def main():
"""
main testing method
:return:
"""
vocab = text_utils.get_model_vocab()
reduced_test_data = data_generator.gen_dataset(os.path.join("Data",
"Meeting Identifier",
"Reduced High Res Test Data"),
vocab=vocab,
shuffle=False,
input_dim=constants.meeting_dimensions_420p)
test_data = data_generator.gen_dataset(os.path.join("Data",
"Meeting Identifier",
"High Res Test Data"),
vocab=vocab,
shuffle=False,
input_dim=constants.meeting_dimensions_420p)
model = load_model(constants.text_recognition)
model.evaluate(reduced_test_data)
model.evaluate(test_data)
"""length_accuracy(training_data)"""
"""print_learning_curves("Data/Meeting Identifier/Reduced High Res Training Data",
def main():
"""
main method
:return:
"""
vocab = text_utils.get_model_vocab()
read_img_analysis(
os.path.join("Data", "Meeting Identifier",
"Reduced High Res Training Data"), vocab)
"""text = read_img(os.path.join("Data",
def main():
"""
main testing method
:return:
"""
path = "Data/YOLO/Training Data"
vocab = text_utils.get_model_vocab()
dataset = gen_dataset(path, batch_size=1, vocab=vocab, shuffle=False)
for x, y in dataset:
print(y.shape)
print(y)
break
def main():
"""
main testing method
:return:
"""
training_path = "Data/YOLO/Training Data"
vocab = text_utils.get_model_vocab()
dataset = data_generator.gen_dataset(
training_path,
vocab=vocab,
batch_size=1,
shuffle=False,
image_dim=constants.meeting_dimensions_420p)
model = train_network(dataset.take(1), vocab, reload=False)
model.save(constants.letter_detection)
def main():
"""
main testing method
:return:
"""
vocab = text_utils.get_model_vocab()
dataset = data_generator.gen_dataset("Data/YOLO/Training Data",
vocab,
batch_size=100,
verbose=False,
shuffle=False,
image_dim=constants.meeting_dimensions_420p)
get_letters(dataset.take(1),
vocab,
load())
"""model = load()
def main():
"""
main method
:return:
"""
vocab = text_utils.get_model_vocab()
training_data = data_generator.gen_dataset(
os.path.join("Data", "Meeting Identifier",
"Reduced High res Training Data"),
input_dim=constants.meeting_dimensions_420p,
# batch_size=None,
vocab=vocab)
test_data = data_generator.gen_dataset(
os.path.join("Data", "Meeting Identifier", "High Res Test Data"),
input_dim=constants.meeting_dimensions_420p,
vocab=vocab)
"""model = initalizer.init_nn(vocab,
image_dimensions=constants.meeting_dimensions_420p)
model.summary()
fitter = ModelFitter(model)
fitter.fit(training_data,
epochs=300,
validation_data=test_data)"""
model = train_model(training_data,
test_data,
vocab,
resolution=constants.meeting_dimensions_420p)
model.save(constants.text_recognition)
"""tuner = BayesianOptimization(lambda hp: initalizer.init_nn(vocab,
def print_learning_curves(training_path,
test_path,
run_name,
steps=10,
trials=5,
input_shape=constants.meeting_dimensions):
"""
print the learning curves of a model
:param training_path: path to the training data
:param test_path: path to the test data
:param run_name: name of the current treatment group
:param steps: number of dataset steps to take
:param trials: number of trials to take
:param input_shape: shape of the input images
:return: None
"""
# initialize constants
subset_sizes = data_generator.get_dataset_sizes(training_path)
vocab = text_utils.get_model_vocab()
# load the data
training_data = [data_generator.gen_dataset_batchless(training_path,
i + 1,
vocab,
1,
subset_sizes[i],
input_shape)
for i in range(constants.name_length)]
test_data = data_generator.gen_dataset(test_path,
input_dim=input_shape,
vocab=vocab)
print("treatment", "Dataset Size", "training accuracy", "test accuracy", sep=", ")
# take subsets from the dataset
for i in range(steps):
# increment i
i += 1
# take a random sample from each length of dataset
dataset = take_dataset_sample(training_data, subset_sizes, float(i) / steps)
for j in range(trials):
# initialize the model
model = initalizer.init_nn(vocab,
image_dimensions=input_shape)
# train a model on the dataset
model.fit(dataset,
verbose=0,
epochs=300)
training_acc = model.evaluate(dataset,
verbose=0)
test_acc = model.evaluate(test_data,
verbose=0)
print(run_name,
int(float(i) * sum(subset_sizes) / steps),
training_acc[1],
test_acc[1],
sep=", ")