def load_custom_model(root_path, object_image_features_filename,
model_filename):
from Attention import Attention
from models import NIC
from generator import Generator
preprocessed_data_path = root_path + 'preprocessed_data/'
generator = Generator(
data_path=preprocessed_data_path,
batch_size=configs["batch_size"],
image_features_filename=object_image_features_filename)
if (configs['w2v_weights']):
embedding_weights = generator.embedding_matrix
else:
embedding_weights = None
model = NIC(max_token_length=generator.MAX_TOKEN_LENGTH,
vocabulary_size=generator.VOCABULARY_SIZE,
tweet_max_len=configs['tweet_max_len'],
tweet_max_words=configs['tweet_max_words'],
rnn='gru',
num_image_features=generator.IMG_FEATS,
hidden_size=256,
embedding_size=128,
embedding_weights=embedding_weights)
model.load_weights(model_filename)
# model.load_model(model_filename)
return model
preprocessed_data_path = root_path + 'preprocessed_data/'
generator = Generator(data_path=preprocessed_data_path,
batch_size=batch_size,
image_features_filename=object_image_features_filename)
num_training_samples = generator.training_dataset.shape[0]
num_validation_samples = generator.validation_dataset.shape[0]
print('Number of training samples:', num_training_samples)
print('Number of validation samples:', num_validation_samples)
print(generator.VOCABULARY_SIZE)
print(generator.IMG_FEATS)
model = NIC(max_token_length=generator.MAX_TOKEN_LENGTH,
vocabulary_size=generator.VOCABULARY_SIZE,
rnn='gru',
num_image_features=generator.IMG_FEATS,
hidden_size=180,
embedding_size=150)
# print("test")
# model.load_weights("../trained_models/hashtag/hashtag_weights.240-4.3745.hdf5")
# model.compile(loss='categorical_crossentropy',
# optimizer = 'adam',
# metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print(model.summary())
print('Number of parameters:', model.count_params())
plot_model(model, show_shapes=True, to_file='./nuswide_NIH.png')
print('Number of validation samples:', num_validation_samples)
print(generator.VOCABULARY_SIZE)
print(generator.IMG_FEATS)
#generator.flow(mode='train')
#generator.format_to_one_hot("beach sea trip island japan")
if (configs['w2v_weights']):
embedding_weights = generator.embedding_matrix
else:
embedding_weights = None
model = NIC(max_token_length=generator.MAX_TOKEN_LENGTH,
vocabulary_size=generator.VOCABULARY_SIZE,
tweet_max_len=configs['tweet_max_len'],
tweet_max_words=configs['tweet_max_words'],
rnn='gru',
num_image_features=generator.IMG_FEATS,
hidden_size=256,
embedding_size=128,
embedding_weights=embedding_weights)
# print("test")
# model.load_weights("../trained_models/hashtag/hashtag_weights.240-4.3745.hdf5")
# model.compile(loss='categorical_crossentropy',
# optimizer = 'adam',
# metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print(model.summary())
print(generator.VOCABULARY_SIZE)
print(generator.IMG_FEATS)
#generator.flow(mode='train')
#generator.format_to_one_hot("beach sea trip island japan")
# model = NIC(max_token_length=12,
# vocabulary_size=generator.VOCABULARY_SIZE,
# rnn='gru',
# num_image_features=generator.IMG_FEATS,
# hidden_size=128,
# embedding_size=128)
model = NIC(max_token_length=12,
vocabulary_size=generator.VOCABULARY_SIZE,
rnn='gru',
num_image_features=generator.IMG_FEATS,
hidden_size=128,
embedding_size=128)
# print("test")
# model.load_weights("../trained_models/hashtag/hashtag_weights.240-4.3745.hdf5")
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print(model.summary())
print('Number of parameters:', model.count_params())
plot_model(model, show_shapes=True, to_file='../images/NIC.png')
training_history_filename = preprocessed_data_path + 'training_hashtag_history.log'
csv_logger = CSVLogger(training_history_filename, append=False)
model_names = ('../trained_models/custom/' +
data_manager.preprocess()
print(data_manager.captions[0])
print(data_manager.word_frequencies[0:20])
preprocessed_data_path = root_path + 'preprocessed_data/'
generator = Generator(data_path=preprocessed_data_path,
batch_size=batch_size)
num_training_samples = generator.training_dataset.shape[0]
num_validation_samples = generator.validation_dataset.shape[0]
print('Number of training samples:', num_training_samples)
print('Number of validation samples:', num_validation_samples)
model = NIC(max_token_length=generator.MAX_TOKEN_LENGTH,
vocabulary_size=generator.VOCABULARY_SIZE,
rnn='gru',
num_image_features=generator.IMG_FEATS,
hidden_size=128,
embedding_size=128)
model.compile(loss='categorical_crossentropy',
optimizer = 'adam',
metrics=['accuracy'])
print(model.summary())
print('Number of parameters:', model.count_params())
training_history_filename = preprocessed_data_path + 'training_history.log'
csv_logger = CSVLogger(training_history_filename, append=False)
model_names = ('../trained_models/IAPR_2012/' +
'iapr_weights.{epoch:02d}-{val_loss:.2f}.hdf5')
model_checkpoint = ModelCheckpoint(model_names,