save(sess_sc,
train_dir_sc + "/model",
global_step=results_sc['global_step'])
saver_ge.save(sess_ge,
train_dir_ge + "/model",
global_step=train_step)
saver_srl_ge.save(sess_srl_ge,
train_dir_srl_ge + "/model",
global_step=train_srl_step)
srl_generated.generator_max_example(
srl_batcher.get_batches("validation"),
"to_seq_max_generated/valid/" +
str(int(train_srl_step / 30000)) + "_positive",
"to_seq_max_generated/valid/" +
str(int(train_srl_step / 30000)) + "_negative")
srl_generated.generator_max_example(
srl_batcher.get_batches("test"),
"to_seq_max_generated/test/" +
str(int(train_srl_step / 30000)) + "_positive",
"to_seq_max_generated/test/" +
str(int(train_srl_step / 30000)) + "_negative")
whole_generated.generator_max_example(
batcher.get_batches("test-validation"),
"max_generated_final/valid/" +
str(int(train_srl_step / 30000)) + "_positive",
"max_generated_final/valid/" +
str(int(train_srl_step / 30000)) + "_negative")
whole_generated.generator_max_example(
batcher.get_batches("test-test"),
"max_generated_final/test/" +
str(int(train_srl_step / 30000)) + "_positive",
"max_generated_final/test/" +
str(int(train_srl_step / 30000)) + "_negative")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-train"),
"data/" + str(0) + "/train_skeleton.txt")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-valid"),
"data/" + str(0) + "/valid_skeleton.txt")
sc_generated.generator_max_example_test(
sc_batcher.get_batches("pre-test"),
"data/" + str(0) + "/test_skeleton.txt")
merge("data/story/train_process.txt", "data/0/train_skeleton.txt",
"data/0/train.txt")
merge("data/story/validation_process.txt",
"data/0/valid_skeleton.txt", "data/0/valid.txt")
merge("data/story/test_process.txt", "data/0/test_skeleton.txt",
"data/0/test.txt")
else:
raise ValueError("The 'mode' flag must be one of train/eval/decode")
def run_eval(model, batcher, word_vector):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph() # build the graph
saver = tf.train.Saver(
max_to_keep=3) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
if FLAGS.embedding:
sess.run(tf.global_variables_initializer(),
feed_dict={model.embedding_place: word_vector})
eval_dir = os.path.join(
FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = restore_best_eval_model(
) # will hold the best loss achieved so far
train_step = 0
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
processed_batch = 0
avg_losses = []
# evaluate for 100 * batch_size before comparing the loss
# we do this due to memory constraint, best to run eval on different machines with large batch size
# while processed_batch < 100 * FLAGS.batch_size:
processed_batch += FLAGS.batch_size
batch = batcher.next_batch() # get the next batch
tf.logging.info('run eval step on seq2seq model.')
t0 = time.time()
results = model.run_eval_step(sess, batch, train_step)
t1 = time.time()
tf.logging.info('experiment: {}'.format(FLAGS.exp_name))
tf.logging.info('processed_batch: {}, seconds for batch: {}'.format(
processed_batch, t1 - t0))
printer_helper = {}
loss = printer_helper['pgen_loss'] = results['pgen_loss']
if FLAGS.coverage:
printer_helper['coverage_loss'] = results['coverage_loss']
if FLAGS.rl_training or FLAGS.ac_training:
printer_helper['rl_cov_total_loss'] = results[
'reinforce_cov_total_loss']
loss = printer_helper['pointer_cov_total_loss'] = results[
'pointer_cov_total_loss']
if FLAGS.rl_training or FLAGS.ac_training:
printer_helper['shared_loss'] = results['shared_loss']
printer_helper['rl_loss'] = results['rl_loss']
printer_helper['rl_avg_logprobs'] = results['rl_avg_logprobs']
if FLAGS.rl_training:
printer_helper['sampled_r'] = np.mean(
results['sampled_sentence_r_values'])
printer_helper['greedy_r'] = np.mean(
results['greedy_sentence_r_values'])
printer_helper['r_diff'] = printer_helper[
'greedy_r'] - printer_helper['sampled_r']
for (k, v) in printer_helper.items():
if not np.isfinite(v):
raise Exception("{} is not finite. Stopping.".format(k))
tf.logging.info('{}: {}\t'.format(k, v))
# add summaries
summaries = results['summaries']
train_step = results['global_step']
print(train_step)
summary_writer.add_summary(summaries, train_step)
# calculate running avg loss
avg_losses.append(
calc_running_avg_loss(np.asscalar(loss), running_avg_loss,
train_step, summary_writer))
tf.logging.info('-------------------------------------------')
running_avg_loss = np.mean(avg_losses)
tf.logging.info('==========================================')
tf.logging.info('best_loss: {}\trunning_avg_loss: {}\t'.format(
best_loss, running_avg_loss))
tf.logging.info('==========================================')
# If running_avg_loss is best so far, save this checkpoint (early stopping).
# These checkpoints will appear as bestmodel-<iteration_number> in the eval dir
if best_loss is None or running_avg_loss < best_loss:
print(train_step)
tf.logging.info(
'Found new best model with %.3f running_avg_loss. Saving to %s',
running_avg_loss, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = running_avg_loss
# flush the summary writer every so often
if train_step % 100 == 0:
summary_writer.flush()
def run_eval(model, batcher, re_vocab, embed):
"""Repeatedly runs eval iterations, logging to screen and writing summaries. Saves the model with the best loss seen so far."""
model.build_graph(embed) # build the graph
saver = tf.train.Saver(
max_to_keep=10) # we will keep 3 best checkpoints at a time
sess = tf.Session(config=util.get_config())
eval_dir = os.path.join(
FLAGS.log_root, "eval") # make a subdir of the root dir for eval data
bestmodel_save_path = os.path.join(
eval_dir,
'bestmodel') # this is where checkpoints of best models are saved
summary_writer = tf.summary.FileWriter(eval_dir)
running_avg_loss = 0 # the eval job keeps a smoother, running average loss to tell it when to implement early stopping
best_loss = None # will hold the best loss achieved so far
count = 0
while True:
_ = util.load_ckpt(saver, sess) # load a new checkpoint
batch = batcher.next_batch() # get the next batch
if not os.path.exists(FLAGS.model_dir):
os.makedirs(FLAGS.model_dir)
os.makedirs(FLAGS.system_dir)
# run eval on the batch
t0 = time.time()
# results = model.run_eval_step(sess, batch)
step_output = model.run_eval_step(sess, batch)
t1 = time.time()
#tf.logging.info('seconds for batch: %.2f', t1-t0)
(summaries, loss, train_step) = step_output[0]
(out_decoder_outputs, out_sent_decoder_outputs,
final_dists) = step_output[1]
(step_loss, word_loss, sent_loss, word_loss_null, sent_loss_null,
switch_loss) = step_output[2]
coverage_loss = 0.0
running_avg_loss = calc_running_avg_loss(np.asscalar(loss),
running_avg_loss,
summary_writer, train_step)
if best_loss is None or running_avg_loss < (best_loss):
if best_loss is None:
best_loss = 0.0
tf.logging.info(
'Found new best model with %.3f running_avg_loss. Saving to %s',
running_avg_loss, bestmodel_save_path)
saver.save(sess,
bestmodel_save_path,
global_step=train_step,
latest_filename='checkpoint_best')
best_loss = running_avg_loss
last_step = train_step
tf.logging.info('loss: %f rloss: %f', loss,
rloss) # print the loss to screen
tf.logging.info(
'step_loss: %f word_loss: %f ,sent_loss: %f ,word_loss_null: %f,sent_loss_null: %f ,switch_loss: %f,cover_loss: %f',
step_loss, word_loss, sent_loss, word_loss_null, sent_loss_null,
switch_loss, coverage_loss)
os.system("rm -rf " + FLAGS.model_dir + ' ' + FLAGS.system_dir)
count = count + 1
if train_step % 100 == 0:
summary_writer.flush()
def decode(self):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely, loading latest checkpoint at regular intervals"""
t0 = time.time()
counter = 0
total = len(glob.glob(self._batcher._data_path)) * 1000
pbar = tqdm(total=total)
while True:
batch = self._batcher.next_batch(
) # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
logging.info(
"Decoder has finished reading dataset for single_pass.")
logging.info("Output has been saved in %s and %s.",
self._rouge_ref_dir, self._rouge_dec_dir)
if len(os.listdir(self._rouge_ref_dir)) != 0:
logging.info("Now starting ROUGE eval...")
results_dict = rouge_functions.rouge_eval(
self._rouge_ref_dir, self._rouge_dec_dir)
rouge_functions.rouge_log(results_dict, self._decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
all_original_abstract_sents = batch.all_original_abstracts_sents[0]
raw_article_sents = batch.raw_article_sents[0]
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
abstract_withunks = data.show_abs_oovs(
original_abstract, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
decoded_words, decoded_output, best_hyp = decode_example(
self._sess, self._model, self._vocab, batch, counter,
self._batcher._hps)
if FLAGS.single_pass:
if counter < 1000:
self.write_for_human(raw_article_sents,
all_original_abstract_sents,
decoded_words, counter)
rouge_functions.write_for_rouge(
all_original_abstract_sents,
None,
counter,
self._rouge_ref_dir,
self._rouge_dec_dir,
decoded_words=decoded_words
) # write ref summary and decoded summary to file, to eval with pyrouge later
if FLAGS.attn_vis:
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens, counter
) # write info to .json file for visualization tool
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks,
decoded_output) # log output to screen
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens,
counter) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
logging.info(
'We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(self._saver, self._sess)
t0 = time.time()
pbar.update(1)
pbar.close()
def decode(self):
"""Decode examples until data is exhausted (if FLAGS.single_pass) and return, or decode indefinitely, loading latest checkpoint at regular intervals"""
t0 = time.time()
counter = 0
while True:
batch = self._batcher.next_batch(
) # 1 example repeated across batch
if batch is None: # finished decoding dataset in single_pass mode
assert FLAGS.single_pass, "Dataset exhausted, but we are not in single_pass mode"
tf.logging.info(
"Decoder has finished reading dataset for single_pass.")
tf.logging.info(
"Output has been saved in %s and %s. Now starting ROUGE eval...",
self._rouge_ref_dir, self._rouge_dec_dir)
results_dict = rouge_eval(self._rouge_ref_dir,
self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
return
original_article = batch.original_articles[0] # string
original_abstract = batch.original_abstracts[0] # string
original_abstract_sents = batch.original_abstracts_sents[
0] # list of strings
article_withunks = data.show_art_oovs(original_article,
self._vocab) # string
abstract_withunks = data.show_abs_oovs(
original_abstract, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None)) # string
# Run beam search to get best Hypothesis
best_hyp = beam_search.run_beam_search(self._sess, self._model,
self._vocab, batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_hyp.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self._vocab,
(batch.art_oovs[0] if FLAGS.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(
data.STOP_DECODING) # index of the (first) [STOP] symbol
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
decoded_output = ' '.join(decoded_words) # single string
if FLAGS.single_pass:
self.write_for_rouge(
original_abstract_sents, decoded_words, counter
) # write ref summary and decoded summary to file, to eval with pyrouge later
counter += 1 # this is how many examples we've decoded
else:
print_results(article_withunks, abstract_withunks,
decoded_output) # log output to screen
self.write_for_attnvis(
article_withunks, abstract_withunks, decoded_words,
best_hyp.attn_dists, best_hyp.p_gens
) # write info to .json file for visualization tool
# Check if SECS_UNTIL_NEW_CKPT has elapsed; if so return so we can load a new checkpoint
t1 = time.time()
if t1 - t0 > SECS_UNTIL_NEW_CKPT:
tf.logging.info(
'We\'ve been decoding with same checkpoint for %i seconds. Time to load new checkpoint',
t1 - t0)
_ = util.load_ckpt(self._saver, self._sess)
t0 = time.time()
def run_epoch(lm,
session,
batches,
summary_writer,
train_dir,
train_step,
saver,
hps,
best_loss,
avg_loss,
save_all=True):
print_interval = 10.0
total_batches = 0
total_words = 0
start_time = time.time()
tick_time = start_time # for showing status
i = 0
exception_count = 0
init_exception_count = 0
batches_skipped = 0
for batch in batches:
try:
results = runTrainStep(lm, session, batch)
loss = results['loss']
coverage_loss = results['coverage_loss']
if not np.isfinite(loss):
raise Exception("Loss is not finite.")
summaries = results['summaries']
train_step = results['global_step']
summary_writer.add_summary(summaries, train_step)
avg_loss = util.running_avg_loss(np.asscalar(loss), avg_loss,
summary_writer, train_step)
if best_loss is None or avg_loss < best_loss:
#saver.save(session, train_dir, global_step=train_step, latest_filename='checkpoint_best')
best_loss = avg_loss
total_batches = i + 1
total_words += len(batch.original_articles)
i = i + 1
if (time.time() - tick_time >= print_interval):
avg_wps = total_words / (time.time() - start_time)
print(
" [batch {:d}]: seen {:d} examples : {:.1f} eps, Loss: {:.3f}, Avg loss: {:.3f}, Best loss: {:.3f}, cov loss: {:.3f}"
.format(i, total_words, avg_wps, loss, avg_loss, best_loss,
coverage_loss))
tick_time = time.time() # reset time ticker
if save_all:
saver.save(session,
train_dir,
global_step=train_step,
latest_filename='checkpoint')
if train_step % 100 == 0:
summary_writer.flush()
except Exception as e:
if (exception_count <= 10):
print(f' [EXCEPTION]: ', str(e), '; Restoring model params')
exception_count = exception_count + 1
batches_skipped = batches_skipped + 1
util.load_ckpt(saver, session, hps, hps.log_root)
continue
else:
print(' [EXCEPTION LIMIT EXCEEDED]: Batches skipped:',
batches_skipped, '; Error : ', str(e))
raise e
time_total = pretty_timedelta(since=start_time)
print(
f" [END] Training complete: Total examples : {total_words}; Total time: {time_total}"
)
return avg_loss, best_loss, train_step