def train():
# Download and split data.
common.split_data(TIME_STEPS)
# Build and compile the model.
model = build_model()
# model.save_weights(CKP_PATH.format(epoch=0))
# Load last checkpoint if any.
# model.load_weights(
# tf.train.latest_checkpoint(
# os.path.dirname(CKP_PATH)
# )
# )
train_idg = generators.TimeDistributedImageDataGenerator(
rotation_range=30,
zoom_range=0.15,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.15,
horizontal_flip=True,
rescale=1. / 255,
time_steps=TIME_STEPS,
)
validation_idg = generators.TimeDistributedImageDataGenerator(
time_steps=TIME_STEPS,
)
history = model.fit(
train_idg.flow_from_directory(
common.TRAIN_DATA_PATH,
target_size=(48, 48),
batch_size=BATCH_SIZE,
class_mode='sparse',
shuffle=False,
color_mode='rgb',
# classes=['agree_pure', 'agree_considered'],
# save_to_dir='./data/train'
),
validation_data=validation_idg.flow_from_directory(
common.VALIDATION_DATA_PATH,
target_size=(48, 48),
batch_size=BATCH_SIZE,
class_mode='sparse',
shuffle=False,
color_mode='rgb',
# classes=['agree_pure', 'agree_considered'],
# save_to_dir='./data/test'
),
callbacks=CALLBACKS,
epochs=EPOCHS,
)
model.save(SVD_PATH)
common.plot_acc_loss(history, PLT_PATH)
import tensorflow as tf
import pandas as pd
import common
class History:
history = {}
history = History()
df = pd.read_csv('../models/trial-final/32/12/log.csv')
df = df[['accuracy', 'loss', 'val_accuracy', 'val_loss']]
d = df.to_dict()
for x in d:
history.history[x] = [v for k, v in d[x].items()]
common.plot_acc_loss(history, '../models/trial-final/32/12/plot.png')
def train():
tf.keras.utils.get_file(
fname='cec-videos-augmented.tar.gz',
origin='https://unir-tfm-cec.s3.us-east-2.amazonaws.com/cec-videos-augmented.tar.gz',
extract=True
)
for batch_size in BATCH_SIZE:
for time_steps in TIME_STEPS:
path = TRL_PATH + f'/{batch_size}/{time_steps}'
os.makedirs(path, exist_ok=True)
# Build and compile the model.
model = build_model(time_steps, len(CLASSES))
data_aug = tf.keras.preprocessing.image.ImageDataGenerator(
preprocessing_function=tf.keras.applications.mobilenet.preprocess_input
)
train_idg = SlidingFrameGenerator(
classes=CLASSES,
glob_pattern=common.HOME + '/.keras/datasets/cec-videos-augmented/{classname}/*.avi',
nb_frames=time_steps,
split_val=.2,
shuffle=True,
batch_size=batch_size,
target_shape=(224, 224),
nb_channel=3,
transformation=data_aug,
use_frame_cache=False
)
validation_idg = train_idg.get_validation_generator()
# Configure callbacks
callbacks = [
tf.keras.callbacks.ModelCheckpoint(
filepath=path + '/model',
monitor='val_accuracy',
mode='max',
save_best_only=True,
verbose=1,
),
tf.keras.callbacks.EarlyStopping(
monitor='val_loss',
mode='min',
verbose=1,
patience=int(EPOCHS * .01)
),
tf.keras.callbacks.CSVLogger(
filename=path + '/log.csv'
),
tf.keras.callbacks.TensorBoard(
log_dir=path + '/tb',
histogram_freq=1
),
tf.keras.callbacks.ReduceLROnPlateau(
verbose=1
),
]
history = model.fit(
train_idg,
validation_data=validation_idg,
callbacks=callbacks,
epochs=EPOCHS,
)
common.plot_acc_loss(history, path + '/plot.png')
def train():
tf.keras.utils.get_file(
fname='cec-videos.tar.gz',
origin=
'https://unir-tfm-cec.s3.us-east-2.amazonaws.com/cec-videos.tar.gz',
extract=True)
data = pandas.DataFrame(None,
columns=[
'trial', 'batch_size', 'time_steps', 'cycle',
'files', 'sequences'
])
data['trial'] = TRIAL
for batch_size in BATCH_SIZE:
for time_steps in TIME_STEPS:
path = TRL_PATH + f'/{batch_size}/{time_steps}'
os.makedirs(path, exist_ok=True)
# Build and compile the model.
model = build_model(time_steps, len(CLASSES))
data_aug = tf.keras.preprocessing.image.ImageDataGenerator(
zoom_range=.1,
horizontal_flip=True,
rotation_range=8,
width_shift_range=.2,
height_shift_range=.2,
preprocessing_function=tf.keras.applications.mobilenet_v2.
preprocess_input)
train_idg = SlidingFrameGenerator(classes=CLASSES,
glob_pattern=common.VIDEOS_PATH,
nb_frames=time_steps,
split_val=.2,
shuffle=True,
batch_size=batch_size,
target_shape=(224, 224),
nb_channel=3,
transformation=data_aug,
use_frame_cache=False)
keras_video.utils.show_sample(train_idg)
validation_idg = train_idg.get_validation_generator()
row = {
'trial': TRIAL,
'batch_size': batch_size,
'cycle': 'training',
'time_steps': time_steps,
'files': train_idg.files_count,
'sequences': len(train_idg.vid_info)
}
data = data.append(row, ignore_index=True)
row = {
'trial': TRIAL,
'batch_size': batch_size,
'cycle': 'validation',
'time_steps': time_steps,
'files': validation_idg.files_count,
'sequences': len(validation_idg.vid_info)
}
data = data.append(row, ignore_index=True)
# Configure callbacks
callbacks = [
tf.keras.callbacks.ModelCheckpoint(
filepath=path + '/model',
monitor='val_accuracy',
mode='max',
save_best_only=True,
verbose=1,
),
tf.keras.callbacks.EarlyStopping(monitor='val_loss',
mode='min',
verbose=1,
patience=int(EPOCHS * .01)),
tf.keras.callbacks.CSVLogger(filename=path + '/log.csv'),
tf.keras.callbacks.TensorBoard(log_dir=path + '/tb',
histogram_freq=1),
tf.keras.callbacks.ReduceLROnPlateau(verbose=1),
]
history = model.fit(
train_idg,
validation_data=validation_idg,
callbacks=callbacks,
epochs=EPOCHS,
)
common.plot_acc_loss(history, path + '/plot.png')
data.to_csv('trial_01.csv')
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
history = model.fit(
train_idg.flow_from_directory(
common.TRAIN_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/train'
),
epochs=50,
validation_data=validation_idg.flow_from_directory(
common.TEST_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/test'
))
model.save('models/01/ResNet50')
common.plot_acc_loss(history, '../models/01/ResNet50/plot.png')
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
history = model.fit(
train_idg.flow_from_directory(
common.TRAIN_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/train'
),
epochs=50,
validation_data=validation_idg.flow_from_directory(
common.TEST_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/test'
))
model.save('models/04/ResNet152')
common.plot_acc_loss(history, '../models/04/ResNet152/plot.png')
metrics=['accuracy']
)
history = model.fit(
train_idg.flow_from_directory(
common.TRAIN_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/train'
),
epochs=50,
validation_data=validation_idg.flow_from_directory(
common.TEST_DATA_PATH,
target_size=(224, 224),
batch_size=32,
class_mode='categorical',
shuffle=True,
seed=common.SEED_VALUE,
# classes=['agree_pure']
# save_to_dir='./data/test'
)
)
model.save('models/03/DenseNet201')
common.plot_acc_loss(history, '../models/03/DenseNet201/plot.png')