def to(self, *args: Any, **kwargs: Any) -> 'AnalogContext':
"""Move analog tiles of the current context to a device.
Note:
Please be aware that moving analog tiles from GPU to CPU is
currently not supported.
Caution:
Other tensor conversions than moving the device to CUDA,
such as changing the data type are not supported for analog
tiles and will be simply ignored.
Returns:
This module in the specified device.
"""
# pylint: disable=invalid-name, attribute-defined-outside-init
self.data = self.data.to(*args, **kwargs)
device = None
if 'device' in kwargs:
device = kwargs['device']
elif len(args) > 0 and not isinstance(args[0], (Tensor, dtype)):
device = torch_device(args[0])
if device is not None:
device = torch_device(device)
if device.type == 'cuda' and not self.analog_tile.is_cuda:
self.analog_tile = self.analog_tile.cuda(device)
self.reset(self.analog_tile)
return self
def __init__(
self,
dataset: Dataset,
instance: Instance,
model: Model,
optimizer: Optimizer,
scheduler: Any,
epochs: int,
batch_size: int,
run_dir_path: Path,
eval_every: int,
limit_epochs_at: Optional[int],
train_eval_split: float,
metric_names: List[str],
selection_metric: str,
kept_checkpoints: int,
cuda_device: Optional[int],
) -> None:
self.dataset = dataset
self.instance = instance
self.optimizer = optimizer
self.scheduler = scheduler
self.epochs = epochs
self.batch_size = batch_size
self.run_dir_path = run_dir_path
self.eval_every = eval_every
self.train_eval_split = train_eval_split
self.limit_epochs_at = 10e1000 if limit_epochs_at is None else limit_epochs_at
self.selection_metric = selection_metric
self.kept_checkpoints = kept_checkpoints
self._checkpoints_stats: List[Tuple[float, str]] = []
if cuda_device is not None:
if not cuda_is_available():
raise RuntimeError("CUDA is not available on this system.")
self.use_cuda = True
self.device = torch_device("cuda:%d" % cuda_device)
else:
self.use_cuda = False
self.device = torch_device("cpu")
self.model = model.to(self.device)
self._checkpoints_dir = self.run_dir_path / "checkpoints"
self._writers: Dict[DataType, SummaryWriter] = {}
self._accumulated_metrics: Dict[DataType, Dict[
Metric,
List[float]]] = defaultdict(lambda: defaultdict(lambda: []))
self._metrics = [_metrics[metric_name] for metric_name in metric_names]
self._metric_names = {
metric: metric_name
for metric, metric_name in zip(self._metrics, metric_names)
}
self._epochs_size = len(str(epochs))
def __init__(
self,
out_size: int,
in_size: int,
rpu_config: RPUConfigGeneric,
bias: bool = True,
in_trans: bool = False,
out_trans: bool = False
):
self.out_size = out_size
self.in_size = in_size
self.rpu_config = rpu_config
self.bias = bias
self.in_trans = in_trans
self.out_trans = out_trans
# Only used for indexed.
self.image_sizes = [] # type: List[int]
x_size = in_size + 1 if self.bias else in_size
d_size = out_size
self.tile = self._create_simulator_tile(x_size, d_size, rpu_config)
self.tile.set_learning_rate(0.01)
self.tile.set_weights_uniform_random(-0.01, 0.01)
self.device = torch_device('cpu')
self.is_cuda = False
self.shared_weights = None # type: Optional[Tensor]
# create analog context
self.analog_ctx = AnalogContext(self)
def cuda(
self,
device: Optional[Union[torch_device, str,
int]] = None) -> 'BaseTile':
"""Return a copy of this tile in CUDA memory.
Args:
device: CUDA device
Returns:
Self with the underlying C++ tile moved to CUDA memory.
Raises:
CudaError: if the library has not been compiled with CUDA.
"""
if not cuda.is_compiled():
raise CudaError('aihwkit has not been compiled with CUDA support')
device = torch_device('cuda', cuda_device(device).idx)
if self.is_cuda and device != self.device:
raise CudaError(
'Cannot switch CUDA devices of existing Cuda tiles')
if isinstance(self.tile, tiles.AnalogTile):
with cuda_device(device):
self.tile = tiles.CudaAnalogTile(self.tile)
self.is_cuda = True
self.device = device
self.analog_ctx.cuda(device)
return self
def __init__(self,
model,
trainset_cls,
testset_cls=None,
test_apply=None,
batch_size=8,
lr=0.01,
clip_grad=0,
weight_decay=0,
optimizer='sgd',
criterion=nn.CrossEntropyLoss,
device='cuda',
dataset_caching=True):
assert isinstance(model, nn.Module)
self.trainset_cls = trainset_cls
self.testset_cls = testset_cls or trainset_cls
self.test_apply = test_apply
self.batch_size = batch_size
self.clip_grad = clip_grad
self.device = torch_device(device)
self.model = model
self.model.to(device)
parameters = filter(lambda p: p.requires_grad, self.model.parameters())
self.optimizer = OPTIMIZERS[optimizer](parameters,
lr=lr,
weight_decay=weight_decay)
self.criterion = criterion()
self._dataset_caching = dataset_caching
self._dataset_cache = {}
self._dataset_hash = {}
def to(
self,
device: Optional[Union[torch_device, str, int]] = None
) -> 'AnalogSequential':
"""Move and/or cast the parameters, buffers and analog tiles.
Note:
Please be aware that moving analog layers from GPU to CPU is
currently not supported.
Args:
device: the desired device of the parameters, buffers and analog
tiles in this module.
Returns:
This module in the specified device.
"""
# pylint: disable=arguments-differ
device = torch_device(device)
super().to(device)
if device.type == 'cuda':
self._apply_to_analog(lambda m: m.cuda(device))
elif device.type == 'cpu':
self._apply_to_analog(lambda m: m.cpu())
return self
def __init__(self,
out_size: int,
in_size: int,
rpu_config: RPUConfigGeneric,
bias: bool = True,
in_trans: bool = False,
out_trans: bool = False):
self.out_size = out_size
self.in_size = in_size
self.rpu_config = rpu_config
self.bias = bias
self.in_trans = in_trans
self.out_trans = out_trans
# Only used for indexed.
self.image_sizes = [] # type: List[int]
x_size = in_size + 1 if self.bias else in_size
d_size = out_size
# Cuda tiles are assumed to init `self.tile` manually.
if self.is_cuda:
self.tile = None # type: Union[tiles.FloatingPointTile, tiles.AnalogTile]
else:
self.tile = self._create_simulator_tile(x_size, d_size, rpu_config)
self.tile.set_learning_rate(0.01)
self.tile.set_weights_uniform_random(-0.01, 0.01)
self.device = torch_device('cpu')
def generate_images(model,
batch,
mask_descriptors,
num_samples=64,
temp=1.,
verbose=False):
"""Generates image completions based on the images in batch masked by the
masks in mask_descriptors. This will generate
batch.size(0) * len(mask_descriptors) * num_samples completions, i.e.
num_samples completions for every image and mask combination.
Parameters
----------
model : pixconcnn.models.pixel_constrained.PixelConstrained instance
batch : torch.Tensor
mask_descriptors : list of mask_descriptor
See utils.masks.MaskGenerator for allowed mask_descriptors.
num_samples : int
Number of samples to generate for a given image-mask combination.
temp : float
Temperature for sampling.
verbose : bool
If True prints progress information while generating images
"""
device = torch_device("cuda" if cuda_is_available() else "cpu")
model.to(device)
outputs = []
for i in range(batch.size(0)):
outputs_per_img = []
for j in range(len(mask_descriptors)):
if verbose:
print("Generating samples for image {} using mask {}".format(
i, mask_descriptors[j]))
# Get image and mask combination
img = batch[i:i + 1]
mask_generator = MaskGenerator(model.prior_net.img_size,
mask_descriptors[j])
mask = mask_generator.get_masks(1)
# Create conditional pixels which will be used to sample completions
cond_pixels = get_repeated_conditional_pixels(
img, mask, model.prior_net.num_colors, num_samples)
cond_pixels = cond_pixels.to(device)
samples, log_probs = model.sample(cond_pixels,
return_likelihood=True,
temp=temp)
outputs_per_img.append({
"orig_img": img,
"cond_pixels": cond_pixels,
"mask": mask,
"samples": samples,
"log_probs": log_probs
})
outputs.append(outputs_per_img)
return outputs
def test_enum_sobol_GPEI(self):
"""Tests Soboland GPEI instantiation through the Models enum."""
exp = get_branin_experiment()
# Check that factory generates a valid sobol modelbridge.
sobol = Models.SOBOL(search_space=exp.search_space)
self.assertIsInstance(sobol, RandomModelBridge)
for _ in range(5):
sobol_run = sobol.gen(n=1)
self.assertEqual(sobol_run._model_key, "Sobol")
exp.new_batch_trial().add_generator_run(sobol_run).run()
# Check that factory generates a valid GP+EI modelbridge.
exp.optimization_config = get_branin_optimization_config()
gpei = Models.GPEI(experiment=exp, data=exp.fetch_data())
self.assertIsInstance(gpei, TorchModelBridge)
self.assertEqual(gpei._model_key, "GPEI")
botorch_defaults = "ax.models.torch.botorch_defaults"
# Check that the callable kwargs and the torch kwargs were recorded.
self.assertEqual(
gpei._model_kwargs,
{
"acqf_constructor": {
"is_callable_as_path": True,
"value": f"{botorch_defaults}.get_NEI",
},
"acqf_optimizer": {
"is_callable_as_path": True,
"value": f"{botorch_defaults}.scipy_optimizer",
},
"model_constructor": {
"is_callable_as_path": True,
"value": f"{botorch_defaults}.get_and_fit_model",
},
"model_predictor": {
"is_callable_as_path": True,
"value": f"{botorch_defaults}.predict_from_model",
},
"refit_on_cv": False,
"refit_on_update": True,
"warm_start_refitting": True,
},
)
self.assertEqual(
gpei._bridge_kwargs,
{
"transform_configs": None,
"torch_dtype": torch_float64,
"torch_device": torch_device(type="cpu"),
"status_quo_name": None,
"status_quo_features": None,
"optimization_config": None,
"transforms": Cont_X_trans + Y_trans,
},
)
gpei = Models.GPEI(experiment=exp,
data=exp.fetch_data(),
search_space=exp.search_space)
self.assertIsInstance(gpei, TorchModelBridge)
def compute_pulse_response(analog_tile: BaseTile,
direction: ndarray,
use_forward: bool = False) -> ndarray:
"""Computes the pulse response of a given device configuration.
Args:
analog_tile: Base tile used for computing the weight traces
direction: numpy vector of directions to sequentially apply (-1 or 1)
use_forward: Whether to use the (noisy) forward pass to read out the weights
(otherwise returns exact weight value).
Returns:
An numpy array ``w_trace`` of dimensions ``len(direction) x out_size x in_size``
"""
out_size = analog_tile.out_size
in_size = analog_tile.in_size
if analog_tile.is_cuda:
device = torch_device('cuda')
else:
device = torch_device('cpu')
total_iters = len(direction)
w_trace = np.zeros((total_iters, out_size, in_size))
in_vector = -ones(1, in_size, device=device)
out_vector = ones(1, out_size, device=device)
in_eye = eye(in_size, device=device)
dir_tensor = from_numpy(direction).float().to(device)
for i in range(total_iters):
# Update the pulses.
analog_tile.update(in_vector, out_vector * dir_tensor[i])
if use_forward:
# Save weights by using the forward pass (to get the short-term read noise).
w_trace[i, :, :] = analog_tile.forward(
in_eye).detach().cpu().numpy().T
else:
# Noise free.
w_trace[
i, :, :] = analog_tile.get_weights()[0].detach().cpu().numpy()
return w_trace
def make_sequence_video(cfg):
with open(cfg) as fd:
data_specs = json.load(fd)
temp_size = data_specs['temp_size']
num_of_temp_features = data_specs['temp_features']
m = load_model(data_specs['model_path'],
DOTNetCNN.name(),
# ToNameNet.name(),
"model_params.json")
use_gpu = torch_cuda.is_available()
device = torch_device(torch_cuda.current_device()) if use_gpu else torch_device("cpu")
m.to(device)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=0,
shuffle=False
)
full_df = pd.concat([train_d, val_d, test_d], ignore_index=True)
info_path = data_specs["info_path"]
ds = PixelLevelDs(
full_df,
info_path=info_path,
add_polarity="neg" if num_of_temp_features > 1 else "",
)
# make_video(m, ds)
make_non_repeating_video(m, ds)
def __init__(self, source_tile: FloatingPointTile):
if not cuda.is_compiled():
raise CudaError('aihwkit has not been compiled with CUDA support')
# Create a new instance of the rpu config.
new_rpu_config = deepcopy(source_tile.rpu_config)
# Create the tile, replacing the simulator tile.
super().__init__(source_tile.out_size, source_tile.in_size, new_rpu_config,
source_tile.bias, source_tile.in_trans, source_tile.out_trans)
self.tile = tiles.CudaFloatingPointTile(source_tile.tile)
# Set the cuda properties
self.stream = current_stream()
self.device = torch_device(current_device())
def test_run_example_gpu(self):
"""Test running the example using a local runner."""
training_experiment = self.get_experiment()
local_runner = LocalRunner(device=torch_device('cuda'))
with patch('sys.stdout', new=StringIO()) as captured_stdout:
result = local_runner.run(training_experiment, max_elements_train=10)
# Asserts over stdout.
self.assertIn('Epoch: ', captured_stdout.getvalue())
# Asserts over the returned results.
self.assertEqual(len(result), 1)
self.assertEqual(result[0]['epoch'], 0)
self.assertIn('train_loss', result[0])
self.assertIn('accuracy', result[0])
def __init__(self,
out_size: int,
in_size: int,
resistive_device: Optional[BaseResistiveDevice] = None,
bias: bool = False,
in_trans: bool = False,
out_trans: bool = False):
if not cuda.is_compiled():
raise RuntimeError(
'aihwkit has not been compiled with CUDA support')
super().__init__(out_size, in_size, resistive_device, bias, in_trans,
out_trans)
self.tile = tiles.CudaAnalogTile(self.tile)
self.stream = current_stream()
self.device = torch_device(current_device())
def __init__(self, source_tile: AnalogTile):
if not cuda.is_compiled():
raise RuntimeError(
'aihwkit has not been compiled with CUDA support')
# Create a new instance of the resistive device.
new_resistive_device = deepcopy(source_tile.resistive_device)
# Create the tile, replacing the simulator tile.
super().__init__(source_tile.out_size, source_tile.in_size,
new_resistive_device, source_tile.bias,
source_tile.in_trans, source_tile.out_trans)
self.tile = tiles.CudaAnalogTile(source_tile.tile)
# Set the cuda properties
self.stream = current_stream()
self.device = torch_device(current_device())
def __init__(self, path):
self.config_path = os.path.join(path, 'config.json')
self.model_path = os.path.join(path, 'model_best.pth')
config = json.load(open(self.config_path))
checkpoint = torch_load(self.model_path, map_location='cpu')
m_name, sd, self.classes = _get_model_att(checkpoint)
model = vgg11_bn(self.classes, pretrained=False)
model.load_state_dict(checkpoint['state_dict'])
tsf = _get_transform(config)
# self.device = torch_device("cuda" if torch_cuda.is_available() else "cpu")
self.device = torch_device("cpu")
self.model = model.to(self.device)
self.model.eval()
self.transforms = tsf
def get_answer(sentence):
with open(
os.path.join(os.path.dirname(__file__), '..', 'data',
'intents.json'), 'r') as f:
intents = json.load(f)
FILE = os.path.join(os.path.dirname(__file__), '..', 'data', 'data.pth')
data = torch_load(FILE)
input_size = data['input_size']
hidden_size = data['hidden_size']
output_size = data['output_size']
all_words = data['all_words']
tags = data['tags']
model_state = data['model_state']
device = torch_device('cuda' if torch_cuda.is_available() else 'cpu')
model = NeuralNet(input_size, hidden_size, output_size).to(device)
model.load_state_dict(model_state)
model.eval()
sentence = tokenize(sentence)
X = bag_of_words(sentence, all_words)
X = X.reshape(1, X.shape[0])
X = torch_from_numpy(X)
output = model(X)
_, predicted = torch_max(output, dim=1)
tag = tags[predicted.item()]
answer = ''
probs = torch_softmax(output, dim=1)
prob = probs[0][predicted.item()]
if prob.item() > 0.75:
for intent in intents['intents']:
if tag == intent['tag']:
answer = random.choice(intent['responses'])
else:
answer = "I don't understand..."
return answer, prob
def predict_dataset(dataset, model, device, test_apply=None, batch_size=4):
loader = DataLoader(dataset,
batch_size=batch_size,
shuffle=False,
num_workers=0,
pin_memory=cuda.is_available())
y_pred = []
with test_mode(model):
for x in loader:
x = x.to(device)
if test_apply is not None:
out = test_apply(x, model)
else:
out = model(x)
y_pred.append(out.to(torch_device('cpu')).numpy())
y_pred = np.vstack(y_pred)
return y_pred
def __init__(self,
out_size: int,
in_size: int,
resistive_device: BaseResistiveDevice,
bias: bool = True,
in_trans: bool = False,
out_trans: bool = False,
_from_tile: Optional[Any] = None):
self.out_size = out_size
self.in_size = in_size
self.resistive_device = resistive_device
self.bias = bias
self.in_trans = in_trans
self.out_trans = out_trans
x_size = in_size + 1 if self.bias else in_size
d_size = out_size
if _from_tile is not None:
self.tile = _from_tile
else:
self.tile = self.resistive_device.create_tile(x_size, d_size)
self.device = torch_device('cpu')
def __init__(self,
out_size: int,
in_size: int,
resistive_device: BaseResistiveDevice,
bias: bool = True,
in_trans: bool = False,
out_trans: bool = False):
self.out_size = out_size
self.in_size = in_size
self.resistive_device = resistive_device
self.bias = bias
self.in_trans = in_trans
self.out_trans = out_trans
x_size = in_size + 1 if self.bias else in_size
d_size = out_size
# Cuda tiles are assumed to init `self.tile` manually.
if self.is_cuda:
self.tile = None # type: Union[tiles.FloatingPointTile, tiles.AnalogTile]
else:
self.tile = self.resistive_device.create_tile(x_size, d_size)
self.device = torch_device('cpu')
class GazeDetector:
"""
Detect a user's gaze to see if they are engaged
or not
"""
weights_path: str = os.path.abspath("machine_learning/models/weights/gazenet.pth")
coordinate_length: float = 200
device: torch_device = torch_device("cuda:0" if torch_gpu.is_available() else "cpu")
model: gaze_cnn.GazeNet = gaze_cnn.GazeNet(device=device)
model.load_state_dict(torch_load(weights_path, map_location=device))
model.eval()
def __init__(self, logger, base_64_image):
"""
Base 64 image from the student app
:param base_64_image: string from client in request
"""
self.logger = logger
self.opencv_image = self._read_base64(base_64_image)
self.face_detector = FaceDetector(device=self.device)
@staticmethod
def _read_base64(base64_string: str) -> cvtColor:
"""
Read base64 string into CV2
:param base64_string: base 64 string
:return: cv2 image
"""
byte_buffer: BytesIO = BytesIO()
byte_buffer.write(base64_decode(base64_string))
pil_img: Image = Image.open(byte_buffer)
return cvtColor(np.array(pil_img), COLOR_RGB2BGR)
@staticmethod
def _normalize_face(landmarks, frame) -> tuple:
"""
Normalize the face to a standard map
:param landmarks: landmarks on the face
:param frame: image frame
:return: the face, gaze origin and
"""
left_eye_coord: tuple = (0.70, 0.35)
lcenter: tuple = tuple([landmarks[0], landmarks[5]])
rcenter: tuple = tuple([landmarks[1], landmarks[6]])
gaze_origin: tuple = (int((lcenter[0] + rcenter[0]) / 2), int((lcenter[1] + rcenter[1]) / 2))
dY: float = rcenter[1] - lcenter[1]
dX: float = rcenter[0] - lcenter[0]
angle = np.degrees(np.arctan2(dY, dX)) - 180
right_eye_x = 1.0 - left_eye_coord[0]
dist: np.float32 = np.sqrt((dX ** 2) + (dY ** 2))
new_dist = (right_eye_x - left_eye_coord[0])
new_dist *= 112
scale: np.float32 = new_dist / dist
M = getRotationMatrix2D(gaze_origin, angle, scale)
tX = 112 * 0.5
tY = 112 * left_eye_coord[1]
M[0, 2] += (tX - gaze_origin[0])
M[1, 2] += (tY - gaze_origin[1])
face = warpAffine(frame, M, (112, 112), flags=INTER_CUBIC)
return face, gaze_origin, M
@staticmethod
def _get_vector(pitch_yaw: List) -> (np.float32, np.float32):
"""
Get the deltas in the x and y direction to measure gaze
:param pitch_yaw: gaze coords from CNN
:return: dx, dy
"""
dx: np.float32 = -GazeDetector.coordinate_length * np.sin(pitch_yaw[1])
dy: np.float32 = -GazeDetector.coordinate_length * np.sin(pitch_yaw[0]) # coords start in upper left corner
return dx, dy
@staticmethod
def _compute_angle(dx: float, dy: float) -> np.float32:
"""
Computes the angle in degrees of the vector defined
by dx and dy
:param dx: delta x for vector
:param dy: delta y for vector
:return: angle between standard position and the vector in degs
"""
return np.arctan2(dy, dx) * (180 / np.pi) # convert to degrees from radians
@staticmethod
def _compute_magnitude(dx: float, dy: float) -> float:
"""
Compute the magnitude of vector <dy, dx>
:param dx: delta x for vector
:param dy: delta y for vector
:return: magnitude of vector <dx, dy>
"""
return np.sqrt(dx ** 2 + dy ** 2) # find magnitude of vector
def predict_gaze(self) -> (float, float):
"""
Predict the Gaze trajectory as a vector using our GazeNet CNN
and return the angle and magnitude of the vector
:return: angle between delta x and delta y
"""
self.opencv_image = flip_image(self.opencv_image[:, :, ::-1], 1) # helps to rectify rectangular coordinates
faces, landmarks = self.face_detector.detect(Image.fromarray(self.opencv_image))
if len(faces) == 0:
self.logger.info("No faces found!")
return 0, 0, "NOT ENGAGED", 0
for face, landmark in zip(faces, landmarks):
if face[-1] > 0.98: # confidence check
normal_face, gaze_origin, _ = GazeDetector._normalize_face(landmark, self.opencv_image)
# Predict Gaze
with torch_no_grad():
gaze = GazeDetector.model.get_gaze(normal_face)
gaze = gaze[0].data.cpu()
#self.logger("Predicted on CNN: Gaze com puted to be " + str(gaze))
dx, dy = self._get_vector(gaze)
dy *= -1
#self.logger.info("Original Gaze Location: " + str(gaze_origin))
#self.logger.info("Delta X: " + str(dx) + "; Delta Y: " + str(dy))
computed_angle = self._compute_angle(dx, dy).cpu().numpy()
computed_magnitude = self._compute_magnitude(dx, dy).cpu().numpy()
classification, score = self.interpret_vectors(computed_angle, computed_magnitude)
#self.logger.info("Trying to save gaze computation")
# showComputedGaze(Image.fromarray(self.opencv_image), gaze_origin, gaze, computed_angle, computed_magnitude, score, classification)
return computed_angle, computed_magnitude, classification, score
else:
return 0, 0, "NOT ENGAGED", 0
def interpret_vectors(self, angle: np.float32, magnitude: np.float32):
"""
Interpret the Vector's Position and Magnitude for engagement/disengagement
:param angle: the angle of the gaze vector
:param magnitude: the magnitude of the gaze vector
:return: student state and confidence
"""
Rule = namedtuple('Rule', ['trigger_str', 'confidence', 'result', 'score'])
rules: List[Rule] = [ # in order of precedence
Rule('angle == -1.0 and magnitude == -1.0', 0.0, "NOT ENGAGED", 0),
Rule('magnitude <= 30.0 or 0 < angle <= 15', 0.95, "ENGAGED", 10),
Rule('magnitude <= 40.0', 0.95, "ENGAGED", 9),
Rule('magnitude <= 50.0', 0.60, "ENGAGED",8),
Rule('magnitude <= 60.0', 0.65, "ENGAGED", 7),
Rule('15 < angle < 50', 0.70, "SOMEWHAT ENGAGED", 6),
Rule('60 < magnitude <= 70', 0.65, "SOMEWHAT ENGAGED", 5),
Rule('70 < magnitude <= 120', 0.70, "NOT ENGAGED", 4),
Rule('120 < magnitude < 360', 0.70, "NOT ENGAGED", 3),
Rule('magnitude > 60.0', 0.70, "NOT ENGAGED", 2),
Rule('magnitude > 50.0', 0.85, "NOT ENGAGED", 1),
Rule('True', 0.0, "ENGAGED", 10) # default condition (base case-- should never be called)
]
for rule_no, rule in enumerate(rules):
if eval(rule.trigger_str):
self.logger.info("Triggered rule: " + rule.trigger_str + ": " + str(rule))
return rule.result, rule.score
def process_each_frequency(model_dirname, stft, frequency, using_cuda=True):
'''
Setter method on stft.
'''
is_using_cuda = using_cuda and torch_cuda_is_cuda_available()
my_device = torch_device('cuda:0' if is_using_cuda else 'cpu')
# 1. Instantiate Neural Network Model
model_params_fname = os_path_join(
os_path_join(model_dirname, 'k_' + str(frequency)), MODEL_PARAMS_FNAME)
model_save_fpath = os_path_join(model_dirname, 'k_' + str(frequency),
MODEL_SAVE_FNAME)
model = get_which_model_from_params_fname(model_params_fname)
model.load_state_dict(torch_load(os_path_join(
os_path_dirname(model_save_fpath), 'model.dat'),
map_location=my_device),
strict=True)
model.eval()
model = model.to(my_device)
if False:
model.printing = True
from lib.print_layer import PrintLayer
new_model_net = []
for layer in model.net:
new_model_net.append(layer)
new_model_net.append(PrintLayer(layer))
from torch.nn import Sequential
model.net = Sequential(*new_model_net)
# 2. Get X_test
LOGGER.debug('r3.process_each_frequency: stft.shape = {}'.format(
stft.shape))
aperture_data = stft[:, :, frequency] # or stft_frequency
# 2.1. normalize by L1 norm
aperture_data_norm = np_linalg_norm(aperture_data, ord=np_inf, axis=1)
aperture_data /= aperture_data_norm[:, np_newaxis]
# load into torch and onto gpu
aperture_dataset_eval = ApertureDatasetEval(aperture_data)
aperture_dataset_loader = DataLoader(aperture_dataset_eval,
batch_size=EVAL_BATCH_SIZE,
shuffle=False,
num_workers=DATALOADER_NUM_WORKERS,
pin_memory=using_cuda)
# 3. Predict
if is_using_cuda is True:
torch_cuda_empty_cache()
aperture_data_new = predict(model, aperture_dataset_loader, my_device)
del aperture_data, model, aperture_dataset_eval, aperture_dataset_loader, my_device
if is_using_cuda is True:
torch_cuda_empty_cache()
# 4. Postprocess on y_hat
# rescale the data and store new data in stft
stft[:, :,
frequency] = aperture_data_new * aperture_data_norm[:, np_newaxis]
del aperture_data_new, aperture_data_norm
if config.color:
config.tau_size = 3
else:
config.tau_size = 2
config.eof_size = 15
config.aux_size = 2
config.out_size = 7
else:
config.eof_size = 15
config.tau_size = 3
config.aux_size = 6
config.out_size = 6
if use_cuda():
config.device = torch_device(args.device)
else:
config.device = torch_device('cpu')
# Compilation params
config.dest_dir = config.data_file
config.test_cases = ()
config.split_percen = .99
config.simulation = config.sim
config.max_traj = args.num_traj
# Data gathering params
config.normalize = False
config.num_traj = args.num_traj
config.framerate = args.framerate
config.save_folder = config.root_dir
def train(
*,
instance_file: str,
tensors_dir: str,
train_dir: str,
configs_dir: str,
model_encoder_iterations: int,
model_encoder_output_dim: int,
model_encoder_message_dim: int,
model_decoder_type: str,
model_learning_rate: float,
model_batch_size: int,
trainer_epochs: int,
trainer_eval_every: int,
trainer_limit_epochs_at: Optional[int],
trainer_train_eval_split: float,
trainer_selection_metric: str,
trainer_kept_checkpoints: int,
trainer_cuda: Optional[int],
log_level: str,
) -> None:
"""Run the training."""
Config.from_arguments(locals(),
["instance_file", "tensors_dir", "train_dir"],
"configs_dir").save(
Path(configs_dir) / "train.json")
logger = setup_logging(__name__, log_level)
tensors_dir_path = Path(tensors_dir).expanduser().resolve()
train_dir_path = Path(train_dir).expanduser().resolve()
train_dir_path.mkdir(parents=True, exist_ok=True)
with bz2_open(instance_file, "rb") as fh:
instance = pickle_load(fh)
dataset = CodRepDataset(input_dir=tensors_dir_path)
logger.info("Dataset of size %d", len(dataset))
train_length = round(0.9 * len(dataset))
eval_length = round(0.05 * len(dataset))
test_length = len(dataset) - train_length - eval_length
train_dataset, eval_dataset, test_dataset = random_split(
dataset, [train_length, eval_length, test_length])
if trainer_cuda is not None:
if not cuda_is_available():
raise RuntimeError("CUDA is not available on this system.")
device = torch_device("cuda:%d" % trainer_cuda)
else:
device = torch_device("cpu")
model = build_model(
instance=instance,
model_encoder_iterations=model_encoder_iterations,
model_encoder_output_dim=model_encoder_output_dim,
model_encoder_message_dim=model_encoder_message_dim,
model_decoder_type=model_decoder_type,
model_learning_rate=model_learning_rate,
model_batch_size=model_batch_size,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
test_dataset=test_dataset,
)
# The model needs a forward to be completely initialized.
model.training_step(instance.collate([dataset[0]]), 0)
logger.info("Configured model %s", model)
checkpoint_callback = ModelCheckpoint(
filepath=train_dir,
save_best_only=True,
verbose=True,
monitor="eval_mrr",
mode="max",
prefix="",
)
trainer = Trainer(default_save_path=train_dir,
checkpoint_callback=checkpoint_callback)
trainer.fit(model)
# this dictionary contains the basic parameters for training the model
custom_config = dict(dim=1000,
margin=10,
batch_size=1000,
learning_rate=0.2,
num_epochs=200,
corruption_factor=5,
filter_validation_triples=False,
early_stopping=True,
device="cuda:1")
config = BASE_CONFIG.copy()
config.update(custom_config)
config['model'] = list(MODELS)[model_type_train]
if cuda.is_available() and config['device'] != "cpu":
device = torch_device(config['device'])
else:
config['device'] = "cpu"
device = torch_device("cpu")
print("Using {}".format(config['device']))
if SEED:
manual_seed(0) # pytorch seed
seed(0) # numpy seed
if TRAIN_MODEL:
start = time.time()
run_train(config=config,
device=device,
dataset=dataset,
align_dataset=align_data_train,
N_si = 3.48 # silicon
eps_si = N_si**2
N_sio2 = 1.44 # silicon oxide
eps_sio2 = N_sio2**2
N_sinitride = 1.99 # silicon nitride
eps_sinitride = N_sinitride**2
# Physical constants ===================================================================================================
OMEGA_1550 = 1.215e15 # 1550nm frequency
MU0 = 4 * pi * 10**-7 # vacuum permeability
EPSILON0 = 8.854187817620e-12 # vacuum permittivity
C = 299792458.0 # speed of light
BUFFER_PERMITTIVITY = -1e20 # near infinite permittivity for cavity boundaries
# BUFFER_PERMITTIVITY = 1.0 # vacuum permittivity if using PML
# Design space size ====================================================================================================
DEVICE_LENGTH = 64 # length of permittivity region
DEVICE_LENGTH_2D = 32
NPML = 0 # number of PMLs
BUFFER_LENGTH = 4 # buffer size before NPML (reflective boundary if using cavity)
SCALE = 1e-15
L0 = 1e-6
dL = 0.05
PIXEL_SIZE = dL * L0
# Device configuration =================================================================================================
device = torch_device('cuda:0')
# device = torch_device('cpu')
def __setstate__(self, state: Dict) -> None:
"""Set the state after unpickling.
This method recreates the ``tile`` member, creating a new one from
scratch, as the binding Tiles are not serializable.
Caution:
RPU configs are overwritten by loading the state.
Raises:
TileError: if tile class does not match or hidden parameters do not match
"""
# pylint: disable=too-many-locals
# Note: self here is NOT initialized! So we need to recreate
# attributes that were not saved in getstate
current_dict = state.copy()
weights = current_dict.pop('analog_tile_weights')
hidden_parameters = current_dict.pop('analog_tile_hidden_parameters')
hidden_parameters_names = current_dict.pop(
'analog_tile_hidden_parameter_names', [])
alpha_scale = current_dict.pop('analog_alpha_scale', None)
tile_class = current_dict.pop('analog_tile_class',
self.__class__.__name__)
analog_lr = current_dict.pop('analog_lr', 0.01)
analog_ctx = current_dict.pop('analog_ctx')
shared_weights = current_dict.pop('shared_weights')
shared_weights_if = shared_weights is not None
self.__dict__.update(current_dict)
self.device = torch_device('cpu')
self.is_cuda = False
# get the current map location from analog_ctx (which is restored)
to_device = analog_ctx.device
# recreate attributes not saved
# always first create on CPU
x_size = self.in_size + 1 if self.bias else self.in_size
d_size = self.out_size
# Recreate the tile.
# Check for tile mismatch
if tile_class != self.__class__.__name__:
raise TileError(
'Mismatch of tile class: {} versus {}. Can only load analog '
'state from the same tile class.'.format(
self.__class__.__name__, tile_class))
self.tile = self._create_simulator_tile(x_size, d_size,
self.rpu_config)
names = self.tile.get_hidden_parameter_names()
if len(hidden_parameters_names
) > 0 and names != hidden_parameters_names:
# Check whether names match
raise TileError('Mismatch with loaded analog state: '
'Hidden parameter structure is unexpected.')
self.tile.set_hidden_parameters(Tensor(hidden_parameters))
self.tile.set_weights(weights)
self.tile.set_learning_rate(analog_lr)
# re-generate shared weights (CPU)
if shared_weights_if:
if not hasattr(self, 'shared_weights'):
# this is needed when pkl loading
self.shared_weights = shared_weights
with no_grad():
# always new will be populated with set weights.
self.shared_weights.data = zeros(d_size,
x_size,
requires_grad=True)
self.ensure_shared_weights()
else:
self.shared_weights = None
# Regenerate context but keep the object ID
if not hasattr(self, 'analog_ctx'): # when loading
self.analog_ctx = AnalogContext(self, parameter=analog_ctx)
self.analog_ctx.reset(self)
self.analog_ctx.set_data(analog_ctx.data)
if to_device.type.startswith('cuda'):
self.cuda(to_device)
if alpha_scale is not None:
# legacy. We apply the alpha scale instaed of the
# out_scaling_alpha when loading. The alpha_scale
# mechansim is now replaced with the out scaling factors
#
# Caution: will overwrite the loaded out_scaling_alphas
# if they would exist also (should not be for old checkpoints)
self.set_out_scaling_alpha(alpha_scale)
def _get_device(self):
return torch_device("cuda" if self.is_cuda else "cpu")
def main():
temp_size = 5
future_size = 0
net_temp_output = future_size + 1
out_dims = (260, 346)
batch_size = 12
num_classes = 3
# use_gpu = torch_cuda.is_available()
use_gpu = False # pytorch is giving a memory error
device = torch_device(torch_cuda.current_device()) if use_gpu else torch_device("cpu")
with open("cfg.txt") as fd:
data_specs = json.load(fd)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=future_size
)
info_path = data_specs["info_path"]
train_ds = PixelLevelDs(
train_d,
info_path=info_path
)
val_ds = PixelLevelDs(
val_d,
# add_polarity="neg",
info_path=info_path
)
test_ds = PixelLevelDs(
test_d,
info_path=info_path
)
dls = {
"train": DataLoader(train_ds, batch_size=batch_size, num_workers=0, pin_memory=use_gpu),
"val": DataLoader(val_ds, batch_size=batch_size, num_workers=0, pin_memory=use_gpu),
"test": DataLoader(test_ds, batch_size=1)
}
full_forward = SaveStatsForwardCallback(heatmap_CE_accuracy(num_classes))
# to_device = ToDeviceCallback(device)
compose_callbacks = ComposeCallback(
start_callbacks=[StartPrinterCallback()],
train_pData_callbacks=[
# to_device,
full_forward],
val_pData_callbacks=[
# to_device,
full_forward],
nextEpoch_callbacks=[full_forward],
phaseEnded_callbacks=[full_forward],
allEnded_callbacks=[]
)
# TODO: read <A Closer Look at Spatiotemporal Convolutions for Action Recognition>
# https://arxiv.org/pdf/1711.11248.pdf
model = ToNameNet(temp_size, net_temp_output, out_dims, num_classes)
model.to(device)
train(model, dls['train'], dls['val'], compose_callbacks,
# criterion=heatmap_mse(num_classes),
criterion=heatmap_cross_entropy(num_classes, weight=tensor([0.3, 4, 15], dtype=torch_float)),
epochs=1)
do_3state = True
if model_type == '2state':
do_3state = False
elif model_type == '3state':
do_3state = True
else:
print('Provide a model type (2state or 3state) as an argument')
# flag that indicates on which state the trajectory is running
flag_es = 3
# flags for hopping
do_hop = True
hop = False
skip_next = False
# SchNet calculator
model_s = load_model(path1, map_location=torch_device('cpu'))
model2_s = load_model(path2, map_location=torch_device('cpu'))
if do_3state:
model3_s = load_model(path3, map_location=torch_device('cpu'))
model = model_s.double()
model2 = model2_s.double()
if do_3state:
model3 = model3_s.double()
# demon-nano calculator
input_arguments = {
'DFTB': 'SCC LRESP EXST=2',
'CHARGE': '0.0',
'PARAM': 'PTYPE=BIO'
}