def testcg(self):
K.set_floatx('float64')
from dlmri_tutorial.keras.layers import mri
A = mri.MulticoilForwardOp(center=True)
AH = mri.MulticoilAdjointOp(center=True)
dc = DCPM(A, AH, weight_init=1.0, max_iter=50, tol=1e-12)
shape = (5, 10, 10, 1)
kshape = (5, 3, 10, 10)
x = tf.complex(tf.random.normal(shape, dtype=tf.float64),
tf.random.normal(shape, dtype=tf.float64))
y = tf.complex(tf.random.normal(kshape, dtype=tf.float64),
tf.random.normal(kshape, dtype=tf.float64))
mask = tf.ones(kshape, dtype=tf.float64)
smaps = tf.complex(tf.random.normal(kshape, dtype=tf.float64),
tf.random.normal(kshape, dtype=tf.float64))
tf_a = tf.Variable(np.array([1.1]), trainable=True, dtype=tf.float64)
tf_b = tf.Variable(np.array([1.1]), trainable=True, dtype=tf.float64)
# perform a gradient check:
epsilon = 1e-5
def compute_loss(a, b):
arg = dc([x * tf.complex(a, tf.zeros_like(a)), y, mask, smaps],
scale=1 / b) # take 1/b
return 0.5 * tf.math.real(tf.reduce_sum(tf.math.conj(arg) * arg))
with tf.GradientTape() as g:
g.watch(x)
# setup the model
tf_loss = compute_loss(tf_a, tf_b)
# backpropagate the gradient
dLoss = g.gradient(tf_loss, [tf_a, tf_b])
grad_a = dLoss[0].numpy()[0]
grad_b = dLoss[1].numpy()[0]
# numerical gradient w.r.t. the input
l_ap = compute_loss(tf_a + epsilon, tf_b).numpy()
l_an = compute_loss(tf_a - epsilon, tf_b).numpy()
grad_a_num = (l_ap - l_an) / (2 * epsilon)
print("grad_x: {:.7f} num_grad_x {:.7f} success: {}".format(
grad_a, grad_a_num,
np.abs(grad_a - grad_a_num) < 1e-4))
self.assertTrue(np.abs(grad_a - grad_a_num) < 1e-4)
# numerical gradient w.r.t. the weights
l_bp = compute_loss(tf_a, tf_b + epsilon).numpy()
l_bn = compute_loss(tf_a, tf_b - epsilon).numpy()
grad_b_num = (l_bp - l_bn) / (2 * epsilon)
print("grad_w: {:.7f} num_grad_w {:.7f} success: {}".format(
grad_b, grad_b_num,
np.abs(grad_b - grad_b_num) < 1e-4))
self.assertTrue(np.abs(grad_b - grad_b_num) < 1e-4)
def _enable_float32(self):
dtype = 'float32'
K.set_floatx(dtype)
K.set_epsilon(1e-7)
return dtype
def _enable_float16(self):
dtype = 'float16'
K.set_floatx(dtype)
K.set_epsilon(1e-4)
return dtype
def build_model(self, input_layer_size):
"""
creation of a neural network for multiclass classification
:param input_layer_size:
:return:
"""
K.set_floatx('float16')
config = tensorflow.ConfigProto(device_count={'GPU': 1, 'CPU': 56})
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 0.5
K.set_session(tensorflow.Session(config=config))
self.model = Sequential()
self.model.add(
Dense(input_layer_size,
dtype=tensorflow.float64,
activation='relu'))
self.model.add(Dense(512, activation='relu'))
self.model.add(Dense(256, activation='relu'))
self.model.add(Dense(128, activation='relu'))
self.model.add(Dense(3, activation='softmax'))
self.model.compile(loss='categorical_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
self.initial_fit()
def _model_tf_v1(n_input, n_output):
from tensorflow.python.keras.layers import Input, Dense, Dropout
session_config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
sess = tf.Session(config=session_config)
K.set_floatx('float64')
# ---------------------
# CUSTOM CHANGES STARTS
# a simple example
ecg_input = Input(shape=(None, n_input), dtype='float64', name='ecg_input')
x = Dense(16, activation='relu')(ecg_input)
x = Dropout(0.1)(x)
x = Dense(8, activation='relu')(x)
x = Dropout(0.1)(x)
bcg_out = Dense(n_output, activation='linear')(x)
model = Model(inputs=ecg_input, outputs=bcg_out)
# CUSTOM CHANGES ENDS
# ---------------------
return model
def __init__(self,
seed,
cutoff,
loss_function="superset",
num_epochs=5000,
learning_rate=0.05,
batch_size=32,
patience=16,
es_val_ratio=0.3,
early_stop_interval=5,
log_losses=True,
hidden_layer_sizes=[32],
activation_function="sigmoid"):
self.network = None
self.logger = logging.getLogger("NeuralNetSuperSetLoss")
self.loss_history = []
self.es_val_history = []
self.seed = seed
self.cutoff = cutoff
self.loss_function = loss_function
self.num_epochs = num_epochs
self.learning_rate = learning_rate
self.batch_size = batch_size
self.patience = patience
self.es_val_ratio = es_val_ratio
self.early_stop_interval = early_stop_interval
self.log_losses = log_losses
self.hidden_layer_sizes = hidden_layer_sizes
self.activation_function = activation_function
K.set_floatx("float64")
def BuildModel(dataShape, modelName, learningRate):
K.set_floatx('float16')
K.set_epsilon(1e-4)
input0 = tf.keras.Input(shape=(dataShape, dataShape, 3),
name='input_0',
dtype='float16') #Scene color
input1 = tf.keras.Input(shape=(dataShape, dataShape, 1),
name='input_1',
dtype='float16') #Depth 0
input2 = tf.keras.Input(shape=(dataShape, dataShape, 1),
name='input_2',
dtype='float16') #Depth -1
input3 = tf.keras.Input(shape=(dataShape, dataShape, 1),
name='input_3',
dtype='float16') #Depth -2
modelFunc = importlib.import_module('Models.' + modelName)
model = modelFunc.MakeModel([input0, input1, input2, input3], dataShape,
modelName)
print("Loaded model from disk")
model.compile(loss=Loss,
optimizer=LossScaleOptimizer(
RMSprop(lr=learningRate, epsilon=1e-4), 1000))
model.summary()
return model
def train_conv_net_cpu(train_data, train_labels, val_data, val_labels,
conv_net_hyperparameters, num_processors, seed, dtype="float32", inter_op_threads=2,trial=None):
"""
Train a convolutional neural network on the CPU.
"""
np.random.seed(seed)
if "get_visible_devices" in dir(tf.config.experimental):
gpus = tf.config.experimental.get_visible_devices("GPU")
else:
gpus = tf.config.get_visible_devices("GPU")
if len(gpus) > 0:
for device in gpus:
tf.config.experimental.set_memory_growth(device, True)
tf.config.threading.set_inter_op_parallelism_threads(inter_op_threads)
tf.config.threading.set_intra_op_parallelism_threads(num_processors)
if tf.__version__[0] == "1":
tf.set_random_seed(seed)
else:
tf.random.set_seed(seed)
K.set_floatx(dtype)
with tf.device("/CPU:0"):
scn = ResNet(**conv_net_hyperparameters)
history = scn.fit(train_data, train_labels, val_x=val_data, val_y=val_labels)
#scn.fit(train_data, train_labels, val_x=val_data, val_y=val_labels)
time_str = pd.Timestamp.now().strftime("%d/%m/%Y %I:%M:%S")
with open('AUC_history.csv','w') as f:
for key in history.history.keys():
f.write("%s,%s\n"%(key,history.history[key]))
epoch_times = scn.time_history.times
batch_loss = np.array(scn.loss_history.losses).ravel().tolist()
epoch_loss = np.array(scn.loss_history.val_losses).ravel().tolist()
date_time = str(datetime.now())
save_model(scn.model, 'goes16ci_model_cpu' + date_time + '.h5', save_format = 'h5')
return epoch_times, batch_loss, epoch_loss
def get_weights(save_dir: Path, model_name: str, dtype: str) -> str:
"""Download pre-trained imagenet weights for model.
Args:
save_dir: Path to where checkpoint must be downloaded.
model_name: Type of image classification model, must be one of
("GoogleNet", "InceptionV1", "MobileNet", "MobileNetV2", "NASNetMobile", "DenseNet121",
"ResNet50", "Xception", "InceptionV3") in all lower case.
dtype: Data type of the network.
Returns: Path to checkpoint file.
"""
if isinstance(save_dir, str):
save_dir = Path(save_dir)
g = tf.Graph()
with tf.Session(graph=g) as sess:
keras_backend.set_floatx(dtype)
keras_backend.set_session(sess)
if model_name == "mobilenet":
MobileNet(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "mobilenetv2":
MobileNetV2(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "nasnetmobile":
NASNetMobile(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "densenet121":
DenseNet121(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "resnet50":
ResNet50(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "xception":
Xception(weights='imagenet')
saver = tf.train.Saver()
elif model_name == "inceptionv3":
InceptionV3(weights='imagenet')
saver = tf.train.Saver()
elif model_name in ("googleNet", "inceptionv1"):
tar_file = get_file(
fname='inceptionv1_tar.gz',
origin=
'http://download.tensorflow.org/models/inception_v1_2016_08_28.tar.gz'
)
tar_file_reader = tarfile.open(tar_file)
tar_file_reader.extractall(save_dir)
if dtype == 'float16':
saver = convert_ckpt_to_fp16(
Path(save_dir, 'inception_v1.ckpt').as_posix())
sess.run(tf.global_variables_initializer())
else:
raise ValueError("""Requested model type = %s not one of
["GoogleNet", "InceptionV1", "MobileNet", "MobileNetV2", "NASNetMobile", "DenseNet121",
"ResNet50", "Xception", "InceptionV3"].""" % model_name)
save_dir.mkdir(parents=True, exist_ok=True)
return saver.save(sess,
Path(save_dir, f"{model_name}.ckpt").as_posix())
def use_floatx(x: str):
"""Context manager to temporarily
set the keras backend precision.
"""
_floatx = floatx()
set_floatx(x)
yield
set_floatx(_floatx)
def _set_precision(calculation_dtype, calculation_epsilon):
# enable single/half/double precision
K.set_floatx(calculation_dtype)
K.set_epsilon(calculation_epsilon)
# enable mixed precission
if "float16" in calculation_dtype:
mixed_precision.set_global_policy("mixed_float16")
def __init__(self, state_dim, action_dim, lr):
K.set_floatx('float32')
self.state_dim = state_dim
self.action_dim = action_dim
self.model = self.build_model()
self.target = self.build_model()
self.adam_optimizer = Adam(learning_rate=lr)
def main(use_mixed_precision=False,
training_batch_size=TRAINING_BATCH_SIZE,
generation_batch_size=generation_batch_size,
generator_optimizer=generator_optimizer,
discriminator_optimizer=discriminator_optimizer):
if use_mixed_precision:
print('Using Mixed Precision')
policy = mixed_precision.Policy('mixed_float16')
mixed_precision.set_policy(policy)
else:
policy = mixed_precision.Policy('float32')
mixed_precision.set_policy(policy)
epsilon = 1e-7
dtype = 'float32'
K.set_epsilon(epsilon)
K.set_floatx(dtype)
print(K.floatx(), K.epsilon(), training_batch_size)
print('Compute dtype: %s' % policy.compute_dtype)
print('Variable dtype: %s' % policy.variable_dtype)
tf.autograph.set_verbosity(0, False)
test_datagen = ImageDataGenerator(rescale=1. / 255)
# if start_from_scratch:
# initializeSnakeIdentifier(
# train_datagen,
# test_datagen
# )
snek_generator = createSnekMaker()
snek_discriminator = initializeSnakeIdentifier()
# snek_discriminator.predict([baby_noise, tf.constant([0]*32)])
gan = make_gan(snek_discriminator, snek_generator)
snek_discriminator.compile(optimizer=discriminator_optimizer,
loss='binary_crossentropy')
train_df = pd.read_csv('./classes_train.csv')
checkpoint_dir = './snek_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
print(checkpoint_prefix)
checkpoint = tf.train.Checkpoint(
generator_optimizer=generator_optimizer,
discriminator_optimizer=discriminator_optimizer,
snek_checker=snek_discriminator,
snek_generator=snek_generator,
gan=gan)
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
# checkpoint.save(file_prefix=checkpoint_prefix)
# trainSnekMaker(
# train_datagen,
# check_model=snek_checker,
# gen_model=snek_generator,
# train_model=train_model
# )
train(train_df, EPOCHS, snek_generator, snek_discriminator, gan,
checkpoint, checkpoint_prefix)
def __init__(self, rows, cols, channels, classes, latent, tpu=False):
if tpu:
set_floatx('float16')
set_epsilon(1e-4)
# Input shape
self.img_rows = rows
self.img_cols = cols
self.channels = channels
self.img_shape = (self.img_rows, self.img_cols, self.channels)
self.num_of_classes = classes
# size of the vector to fid the generator (z)
self.latent_dim = latent
#optimizer = Adam(0.0002, 0.5)
optimizer = tf.train.AdamOptimizer(0.0002, 0.5)
loss_scale_manager = FixedLossScaleManager(5000)
loss_scale_optimizer = LossScaleOptimizer(optimizer,
loss_scale_manager)
losses = ['binary_crossentropy', 'sparse_categorical_crossentropy']
# Build and compile the discriminator
self.discriminator = self.build_discriminator()
self.discriminator.compile(loss=losses,
optimizer=loss_scale_optimizer,
metrics=['accuracy'])
# Build the generator
self.generator = self.build_generator()
# The generator takes noise and the target label as input
# and generates the corresponding digit of that label
noise = Input(shape=(self.latent_dim, ))
label = Input(shape=(1, ))
img = self.generator([noise, label])
# For the combined model we will only train the generator
self.discriminator.trainable = False
# The discriminator takes generated image as input and determines validity
# and the label of that image
valid, target_label = self.discriminator(img)
# The combined model (stacked generator and discriminator)
# Trains the generator to fool the discriminator
self.combined = Model([noise, label], [valid, target_label])
self.combined.compile(loss=losses, optimizer=loss_scale_optimizer)
def _set_precision(calculation_dtype, calculation_epsilon):
# enable single/half/double precision
import tensorflow.keras.backend as K
K.set_floatx(calculation_dtype)
K.set_epsilon(calculation_epsilon)
# enable mixed precission
if "float16" in calculation_dtype:
import tensorflow.keras.mixed_precision as mixed_precision
policy = mixed_precision.Policy("mixed_float16")
mixed_precision.set_global_policy(policy)
def _model_tf_v1(n_input, n_output):
"""
Initialize the tensorflow 1.1X version of the model
:param int n_input: number of input dimensions (number of ECG + aux channels)
:param int n_output: number of output (number of EEG channels)
:return: initialized model
"""
from tensorflow.python.keras.layers import Input, Bidirectional, CuDNNGRU, Dense, Dropout
session_config = tf.ConfigProto(gpu_options=tf.GPUOptions(
allow_growth=True))
sess = tf.Session(config=session_config)
K.set_floatx('float64')
ecg_input = Input(shape=(None, n_input),
dtype='float64',
name='ecg_input')
x = Bidirectional(
CuDNNGRU(16,
return_sequences=True,
recurrent_regularizer=l2(0.096),
activity_regularizer=l2(0.030)))(ecg_input)
x = Bidirectional(
CuDNNGRU(16,
return_sequences=True,
recurrent_regularizer=l2(0.090),
activity_regularizer=l2(0.013)))(x)
x = Dense(8, activation='relu')(x)
x = Dropout(0.327)(x)
x = Bidirectional(
CuDNNGRU(16,
return_sequences=True,
recurrent_regularizer=l2(0.024),
activity_regularizer=l2(0.067)))(x)
x = Bidirectional(
CuDNNGRU(64,
return_sequences=True,
recurrent_regularizer=l2(2.48e-07),
activity_regularizer=l2(0.055)))(x)
bcg_out = Dense(n_output, activation='linear')(x)
model = Model(inputs=ecg_input, outputs=bcg_out)
return model
def __init__(self, save_path=None):
K.set_floatx('float32')
if save_path is not None:
self.save_path = save_path
self.img_rows = 256
self.img_cols = 256
self.channels = 1
self.AUTOTUNE = tf.data.experimental.AUTOTUNE
self.optimizer = 'Adam'
self.loss = 'binary_crossentropy'
self.metrics = [tf.keras.metrics.BinaryAccuracy()]
def main():
# parse args --------------------------------------------------------------
args = _parse_args()
# set device --------------------------------------------------------------
os.environ['CUDA_VISIBLE_DEVICES'] = args.devices
# set learning phase ------------------------------------------------------
K.set_learning_phase(0)
# set data format ---------------------------------------------------------
if args.devices == '' or args.model == 'mobilenet_v2':
# note: tensorflow supports b01c pooling on cpu only
K.set_image_data_format('channels_last')
else:
K.set_image_data_format('channels_first')
# set dtype ---------------------------------------------------------------
K.set_floatx(args.dtype)
# load model --------------------------------------------------------------
model_module = globals()[args.model]
model_kwargs = {}
if args.model == 'mobilenet_v2':
model_kwargs['alpha'] = args.mobilenet_v2_alpha
model = model_module.get_model(input_type=args.input_type,
input_shape=(args.input_height,
args.input_width),
output_type=args.output_type,
n_classes=args.n_classes,
sampling=False,
**model_kwargs)
# create frozen graph
sess = K.get_session()
out_name = [out.op.name for out in model.outputs]
frozen_graph = _freeze_session(sess, output_names=out_name)
dirname = os.path.dirname(args.output_filepath)
filename = os.path.basename(args.output_filepath)
assert os.path.splitext(filename)[1] == '.pb'
tf.train.write_graph(frozen_graph, dirname, filename, as_text=False)
# tf.train.write_graph(frozen_graph, dirname, filename, as_text=False)
# store input and output names as json file
write_json(
args.output_filepath + '.json', {
'input_names': [input.op.name for input in model.inputs],
'output_names': [output.op.name for output in model.outputs]
})
def __init__(self,
data: Tuple[tf.Tensor, tf.Tensor],
m: int = 20,
d: int = 1,
alpha: np.float = 1./np.sqrt(2.),
eps_sq: np.float = 1,
sigma_n_sq: np.float = 1,
sigma_f_sq: np.float = 1,
dir_weights: str = None):
if data[1].dtype == np.float64:
K_bd.set_floatx('float64')
else:
set_default_float(np.float32)
self.num_data = tf.cast(data[1].shape[0], default_float())
self.data = (tf.cast(data[0], default_float()), tf.cast(data[1], default_float()))
self.const = tf.cast(0.5*data[1].size*np.log(2*np.pi), default_float())
self.flag_1d = d == 1
self.alpha = tf.cast(alpha, default_float())
self.alpha_sq = tf.square(self.alpha)
self.m = tf.cast(m, default_float())
self.this_range = tf.constant(np.asarray(list(product(range(1, m + 1), repeat=d))).squeeze(), dtype=default_float())
self.this_range_1 = self.this_range - 1.
self.this_range_1_2 = self.this_range_1 if self.flag_1d else tf.range(m, dtype=default_float())
self.this_range_1_int = tf.cast(self.this_range_1, tf.int32)
self.tf_range_dnn_out = tf.range(d)
self.this_range_1_ln2 = np.log(2.)*self.this_range_1
self.vander_range = tf.range(m+1, dtype=default_float())
self.eye_k = tf.eye(m**d, dtype=default_float())
self.yTy = tf.reduce_sum(tf.math.square(self.data[1]))
self.coeff_n_tf = tf.constant(np.load(os.path.dirname(os.path.realpath(__file__)) + '/hermite_coeff.npy')[:m, :m], dtype=default_float())
eps_sq = eps_sq*np.ones(d) if d > 1 else eps_sq
self.eps_sq = Parameter(eps_sq, transform=positive(), dtype=default_float())
self.sigma_f_sq = Parameter(sigma_f_sq, transform=positive(), dtype=default_float())
self.sigma_n_sq = Parameter(sigma_n_sq, transform=positive(), dtype=default_float())
model = models.Sequential()
model.add(layers.Dense(512, activation='tanh', input_dim=data[0].shape[1]))
model.add(layers.Dense(256, activation='tanh'))
model.add(layers.Dense(64, activation='tanh'))
model.add(layers.Dense(d))
if dir_weights is not None:
model.load_weights(dir_weights)
self.neural_net = model
def __init__(self, content: np.ndarray, style: np.ndarray):
tf.compat.v1.disable_eager_execution()
K.set_floatx('float64')
self.content = content
self.style = style
print(" Building transfer model.")
self.contentTensor = K.variable(self.content)
self.styleTensor = K.variable(self.style)
self.genTensor = K.placeholder((1, CONTENT_IMG_H, CONTENT_IMG_W, 3))
self.inputTensor = K.concatenate(
[self.contentTensor, self.styleTensor, self.genTensor], axis=0)
self.model = vgg19.VGG19(include_top=False,
input_tensor=self.inputTensor)
self.totalLoss = self.constructTotalLoss()
self.gradient = self.constructGradient()
self.kerasFunction = self.constructKerasFunction()
self.runOutput = None
def __init__(self,
field_size=32,
name="DenseNN",
loss=kls.MeanSquaredError()):
Inputshape = (field_size, field_size, 2)
K.set_floatx('float32')
model = km.Sequential(name=name)
model.add(
kl.Flatten(input_shape=Inputshape, data_format="channels_last"))
model.add(kl.Dense(32, activation='sigmoid', name="Dense1"))
model.add(kl.Dense(32, activation='sigmoid', name="Dense2"))
model.add(kl.Dense(1, activation='linear', name="Prediction"))
model.compile(optimizer=ko.Adam(), loss=loss)
super().__init__(model, loss=loss)
def train_bertffn(model_path='models/bertffn_crossentropy/bertffn',
data_path='data/mqa_csv',
num_epochs=20,
num_gpu=1,
batch_size=64,
learning_rate=2e-5,
validation_split=0.2,
loss='categorical_crossentropy',
pretrained_path='pubmed_pmc_470k/',
max_seq_len=256):
tf.compat.v1.disable_eager_execution()
if loss == 'categorical_crossentropy':
loss_fn = qa_pair_cross_entropy_loss
else:
loss_fn = qa_pair_loss
K.set_floatx('float32')
tokenizer = FullTokenizer(os.path.join(pretrained_path, 'vocab.txt'))
d = create_dataset_for_bert(data_path,
tokenizer=tokenizer,
batch_size=batch_size,
shuffle_buffer=500000,
dynamic_padding=True,
max_seq_length=max_seq_len)
eval_d = create_dataset_for_bert(data_path,
tokenizer=tokenizer,
batch_size=batch_size,
mode='eval',
dynamic_padding=True,
max_seq_length=max_seq_len,
bucket_batch_sizes=[64, 64, 64])
medical_qa_model = MedicalQAModelwithBert(
config_file=os.path.join(pretrained_path, 'bert_config.json'),
checkpoint_file=os.path.join(pretrained_path, 'biobert_model.ckpt'))
optimizer = tf.keras.optimizers.Adam(lr=learning_rate)
medical_qa_model.compile(optimizer=optimizer,
loss=loss_fn,
metrics=[qa_pair_batch_accuracy])
epochs = num_epochs
loss_metric = tf.keras.metrics.Mean()
medical_qa_model.fit(d, epochs=epochs)
medical_qa_model.summary()
medical_qa_model.save_weights(model_path)
medical_qa_model.evaluate(eval_d)
def set_precision(precision):
if precision == 'float16':
dtype = 'float16'
K.set_floatx(dtype)
# default is 1e-7 which is too small for float16. Without adjusting the epsilon, we will get NaN predictions because of divide by zero problems
K.set_epsilon(1e-4)
print_debug('Compute dtype: %s' % 'float16')
print_debug('Variable dtype: %s' % 'float16')
elif precision == 'mixed':
policy = mixed_precision.Policy('mixed_float16')
mixed_precision.set_policy(policy)
print_debug('Compute dtype: %s' % policy.compute_dtype)
print_debug('Variable dtype: %s' % policy.variable_dtype)
else:
policy = mixed_precision.Policy('float32')
mixed_precision.set_policy(policy)
print_debug('Compute dtype: %s' % policy.compute_dtype)
print_debug('Variable dtype: %s' % policy.variable_dtype)
def __init__(self,
keras_input_layer,
alpha=1,
stddev=1.0,
use_float64_ops=False,
**kwargs):
super(RBMBase, self).__init__(keras_input_layer, **kwargs)
self.use_float64_ops = use_float64_ops
self.layers_dtype = tensorflow.complex128 if self.use_float64_ops else tensorflow.complex64
self.alpha = alpha
if self.use_float64_ops:
K.set_floatx('float64')
x = ToComplex128()(self.keras_input_layer)
else:
x = ToComplex64()(self.keras_input_layer)
self.initializer = tensorflow.random_normal_initializer(
stddev=stddev, dtype=self.layers_dtype.real_dtype)
self._predictions = self.build_predictions(x)
def __init__(self, image_paths, labels, test_x, test_y, save_path=None):
K.set_floatx('float32')
self.x = image_paths
self.y = labels
self.xt = test_x
self.yt = test_y
if save_path is not None:
self.save_path = save_path
self.img_rows = 256
self.img_cols = 256
self.channels = 1
self.AUTOTUNE = tf.data.experimental.AUTOTUNE
self.optimizer = 'Adam'
self.loss = 'mse'
self.metrics = ['mse']
def build_model(self, input_layer_size):
K.set_floatx('float32')
# sess_gpu = tf.Session(config=tf.ConfigProto(log_device_placement=True))
print(tf.test.is_gpu_available())
from tensorflow.python.client import device_lib
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
adam = Adam(learning_rate=0.000007)
model = Sequential()
model.add(Dense(input_layer_size, kernel_initializer='uniform', activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(int(input_layer_size / 2), kernel_initializer='uniform', activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(64, kernel_initializer='uniform', activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(32, kernel_initializer='uniform', activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(8, kernel_initializer='uniform', activation='relu'))
model.add(Dense(1, kernel_initializer='uniform', activation='sigmoid'))
# model.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
model.compile(loss='binary_crossentropy', optimizer=adam, metrics=['accuracy'])
self.model = model
X_train, X_test, Y_train, Y_test = self.get_train_test_data(self.X, self.Y)
history = self.model.fit(
X_train, Y_train, epochs=20, batch_size=64
, validation_data=(X_test, Y_test)
)
self.save_model()
print(X_test[0])
y_pred = self.model.predict(X_test)
print(history.history)
print("classification_report\n", classification_report(Y_test, y_pred.round()))
print("accuracy_score\n", accuracy_score(Y_test, y_pred.round()))
# print(confusion_matrix(X_test, Y_test))
self.model.evaluate(X_test, Y_test, verbose=1)
def __init__(self,
n_input=1,
n_output=63,
lr=1e-3,
opt_type='adam',
**kwargs):
"""
Constructor for RNN model
:param int n_input: number of input dimensions (number of ECG + aux channels)
:param int n_output: number of output (number of EEG channels)
:param float lr: learning rate
:param str opt_type: chosen type of optimizer, allowed to be adam, rmsprop or sgd
:param kwargs: clipnorm: normalized value for gradient clipping
clipvalue: numerical value for gradient clipping
and other Keras optimizier parameters for the chosen optimizer
:return: initialized model object
"""
super().__init__()
self.name = 'default_rnn_model'
self.lr = lr
self.opt_type = opt_type.lower()
self.n_input = n_input
self.n_output = n_output
K.set_floatx('float64')
if opt_type.lower() == 'adam':
self.optimizer = optimizers.Adam(lr=lr, **kwargs)
elif opt_type.lower() == 'rmsprop':
self.optimizer = optimizers.RMSprop(lr=lr, **kwargs)
elif opt_type.lower() == 'sgd':
self.optimizer = optimizers.SGD(lr=lr, **kwargs)
else:
raise NotImplementedError
def make_gps_mlp(input_size, output_size, config):
global hidden_layer_nodes
K.set_floatx('float64')
input = Input(shape=(input_size, ))
x = Dense(units=hidden_layer_nodes, activation='relu')(input)
x = Dense(units=hidden_layer_nodes, activation='relu')(x)
mean = Dense(units=output_size, activation='linear')(x)
# mean = Dense(units=output_size, activation='tanh')(x)
# std = Dense(units=output_size, activation='softplus')(x)
# action_out = Lambda(action)([mean, std])
action_out = Lambda(gps_action)(mean)
model_act = Model(inputs=input, outputs=action_out)
if config['train']:
act_taken = Input(shape=(output_size, ))
guiding_prob = Input(shape=(1, ))
# log_prob_out = Lambda(log_prob)([mean, std, act_taken])
prob_out = Lambda(prob)([mean, act_taken])
# Model for full training
reg_weight = K.variable(1e-2, name='regularization_weight')
model_train = Model(inputs=[input, act_taken, guiding_prob],
outputs=prob_out)
adam1 = Adam(lr=config['lr'])
model_train.compile(loss=gps_loss(reg_weight, guiding_prob),
optimizer=adam1)
# Model for pretraining
model_pretrain = Model(inputs=[input, act_taken], outputs=prob_out)
# adam2 = Adam(lr=config['lr'])
adam2 = Adam(lr=0.001)
model_pretrain.compile(loss=binary_crossentropy, optimizer=adam2)
else:
model_train = None
model_pretrain = None
reg_weight = None
return model_act, model_train, model_pretrain, reg_weight
def train_conv_net_cpu(train_data,
train_labels,
val_data,
val_labels,
conv_net_hyperparameters,
num_processors,
seed,
dtype="float32",
inter_op_threads=2):
"""
Train a convolutional neural network on the CPU.
"""
np.random.seed(seed)
if "get_visible_devices" in dir(tf.config.experimental):
gpus = tf.config.experimental.get_visible_devices("GPU")
else:
gpus = tf.config.get_visible_devices("GPU")
if len(gpus) > 0:
for device in gpus:
tf.config.experimental.set_memory_growth(device, True)
tf.config.threading.set_inter_op_parallelism_threads(inter_op_threads)
tf.config.threading.set_intra_op_parallelism_threads(num_processors)
if tf.__version__[0] == "1":
tf.set_random_seed(seed)
else:
tf.random.set_seed(seed)
K.set_floatx(dtype)
with tf.device("/CPU:0"):
scn = ResNet(**conv_net_hyperparameters)
history = scn.fit(train_data,
train_labels,
val_x=val_data,
val_y=val_labels)
history
print(history)
epoch_times = scn.time_history.times
batch_loss = np.array(scn.loss_history.losses).ravel().tolist()
epoch_loss = np.array(scn.loss_history.val_losses).ravel().tolist()
return epoch_times, batch_loss, epoch_loss
def __init__(self, image_paths, labels, save_path=None):
K.set_floatx('float32')
self.image_paths = image_paths
self.labels = labels
if save_path is not None:
self.save_path = save_path
self.img_rows = 256
self.img_cols = 256
self.channels = 1
self.AUTOTUNE = tf.data.experimental.AUTOTUNE
self.optimizer = 'Adam'
self.loss = 'binary_crossentropy'
self.metrics = [
tf.keras.metrics.AUC(),
tf.keras.metrics.FalseNegatives(),
tf.keras.metrics.FalsePositives(),
tf.keras.metrics.TrueNegatives(),
tf.keras.metrics.TruePositives()
]
