reworked first part add config for verbose output (86d557a6) · Commits · Softwareprojekt 2018 - Jason Marinco Rudolf Samuel / O Caption My Caption

caption-lib/lstm/Models/coco_model.py

+5 −0

Original line number	Diff line number	Diff line
		@@ -11,6 +11,7 @@
		# =========================================================================================

		from __future__ import division
		from lstm.configuration import current as _config

		import tensorflow as tf

		@@ -57,6 +58,7 @@ class coco_model(object):
		self.features = tf.placeholder(tf.float32, [None, self.D], name="features_placeholder_model")
		self.captions = tf.placeholder(tf.int32, [None, self.L], name="captions_placeholder_model")

		if _config.trainer.verbose:
		print("\n#>\t CREATE MODEL")
		print("#>\tdim_feature:", dim_feature)
		print("#>\tdim_embed", dim_embed)
		@@ -86,6 +88,7 @@ class coco_model(object):
		b_c = tf.get_variable('b_c', [self.H], initializer=self.const_initializer)
		c = tf.nn.tanh(tf.matmul(features_mean, w_c) + b_c)

		if _config.trainer.verbose:
		print("\n#>\t Initial LSTM")
		print("#>\t:", features_mean[0])
		print("#>\tc:", c)
		@@ -97,6 +100,7 @@ class coco_model(object):
		with tf.variable_scope('word_embedding', reuse=reuse):
		w = tf.get_variable('w', [self.V, self.M], initializer=self.emb_initializer)
		x = tf.nn.embedding_lookup(w, inputs, name='word_vector') # (N, T, M) or (N, M)
		if _config.trainer.verbose:
		print("\n#>\t Embedding Loopkup")
		print("#>\tV:", self.V)
		print("#>\tM:", self.M)
		@@ -188,6 +192,7 @@ class coco_model(object):
		alpha_list = []
		lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(num_units=self.H)

		if _config.trainer.verbose:
		print("\n#>\t Build Model")
		print("#>\tmask:", mask)
		print("#>\tbatch_normalisation:", self._batch_norm)

caption-lib/lstm/Trainers/coco_trainer.py

+6 −3

Original line number	Diff line number	Diff line
		@@ -60,6 +60,7 @@ class coco_trainer(object):
		if not os.path.exists(self.log_path):
		os.makedirs(self.log_path)

		if _config.trainer.verbose:
		print("\n#>\t INIT TRAINING")
		print("#>\ttf.Optimizer", self.optimizer)
		print("#>\tlearningrate", self.learning_rate)
		@@ -85,6 +86,7 @@ class coco_trainer(object):
		loss = self.model.build_model()
		tf.get_variable_scope().reuse_variables()
		_, _, generated_captions = self.model.build_sampler(max_len=20)
		if _config.trainer.verbose:
		print("\n#>\t INIT LOSS")
		print("#>\tloss", loss)

		@@ -95,6 +97,7 @@ class coco_trainer(object):
		grads = tf.gradients(loss, tf.trainable_variables())
		grads_and_vars = list(zip(grads, tf.trainable_variables()))
		train_op = optimizer.apply_gradients(grads_and_vars=grads_and_vars)
		if _config.trainer.verbose:
		print("\n#>\t TRAIN Operation")
		print("#>\toptimizer\n", optimizer)
		# print("#>\tgradients\n", grads)
		@@ -114,7 +117,7 @@ class coco_trainer(object):
		#summary_op = tf.merge_all_summaries()
		summary_op = tf.summary.merge_all()

		print("\n#>\t TRAINING")
		print("\n#> TRAINING")
		print("#>\tTotal number of epochs: %d" %self.n_epochs)
		print("#>\tData size: %d" %n_examples)
		print("#>\tBatch size: %d" %self.batch_size)
		@@ -141,7 +144,7 @@ class coco_trainer(object):
		print("#>\tcaption shape:", captions.shape)
		print("#>\timageId shape:", image_idxs.shape)
		for e in range(self.n_epochs):
		print("#>\tEPOCH")
		print("#> EPOCH")

		rand_idxs = np.random.permutation(n_examples)
		m_captions = captions[rand_idxs]
		@@ -154,7 +157,7 @@ class coco_trainer(object):
		image_idxs_batch = m_image_idxs[iself.batch_size:(i+1)self.batch_size]
		features_batch = m_features[iself.batch_size:(i+1)self.batch_size]
		if i % 10 == 0:
		print("#> Batch Shapes")
		print()
		print("#>\tBatch: " + str(i) + "\tEpoch: " + str(e) + "\tBatch size: " + str(iself.batch_size) + " - " + str((i+1)self.batch_size))
		print("#>\tfrom Feature Batch:", features_batch.shape, features_batch[0].shape)
		print("#>\tfrom Captions Batch:", captions_batch.shape)

caption-lib/lstm/configuration.py

+3 −2

Original line number	Diff line number	Diff line
		@@ -45,6 +45,7 @@ class default_trainer(object):
		num_layers = 2
		num_epochs = 60
		num_steps = 35
		vector_size = 2048
		batch_size = 20
		dropout = 0.5
		init_scale = 0.05
		@@ -84,10 +85,10 @@ class current(default_reader, default_validator, default_trainer, default_paths)
		# e.g. validator.read_test = "/test1234"

		reader.reader_on = False
		reader.verbose = True
		reader.verbose = False

		validator.validator_on = False

		trainer.train_on = True
		trainer.verbose = True
		trainer.verbose = False
		trainer.num_epochs = 3

caption-lib/lstm/lstm_main.py

+3 −3

Original line number	Diff line number	Diff line
		@@ -40,8 +40,8 @@ def main(self, parameter_list):

		model = _model.coco_model( word_to_idx,
		current_reader,
		dim_feature=[20, 2048],
		dim_embed=2048,
		dim_feature=[_config.trainer.batch_size, _config.trainer.vector_size],
		dim_embed=_config.trainer.vector_size,
		dim_hidden=1024,
		n_time_step=16,
		prev2out=True,
		@@ -55,7 +55,7 @@ def main(self, parameter_list):
		train_data,
		val_data,
		n_epochs=3,
		batch_size=20,
		batch_size=_config.trainer.batch_size,
		update_rule='adam',
		learning_rate=0.001,
		print_every=10,

Admin message