def calculate_loss(seg, recon, depth, normals):
((seg_pred, seg_target),
(recon_pred, recon_target),
(depth_pred, depth_target),
(normals_pred, normals_target)) = (seg, recon, depth, normals)
seg_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(seg_pred, seg_target)
ae_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(recon_pred, recon_target)
normals_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(normals_pred, normals_target)
depth_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(depth_pred, depth_target)
pred_soft = torch.sigmoid(seg_pred)
dice = dice_loss(pred_soft, seg_target, epsilon=1)
jaccard = jaccard_loss(pred_soft, seg_target, epsilon=1)
terms = NOD.dict_of(dice,
jaccard,
ae_bce_loss,
seg_bce_loss,
depth_bce_loss,
normals_bce_loss
)
term_weight = 1 / len(terms)
weighted_terms = [term.mean() * term_weight for term in terms.as_list()]
loss = sum(weighted_terms)
return NOD.dict_of(loss, terms)
def convert_to_coco_api(ds):
"""
:param ds:
:type ds:
:return:
:rtype:
"""
coco_ds = COCO()
ann_id = 0
dataset = {"images":[], "categories":[], "annotations":[]}
categories = set()
for img_idx in range(len(ds)):
# find better way to get target
# targets = ds.get_annotations(img_idx)
img, targets = ds[img_idx]
image_id = targets["image_id"].item()
dataset["images"].append(
{"id":image_id, "height":img.shape[-2], "width":img.shape[-1]}
)
bboxes = targets["boxes"]
bboxes[:, 2:] -= bboxes[:, :2]
bboxes = bboxes.tolist()
labels = targets["labels"].tolist()
areas = targets["area"].tolist()
iscrowd = targets["iscrowd"].tolist()
masks = None
keypoints = None
if "masks" in targets:
masks = targets["masks"]
# make masks Fortran contiguous for coco_mask
masks = masks.permute(0, 2, 1).contiguous().permute(0, 2, 1)
if "keypoints" in targets:
keypoints = targets["keypoints"]
keypoints = keypoints.reshape(keypoints.shape[0], -1).tolist()
num_objs = len(bboxes)
for i in range(num_objs):
ann = NOD(
image_id=image_id,
bbox=bboxes[i],
category_id=labels[i],
area=areas[i],
iscrowd=iscrowd[i],
id=ann_id,
segmentation=None,
keypoints=None,
num_keypoints=None,
)
categories.add(labels[i])
if "masks" in targets:
ann.segmentation = coco_mask.encode(masks[i].numpy())
if "keypoints" in targets:
ann.keypoints = keypoints[i]
ann.num_keypoints = sum(k != 0 for k in keypoints[i][2::3])
dataset["annotations"].append(ann)
ann_id += 1
dataset["categories"] = [{"id":i} for i in sorted(categories)]
coco_ds.dataset = dataset
coco_ds.createIndex()
return coco_ds
def session_factory(
agent: Type[AgentType] = None,
config=None,
*,
session: Union[Type[EnvironmentSessionType], EnvironmentSession],
save: bool = True,
has_x_server: bool = True,
skip_confirmation: bool = True,
**kwargs,
):
r"""
Entry point start a starting a training session with the functionality of parsing cmdline arguments and
confirming configuration to use before training and overwriting of default training configurations
"""
if config is None:
config = {}
if isinstance(config, dict):
config = NOD(**config)
else:
config = NOD(config.__dict__)
if has_x_server:
display_env = getenv("DISPLAY", None)
if display_env is None:
config.RENDER_ENVIRONMENT = False
has_x_server = False
config_mapping = config_to_mapping(config)
config_mapping.update(**kwargs)
config_mapping.update(save=save, has_x_server=has_x_server)
if not skip_confirmation:
sprint(f"\nUsing config: {config}\n", highlight=True, color="yellow")
for key, arg in config_mapping:
print(f"{key} = {arg}")
input("\nPress Enter to begin... ")
if session is None:
raise NoProcedure
elif inspect.isclass(session):
session = session(**config_mapping) # Use passed config arguments
elif isinstance(session, GDKC):
session = session(
**kwargs
) # Assume some kw parameters is set prior to passing session, only override with explicit overrides
try:
session(agent, **config_mapping)
except KeyboardInterrupt:
print("Stopping")
torch.cuda.empty_cache()
exit(0)
def main(is_user: bool = False):
""" """
global LOG_WRITER
if is_user:
LOG_WRITER = LogWriter(
ensure_existence(PROJECT_APP_PATH.user_log)
/ f"{PROJECT_NAME}_publisher.log"
)
else:
LOG_WRITER = LogWriter(
ensure_existence(PROJECT_APP_PATH.site_log)
/ f"{PROJECT_NAME}_publisher.log"
)
LOG_WRITER.open()
client = mqtt.Client()
client.on_publish = on_publish
client.on_disconnect = on_disconnect
HEIMDALLR_SETTINGS = HeimdallrSettings() # TODO input scope
client.username_pw_set(
HEIMDALLR_SETTINGS.mqtt_username, HEIMDALLR_SETTINGS.mqtt_password
)
try:
client.connect(
HEIMDALLR_SETTINGS.mqtt_broker, HEIMDALLR_SETTINGS.mqtt_port, keepalive=60
)
except ValueError as ve:
raise ValueError(
f"{HEIMDALLR_SETTINGS._mqtt_settings_path},"
f"{HEIMDALLR_SETTINGS.mqtt_broker},"
f"{HEIMDALLR_SETTINGS.mqtt_port},"
f"{ve}"
)
client.loop_start()
sensor_data = NOD({HOSTNAME: pull_gpu_info()})
next_reading = time.time()
with IgnoreInterruptSignal():
print("Publisher started")
for _ in busy_indicator():
sensor_data[HOSTNAME] = pull_gpu_info()
s = sensor_data.as_dict()
s = json.dumps(s)
client.publish(ALL_CONSTANTS.MQTT_TOPIC, s, ALL_CONSTANTS.MQTT_QOS)
next_reading += ALL_CONSTANTS.MQTT_PUBLISH_INTERVAL_SEC
sleep_time = next_reading - time.time()
if sleep_time > 0:
time.sleep(sleep_time)
# noinspection PyUnreachableCode
LOG_WRITER.close()
client.loop_stop()
client.disconnect()
def sample_transitions():
signals = numpy.array([[1, 2, 3, 4, 5], [5, 4, 3, 2, 1]], numpy.float32)
terminals = numpy.array([[0, 0, 1, 0, 0], [0, 0, 0, 0, 0]], numpy.float32)
values = numpy.array(
[[-100, 10, 20, 30, 40, 50], [-150, 15, 25, 35, 45, 55]], numpy.float32
) # Future values
return NOD({"signals": signals, "terminals": terminals, "values": values})
def asdijhsadasdad():
""" """
from warg import NOD
a = NOD(a=[1, 2, 8], b=[4, 3, 99])
print(f"ValueMapProduct{str(list(map_value_product(a.as_dict())))}")
print(f"MapProduct{str(list(map_product(a.as_dict())))}")
print(f"map_combinations{str(list(map_combinations(a.as_dict())))}")
print(f"map_permutations{str(list(map_permutations(a.as_dict())))}")
print(f"map_combinations_with_replacement{str(list(map_combinations_with_replacement(a.as_dict())))}")
def pull_gpu_info(include_graphics_processes: bool = True) -> dict:
"""Get all information about all your graphics cards.
Returns:
dict: The returned result is a dict with 3 keys: count, driver_version and devices:
count: Number of gpus found
driver_version: The version of the system’s graphics driver
devices: It's a list and every item is a namedtuple Device which has 10 fields, for exzample id,
name and fan_speed etc.
It should be noted that the Process field is also a namedtuple which has 11 fields."""
driver_version, devices = get_nv_info(include_graphics_processes)
info = NOD()
info["count"] = len(devices)
info["driver_version"] = driver_version
info["devices"] = devices
return info.as_dict()
def main(setting_scope: SettingScopeEnum = SettingScopeEnum.user):
""" """
global LOG_WRITER
if setting_scope == SettingScopeEnum.user:
LOG_WRITER = LogWriter(
ensure_existence(PROJECT_APP_PATH.user_log) /
f"{PROJECT_NAME}_publisher.log")
else:
LOG_WRITER = LogWriter(
ensure_existence(PROJECT_APP_PATH.site_log) /
f"{PROJECT_NAME}_publisher.log")
LOG_WRITER.open()
client = mqtt.Client()
client.on_publish = on_publish
client.on_disconnect = on_disconnect
HEIMDALLR_SETTINGS = HeimdallrSettings(setting_scope)
client.username_pw_set(HEIMDALLR_SETTINGS.mqtt_username,
HEIMDALLR_SETTINGS.mqtt_password)
try:
client.connect(HEIMDALLR_SETTINGS.mqtt_broker,
HEIMDALLR_SETTINGS.mqtt_port,
keepalive=60)
except ValueError as ve:
raise ValueError(f"{HEIMDALLR_SETTINGS._mqtt_settings_path},"
f"{HEIMDALLR_SETTINGS.mqtt_broker},"
f"{HEIMDALLR_SETTINGS.mqtt_port},"
f"{ve}")
client.loop_start()
sensor_data = NOD({HOSTNAME: pull_gpu_info()})
if True: # with IgnoreInterruptSignal():
print("Publisher started")
def job():
""" """
sensor_data[HOSTNAME] = pull_gpu_info()
s = sensor_data.as_dict()
s = json.dumps(s)
client.publish(ALL_CONSTANTS.MQTT_TOPIC, s, ALL_CONSTANTS.MQTT_QOS)
schedule.every(ALL_CONSTANTS.MQTT_PUBLISH_INTERVAL_SEC).seconds.do(job)
for _ in busy_indicator():
schedule.run_pending()
time.sleep(1)
# noinspection PyUnreachableCode
LOG_WRITER.close()
client.loop_stop()
client.disconnect()
def create_licenses(self):
"""Creates the "license" portion of the COCO json"""
license_json = self._dataset_info["license"]
return [
NOD(
url=license_json["url"],
id=license_json["id"],
name=license_json["name"],
)
]
def create_info(self):
"""Creates the "info" piece of the COCO json"""
info_json = self._dataset_info["info"]
return NOD(
description=info_json["description"],
version=info_json["version"],
url=info_json["url"],
year=info_json["year"],
contributor=info_json["contributor"],
date_created=info_json["date_created"],
)
def main():
from configs.mobilenet_v2_ssd320_voc0712 import base_cfg
# from configs.efficient_net_b3_ssd300_voc0712 import base_cfg
# from configs.vgg_ssd300_voc0712 import base_cfg
parser = argparse.ArgumentParser(
description="Single Shot MultiBox Detector Training With PyTorch"
)
parser.add_argument("--local_rank", type=int, default=0)
parser.add_argument(
"--log_step", default=10, type=int, help="Print logs every log_step"
)
parser.add_argument(
"--save_step", default=2500, type=int, help="Save checkpoint every save_step"
)
parser.add_argument(
"--eval_step",
default=2500,
type=int,
help="Evaluate dataset every eval_step, disabled when eval_step < 0",
)
parser.add_argument("--use_tensorboard", default=True, type=str2bool)
parser.add_argument(
"--skip-test",
dest="skip_test",
help="Do not test the final model",
action="store_true",
)
args = parser.parse_args()
num_gpus = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
args.distributed = num_gpus > 1
args.num_gpus = num_gpus
set_benchmark_device_dist(args.distributed, args.local_rank)
logger = setup_distributed_logger(
"SSD",
global_distribution_rank(),
ensure_existence(PROJECT_APP_PATH.user_data / "results")
)
logger.info(f"Using {num_gpus} GPUs")
logger.info(args)
with TorchCacheSession():
model = train_ssd(
base_cfg.data_dir,
base_cfg,
base_cfg.solver,
NOD(**args.__dict__)
)
if not args.skip_test:
logger.info("Start evaluating...")
do_ssd_evaluation(base_cfg, model, distributed=args.distributed)
def create_coco_image(image_path, image_id, image_license):
"""Creates the "image" portion of COCO json"""
# Open the image and get the size
image_file = Image.open(image_path)
width, height = image_file.size
return NOD(
license=image_license,
file_name=image_path.name,
width=width,
height=height,
id=image_id,
)
def get_terminal_size() -> NOD:
"""
:return:
:rtype:"""
try:
size = shutil.get_terminal_size()
columns, rows = size.columns, size.lines
except:
rows, columns = (os.getenv("LINES", 25), os.getenv("COLUMNS", 80))
rows, columns = int(rows), int(columns)
return NOD(rows=rows, columns=columns)
def asdijha() -> None:
"""
:rtype: None
"""
from warg import NOD
a = NOD(a=[1], b=[4], c=[8])
print(f"ValueMapProduct{str(list(map_value_product(a.as_dict())))}")
print(f"MapProduct{str(list(map_product(a.as_dict())))}")
print(f"map_combinations{str(list(map_combinations(a.as_dict())))}")
print(f"map_permutations{str(list(map_permutations(a.as_dict())))}")
print(
f"map_combinations_with_replacement{str(list(map_combinations_with_replacement(a.as_dict())))}"
)
def train(self, **explicit_overrides) -> None:
"""
@param explicit_overrides: Accepts kwarg overrides to config
@return:
"""
default_config = NOD(AGENT_CONFIG[self.agent_key])
config_overrides = upper_dict(explicit_overrides)
for key, arg in config_overrides.items():
setattr(default_config, key, arg)
print("Explicit Overrides:")
print(explicit_overrides)
#print(default_config)
self.agent_callable(config=default_config, **explicit_overrides)
def calculate_loss(seg, recon, depth, normals):
"""
:param seg:
:type seg:
:param recon:
:type recon:
:param depth:
:type depth:
:param normals:
:type normals:
:return:
:rtype:
"""
(
(seg_pred, seg_target),
(recon_pred, recon_target),
(depth_pred, depth_target),
(normals_pred, normals_target),
) = (seg, recon, depth, normals)
seg_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(
seg_pred, seg_target
)
ae_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(
recon_pred, recon_target
)
normals_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(
normals_pred, normals_target
)
depth_bce_loss = torch.nn.functional.binary_cross_entropy_with_logits(
depth_pred, depth_target
)
pred_soft = torch.sigmoid(seg_pred)
dice = dice_loss(pred_soft, seg_target, epsilon=1)
jaccard = jaccard_loss(pred_soft, seg_target, epsilon=1)
terms = (dice, jaccard, ae_bce_loss, seg_bce_loss, depth_bce_loss, normals_bce_loss)
term_weight = 1 / len(terms)
weighted_terms = [term.mean() * term_weight for term in terms]
loss = sum(weighted_terms)
return NOD(loss=loss, terms=terms)
def get_ram_config():
config, unparsed = parser.parse_known_args()
ram_base = "ram"
config = NOD(**config.__dict__)
# config.data_dir = PROJECT_APP_PATH.user_data / ram_base / 'data'
config.data_dir = Path.home() / "Data" / "vision_sample_data"
config.ckpt_dir = PROJECT_APP_PATH.user_data / ram_base / "ckpt"
config.logs_dir = PROJECT_APP_PATH.user_log / ram_base / "logs"
config.plot_dir = PROJECT_APP_PATH.user_log / ram_base / "plots"
return config # , unparsed
def __call__(self, image: ndarray, target: Mapping[str, Any]) -> Tuple:
w, h = image.size
image_id = torch.tensor([target["image_id"]])
anno = [obj for obj in target["annotations"] if obj.iscrowd == 0]
boxes = [obj.BoundingBox for obj in anno]
# guard against no boxes via resizing
boxes = torch.as_tensor(boxes, dtype=torch.float32).reshape(-1, 4)
boxes[:, 2:] += boxes[:, :2]
boxes[:, 0::2].clamp_(min=0, max=w)
boxes[:, 1::2].clamp_(min=0, max=h)
classes = torch.tensor([obj.category_id for obj in anno],
dtype=torch.int64)
masks = convert_coco_poly_to_mask([obj.segmentation for obj in anno],
h, w)
keypoints = None
if anno and anno[0].Keypoints is not None:
keypoints = [obj.Keypoints for obj in anno]
keypoints = torch.as_tensor(keypoints, dtype=torch.float32)
num_keypoints = keypoints.shape[0]
if num_keypoints:
keypoints = keypoints.view(num_keypoints, -1, 3)
keep = (boxes[:, 3] > boxes[:, 1]) & (boxes[:, 2] > boxes[:, 0])
target = NOD(
boxes=boxes[keep],
labels=classes[keep],
masks=masks[keep],
image_id=image_id,
area=torch.tensor([obj.area for obj in anno]),
iscrowd=torch.tensor([obj.iscrowd for obj in anno]),
keypoints=None,
)
if keypoints is not None:
target.keypoints = keypoints[keep]
return image, target
def create_categories(self):
"""Creates the "categories" portion of the COCO json
Returns:
categories: category objects that become part of the final json
category_ids_by_name: a lookup dictionary for category ids based
on the name of the category"""
categories = []
category_ids_by_name = dict()
category_id = 1 # 0 is reserved for the background
super_categories = self._mask_definitions["super_categories"]
for super_category, _categories in super_categories.items():
for category_name in _categories:
categories.append(
NOD(supercategory=super_category,
id=category_id,
name=category_name))
category_ids_by_name[category_name] = category_id
category_id += 1
return categories, category_ids_by_name
def main(**config):
"""
Args:
config:
"""
config = NOD(**config)
torch.manual_seed(config.random_seed) # ensure reproducibility
kwargs = {}
if config.use_gpu:
torch.cuda.manual_seed(config.random_seed)
kwargs["num_workers"] = 1
kwargs["pin_memory"] = True
if config.is_train: # instantiate data loaders
data_loader = MNISTDataset.get_train_valid_loader(
config.data_dir,
batch_size=config.batch_size,
random_seed=config.random_seed,
valid_size=config.valid_size,
shuffle=config.shuffle,
**kwargs,
)
else:
data_loader = MNISTDataset.get_test_loader(config.data_dir,
config.batch_size, **kwargs)
trainer = Trainer(data_loader, **config)
with PytorchTensorboardWriter() as writer:
if config.is_train:
trainer.train(writer=writer)
else: # or load a pretrained model and test
trainer.test(writer=writer)
def test_model(model, data_iterator, latest_model_path, num_columns: int = 2):
model = model.eval().to(global_torch_device())
inputs, labels = next(data_iterator)
inputs = inputs.to(global_torch_device())
labels = labels.to(global_torch_device())
with torch.no_grad():
pred = model(inputs)
y_pred = pred.data.to("cpu").numpy()
y_pred_max = numpy.argmax(y_pred, axis=-1)
accuracy_w = accuracy_score(labels, y_pred_max)
precision_a, recall_a, fscore_a, support_a = precision_recall_fscore_support(
labels, y_pred_max)
precision_w, recall_w, fscore_w, support_w = precision_recall_fscore_support(
labels, y_pred_max, average="weighted")
_, predicted = torch.max(pred, 1)
truth_labels = labels.data.to("cpu").numpy()
input_images_rgb = [
default_torch_retransform(x) for x in inputs.to(global_torch_device())
]
cell_width = (800 / num_columns) - 6 - 6 * 2
pyplot.plot(numpy.random.random((3, 3)))
alphabet = string.ascii_lowercase
class_names = numpy.array([*alphabet])
samples = len(y_pred)
predictions = [[None for _ in range(num_columns)]
for _ in range(samples // num_columns)]
for i, a, b, c in zip(range(samples), input_images_rgb, y_pred_max,
truth_labels):
pyplot.imshow(a)
if b == c:
outcome = "tp"
else:
outcome = "fn"
gd = ReportEntry(
name=i,
figure=plt_html(a, format="jpg", size=(cell_width, cell_width)),
prediction=class_names[b],
truth=class_names[c],
outcome=outcome,
explanation=None,
)
predictions[i // num_columns][i % num_columns] = gd
cfmat = confusion_matrix_plot(y_pred_max, truth_labels, class_names)
title = "Classification Report"
model_name = latest_model_path
confusion_matrix = plt_html(cfmat, format="png", size=(800, 800))
accuracy = generate_math_html("\dfrac{tp+tn}{N}"), None, accuracy_w
precision = generate_math_html(
"\dfrac{tp}{tp+fp}"), precision_a, precision_w
recall = generate_math_html("\dfrac{tp}{tp+fn}"), recall_a, recall_w
f1_score = (
generate_math_html("2*\dfrac{precision*recall}{precision+recall}"),
fscore_a,
fscore_w,
)
support = generate_math_html("N_{class_truth}"), support_a, support_w
metrics = NOD.nod_of(accuracy, precision, f1_score, recall,
support).as_flat_tuples()
bundle = NOD.nod_of(title, model_name, confusion_matrix, metrics,
predictions)
file_name = Path(title.lower().replace(" ", "_"))
generate_html(file_name.with_suffix(".html"), **bundle)
generate_pdf(file_name.with_suffix(".html"), file_name.with_suffix(".pdf"))
from draugr.torch_utilities.tensors.tensor_container import NamedTensorTuple
from draugr.torch_utilities import Split
from warg import NOD
class CocoModeEnum(Enum):
instances = "instances"
person_keypoints = "person_keypoints"
CocoPolyAnnotation = NOD({
"image_id": None,
"bbox": None,
"category_id": None,
"area": None,
"iscrowd": None,
"id": None,
"segmentation": None,
"keypoints": None,
"num_keypoints":None
})
CocoMask = NOD({"boxes": None,
"labels": None,
"masks": None,
"image_id": None,
"area": None,
"iscrowd": None,
"keypoints":None
})
from pathlib import Path
from warg import NOD
MQTT_CAM_CONFIG = NOD(
mqtt=NOD(
broker="localhost", port=1883, QOS=1
), # or an ip address like 192.168.1.74
camera=NOD(
video_source=0,
fps=30, # 2
mqtt_topic="video/video0/capture",
# If your desired camera is listed as source 0 you will configure video_source: 0. Alternatively
# you can configure the video source as an MJPEG or RTSP stream. For example in config.yml you may
# configure something like video_source: "rtsp://*****:*****@192.168.1.94:554/11" for a RTSP
# camera.
),
processing=NOD(
subscribe_topic="video/video0/capture",
publish_topic="video/video0/capture/rotated",
),
save_captures=NOD(
mqtt_topic="video/video0/capture", captures_directory=Path("captures")
),
)
TRAIN_DIR.mkdir(parents=True)
DATA_DIR = PROJECT_APP_PATH.user_data / "vanilla_vae" / "data"
if not DATA_DIR.exists():
DATA_DIR.mkdir(parents=True)
cfg = NOD(
latent_size=128,
variational="flow",
flow_depth=2,
data_size=784,
learning_rate=0.001,
batch_size=128,
test_batch_size=512,
max_iterations=100000,
log_interval=10000,
early_stopping_interval=5,
n_samples=128,
use_gpu=True,
train_dir=TRAIN_DIR,
data_dir=DATA_DIR,
seed=42,
)
device = torch.device("cuda" if cfg.use_gpu else "cpu")
torch.manual_seed(cfg.seed)
numpy.random.seed(cfg.seed)
random.seed(cfg.seed)
def __crystallise__(self) -> NOD:
return NOD({k: getattr(self, k) for k in self})
def get_nv_info(include_graphics_processes: bool = True):
devices = []
try:
driver_version = bindings.nvmlSystemGetDriverVersion().decode()
device_count = bindings.nvmlDeviceGetCount()
for device_i in range(device_count):
handle = bindings.nvmlDeviceGetHandleByIndex(device_i)
device_name = bindings.nvmlDeviceGetName(handle).decode()
gpu_mem_info = bindings.nvmlDeviceGetMemoryInfo(handle)
gpu_processes = bindings.nvmlDeviceGetComputeRunningProcesses(
handle)
if include_graphics_processes:
gpu_processes = (
gpu_processes +
bindings.nvmlDeviceGetGraphicsRunningProcesses(handle))
processes_info = []
for p in gpu_processes:
pid = p.pid
used_gpu_mem = p.usedGpuMemory
p = psutil.Process(pid=pid)
_ = p.cpu_percent()
time.sleep(
0.1
) # Recommended to preprobe and sleep for atleast 0.1 seconds.
processes_info.append(
NOD(
used_gpu_mem=used_gpu_mem,
device_idx=device_i,
name=p.name(),
username=p.username(),
memory_percent=p.memory_percent(),
cpu_percent=p.cpu_percent(),
cmdline=" ".join(p.cmdline()),
device_name=device_name,
create_time=p.create_time(),
status=p.status(),
pid=pid,
).as_dict())
"""
try:
fan_speed = pynvml.nvmlDeviceGetFanSpeed(handle)
power_usage = pynvml.nvmlDeviceGetPowerUsage(handle) # milliwatts mW
except pynvml.NVMLError_NotSupported as e:
fan_speed = None
power_usage = None
power_state = pynvml.nvmlDeviceGetPowerState(handle)
temperature = pynvml.nvmlDeviceGetTemperature(handle, pynvml.NVML_TEMPERATURE_GPU)
"""
devices.append(
NOD(
id=device_i,
name=device_name,
free=gpu_mem_info.free,
used=gpu_mem_info.used,
total=gpu_mem_info.total,
processes=processes_info,
).as_dict())
except Exception as e:
print(e)
driver_version = "No nvidia driver"
return driver_version, devices
from neodroidagent.common import CategoricalMLP
from neodroidagent.configs.base_config import *
from warg import NOD
CONFIG_NAME = __name__
import pathlib
CONFIG_FILE_PATH = pathlib.Path(__file__)
# Architecture
POLICY_ARCH_SPEC = GDKC(
CategoricalMLP,
NOD(
input_shape=None, # Obtain from environment
hidden_layers=(32, 32), # Estimate from input and output size
output_shape=None, # Obtain from environment
hidden_layer_activation=torch.relu,
use_bias=True,
),
)
ROLLOUTS = 10000
ENVIRONMENT_NAME = "CartPole-v1"
"""
Description
A pole is attached by an un-actuated joint to a cart, which moves along a frictionless track. The pendulum
starts upright, and the goal is to prevent it from falling over by increasing and reducing the cart's
velocity.
res = []
for kpts in detections:
d = {n: k.round().astype(int).tolist() for (n, k) in zip(names, kpts)}
res.append(d)
return res
def grab_video_frame(cap):
ret, frame = cap.read()
return cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
frame_i = 0
time_s = time.time()
image_axs = NOD()
env = VectorWrapper(UnityEnvironment(connect_to_running=True))
fig = pyplot.figure()
print_obs = False
def update_figures(i):
global time_s, frame_i, image_axs
sample = env.action_space.sample()
obs, signal, terminated, info = env.react(sample).to_gym_like_output()
if print_obs:
print(i)
for obs in info.sensors.values():
print(obs)
"DECORATIONS",
"generate_style",
"sprint",
"PrintStyle",
"scale",
"get_terminal_size",
"hyperlink_path",
"hyperlink_url",
]
COLORS = NOD(
red="31",
green="32",
yellow="33",
# gray='30', #Black,
blue="34",
magenta="35",
cyan="36",
white="37",
crimson="38",
)
DECORATIONS = NOD(
end="0",
bold="1",
dim="2",
italic="3",
underline="4",
underline_end="24", # '4:0',
double_underline="21", # '4:2'
# double_underline_end='24', # '4:0'
SAN_CONFIG = NOD(
dataset_type=ImageNet2012,
dataset_path=Path.home() / "Data" / "Datasets" / "ILSVRC2012",
arch="san",
self_attention_type=0,
layers=[2, 1, 2, 4, 1],
kernels=[3, 7, 7, 7, 7],
ignore_label=2000,
base_lr=0.1,
epochs=100,
start_epoch=0,
step_epochs=[30, 60, 90],
label_smoothing=0.1,
scheduler="cosine",
momentum=0.9,
weight_decay=0.0001,
manual_seed=None,
print_freq=10,
save_freq=1,
train_gpu=[0, 1, 2, 3, 4, 5, 6, 7],
workers=32, # data loader workers
batch_size=256, # batch size for training
batch_size_val=
128, # batch size for validation during training, memory and speed tradeoff
batch_size_test=10, # 100,
evaluate=True,
# evaluate on validation set, extra gpu memory needed and small batch_size_val is recommend
dist_url="tcp://127.0.0.1:6789",
dist_backend="nccl",
multiprocessing_distributed=True,
world_size=1,
rank=0,
test_gpu=[0],
test_workers=10,
mixup_alpha=None, #
model_path=None, #
save_path=None, #
weight=None, # path to initial weight (default=none)
resume=None, # path to latest checkpoint (default=none)
)