def __init__(self, config):
super().__init__(config)
# Create an instance from the Model
self.logger.info("Loading encoder pretrained in imagenet...")
if self.config.pretrained_encoder:
pretrained_enc = torch.nn.DataParallel(
ERFNet(self.config.imagenet_nclasses)).cuda()
pretrained_enc.load_state_dict(
torch.load(self.config.pretrained_model_path)['state_dict'])
pretrained_enc = next(pretrained_enc.children()).features.encoder
else:
pretrained_enc = None
# define erfNet model
self.model = ERF(self.config, pretrained_enc)
# Create an instance from the data loader
self.data_loader = VOCDataLoader(self.config)
# Create instance from the loss
self.loss = CrossEntropyLoss(self.config)
# Create instance from the optimizer
self.optimizer = torch.optim.Adam(
self.model.parameters(),
lr=self.config.learning_rate,
betas=(self.config.betas[0], self.config.betas[1]),
eps=self.config.eps,
weight_decay=self.config.weight_decay)
# Define Scheduler
lambda1 = lambda epoch: pow(
(1 - ((epoch - 1) / self.config.max_epoch)), 0.9)
self.scheduler = lr_scheduler.LambdaLR(self.optimizer,
lr_lambda=lambda1)
# initialize my counters
self.current_epoch = 0
self.current_iteration = 0
self.best_valid_mean_iou = 0
# Check is cuda is available or not
self.is_cuda = torch.cuda.is_available()
# Construct the flag and make sure that cuda is available
self.cuda = self.is_cuda & self.config.cuda
if self.cuda:
torch.cuda.manual_seed_all(self.config.seed)
self.device = torch.device("cuda")
torch.cuda.set_device(self.config.gpu_device)
self.logger.info("Operation will be on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.device = torch.device("cpu")
torch.manual_seed(self.config.seed)
self.logger.info("Operation will be on *****CPU***** ")
self.model = self.model.to(self.device)
self.loss = self.loss.to(self.device)
# Model Loading from the latest checkpoint if not found start from scratch.
self.load_checkpoint(self.config.checkpoint_file)
# Tensorboard Writer
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='FCN8s')
def __init__(self, config):
super().__init__(config)
# Create an instance from the Model
self.model = CondenseNet(self.config)
# Create an instance from the data loader
self.data_loader = Cifar10DataLoader(self.config)
# Create instance from the loss
self.loss = CrossEntropyLoss()
# Create instance from the optimizer
self.optimizer = torch.optim.SGD(self.model.parameters(),
lr=self.config.learning_rate,
momentum=float(self.config.momentum),
weight_decay=self.config.weight_decay,
nesterov=True)
# initialize my counters
self.current_epoch = 0
self.current_iteration = 0
self.best_valid_acc = 0
# Check is cuda is available or not
self.is_cuda = torch.cuda.is_available()
# Construct the flag and make sure that cuda is available
self.cuda = self.is_cuda & self.config.cuda
if self.cuda:
self.device = torch.device("cuda")
torch.cuda.manual_seed_all(self.config.seed)
torch.cuda.set_device(self.config.gpu_device)
self.logger.info("Operation will be on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.device = torch.device("cpu")
torch.manual_seed(self.config.seed)
self.logger.info("Operation will be on *****CPU***** ")
self.model = self.model.to(self.device)
self.loss = self.loss.to(self.device)
# Model Loading from the latest checkpoint if not found start from scratch.
self.load_checkpoint(self.config.checkpoint_file)
# Tensorboard Writer
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='CondenseNet')
class CondenseNetAgent(BaseAgent):
def __init__(self, config):
super().__init__(config)
# Create an instance from the Model
self.model = CondenseNet(self.config)
# Create an instance from the data loader
self.data_loader = Cifar10DataLoader(self.config)
# Create instance from the loss
self.loss = CrossEntropyLoss()
# Create instance from the optimizer
self.optimizer = torch.optim.SGD(self.model.parameters(),
lr=self.config.learning_rate,
momentum=float(self.config.momentum),
weight_decay=self.config.weight_decay,
nesterov=True)
# initialize my counters
self.current_epoch = 0
self.current_iteration = 0
self.best_valid_acc = 0
# Check is cuda is available or not
self.is_cuda = torch.cuda.is_available()
# Construct the flag and make sure that cuda is available
self.cuda = self.is_cuda & self.config.cuda
if self.cuda:
self.device = torch.device("cuda")
torch.cuda.manual_seed_all(self.config.seed)
torch.cuda.set_device(self.config.gpu_device)
self.logger.info("Operation will be on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.device = torch.device("cpu")
torch.manual_seed(self.config.seed)
self.logger.info("Operation will be on *****CPU***** ")
self.model = self.model.to(self.device)
self.loss = self.loss.to(self.device)
# Model Loading from the latest checkpoint if not found start from scratch.
self.load_checkpoint(self.config.checkpoint_file)
# Tensorboard Writer
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='CondenseNet')
def save_checkpoint(self, filename='checkpoint.pth.tar', is_best=0):
"""
Saving the latest checkpoint of the training
:param filename: filename which will contain the state
:param is_best: flag is it is the best model
:return:
"""
state = {
'epoch': self.current_epoch,
'iteration': self.current_iteration,
'state_dict': self.model.state_dict(),
'optimizer': self.optimizer.state_dict(),
}
# Save the state
torch.save(state, self.config.checkpoint_dir + filename)
# If it is the best copy it to another file 'model_best.pth.tar'
if is_best:
shutil.copyfile(self.config.checkpoint_dir + filename,
self.config.checkpoint_dir + 'model_best.pth.tar')
def load_checkpoint(self, filename):
filename = self.config.checkpoint_dir + filename
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
checkpoint = torch.load(filename)
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.logger.info(
"Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir, checkpoint['epoch'],
checkpoint['iteration']))
except OSError as e:
self.logger.info(
"No checkpoint exists from '{}'. Skipping...".format(
self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def run(self):
"""
This function will the operator
:return:
"""
try:
if self.config.mode == 'test':
self.validate()
else:
self.train()
except KeyboardInterrupt:
self.logger.info("You have entered CTRL+C.. Wait to finalize")
def train(self):
"""
Main training function, with per-epoch model saving
"""
for epoch in range(self.current_epoch, self.config.max_epoch):
self.current_epoch = epoch
self.train_one_epoch()
valid_acc = self.validate()
is_best = valid_acc > self.best_valid_acc
if is_best:
self.best_valid_acc = valid_acc
self.save_checkpoint(is_best=is_best)
def train_one_epoch(self):
"""
One epoch training function
"""
# Initialize tqdm
tqdm_batch = tqdm(self.data_loader.train_loader,
total=self.data_loader.train_iterations,
desc="Epoch-{}-".format(self.current_epoch))
# Set the model to be in training mode
self.model.train()
# Initialize your average meters
epoch_loss = AverageMeter()
top1_acc = AverageMeter()
top5_acc = AverageMeter()
current_batch = 0
for x, y in tqdm_batch:
if self.cuda:
x, y = x.cuda(self.config.async_loading), y.cuda(
self.config.async_loading)
# current iteration over total iterations
progress = float(
self.current_epoch * self.data_loader.train_iterations +
current_batch) / (self.config.max_epoch *
self.data_loader.train_iterations)
# progress = float(self.current_iteration) / (self.config.max_epoch * self.data_loader.train_iterations)
x, y = Variable(x), Variable(y)
lr = adjust_learning_rate(self.optimizer,
self.current_epoch,
self.config,
batch=current_batch,
nBatch=self.data_loader.train_iterations)
# model
pred = self.model(x, progress)
# loss
cur_loss = self.loss(pred, y)
if np.isnan(float(cur_loss.item())):
raise ValueError('Loss is nan during training...')
# optimizer
self.optimizer.zero_grad()
cur_loss.backward()
self.optimizer.step()
top1, top5 = cls_accuracy(pred.data, y.data, topk=(1, 5))
epoch_loss.update(cur_loss.item())
top1_acc.update(top1.item(), x.size(0))
top5_acc.update(top5.item(), x.size(0))
self.current_iteration += 1
current_batch += 1
self.summary_writer.add_scalar("epoch/loss", epoch_loss.val,
self.current_iteration)
self.summary_writer.add_scalar("epoch/accuracy", top1_acc.val,
self.current_iteration)
tqdm_batch.close()
self.logger.info("Training at epoch-" + str(self.current_epoch) +
" | " + "loss: " + str(epoch_loss.val) +
"- Top1 Acc: " + str(top1_acc.val) + "- Top5 Acc: " +
str(top5_acc.val))
def validate(self):
"""
One epoch validation
:return:
"""
tqdm_batch = tqdm(self.data_loader.valid_loader,
total=self.data_loader.valid_iterations,
desc="Valiation at -{}-".format(self.current_epoch))
# set the model in training mode
self.model.eval()
epoch_loss = AverageMeter()
top1_acc = AverageMeter()
top5_acc = AverageMeter()
for x, y in tqdm_batch:
if self.cuda:
x, y = x.cuda(self.config.async_loading), y.cuda(
self.config.async_loading)
x, y = Variable(x), Variable(y)
# model
pred = self.model(x)
# loss
cur_loss = self.loss(pred, y)
if np.isnan(float(cur_loss.item())):
raise ValueError('Loss is nan during validation...')
top1, top5 = cls_accuracy(pred.data, y.data, topk=(1, 5))
epoch_loss.update(cur_loss.item())
top1_acc.update(top1.item(), x.size(0))
top5_acc.update(top5.item(), x.size(0))
self.logger.info("Validation results at epoch-" +
str(self.current_epoch) + " | " + "loss: " +
str(epoch_loss.avg) + "- Top1 Acc: " +
str(top1_acc.val) + "- Top5 Acc: " +
str(top5_acc.val))
tqdm_batch.close()
return top1_acc.avg
def finalize(self):
"""
Finalize all the operations of the 2 Main classes of the process the operator and the data loader
:return:
"""
self.logger.info(
"Please wait while finalizing the operation.. Thank you")
self.save_checkpoint()
self.summary_writer.export_scalars_to_json("{}all_scalars.json".format(
self.config.summary_dir))
self.summary_writer.close()
self.data_loader.finalize()
def __init__(self, config):
super().__init__(config)
self.config = config
self.onlineExpert = ComputeECBSSolution(self.config)
self.dataTransformer = DataTransformer(self.config)
self.recorder = MonitoringMultiAgentPerformance(self.config)
self.model = DecentralPlannerNet(self.config)
self.logger.info("Model: \n".format(print(self.model)))
# define data_loader
self.data_loader = DecentralPlannerDataLoader(config=config)
# define loss
self.loss = CrossEntropyLoss()
self.l1_reg = L1Regularizer(self.model)
self.l2_reg = L2Regularizer(self.model)
# define optimizers
self.optimizer = optim.Adam(self.model.parameters(),
lr=self.config.learning_rate,
weight_decay=self.config.weight_decay)
print(self.config.weight_decay)
self.scheduler = optim.lr_scheduler.CosineAnnealingLR(
self.optimizer, T_max=self.config.max_epoch, eta_min=1e-6)
# for param in self.model.parameters():
# print(param)
# for name, param in self.model.state_dict().items():
# print(name, param)
# initialize counter
self.current_epoch = 0
self.current_iteration = 0
self.current_iteration_validStep = 0
self.rateReachGoal = 0.0
# set cuda flag
self.is_cuda = torch.cuda.is_available()
if self.is_cuda and not self.config.cuda:
self.logger.info(
"WARNING: You have a CUDA device, so you should probably enable CUDA"
)
self.cuda = self.is_cuda & self.config.cuda
# set the manual seed for torch
self.manual_seed = self.config.seed
if self.cuda:
torch.cuda.manual_seed_all(self.manual_seed)
self.config.device = torch.device("cuda")
torch.cuda.set_device(self.config.gpu_device)
self.model = self.model.to(self.config.device)
self.loss = self.loss.to(self.config.device)
self.logger.info("Program will run on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.config.device = torch.device("cpu")
torch.manual_seed(self.manual_seed)
self.logger.info("Program will run on *****CPU*****\n")
# Model Loading from the latest checkpoint if not found start from scratch.
if self.config.train_TL or self.config.test_general:
self.load_pretrained_checkpoint(self.config.test_epoch,
lastest=self.config.lastest_epoch,
best=self.config.best_epoch)
else:
self.load_checkpoint(self.config.test_epoch,
lastest=self.config.lastest_epoch,
best=self.config.best_epoch)
# Summary Writer
self.robot = multiRobotSim(self.config)
self.switch_toOnlineExpert = False
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='NerualMAPP')
self.plot_graph = True
self.save_dump_input = False
self.dummy_input = None
self.dummy_gso = None
self.time_record = None
class DecentralPlannerAgentLocalWithOnlineExpert(BaseAgent):
def __init__(self, config):
super().__init__(config)
self.config = config
self.onlineExpert = ComputeECBSSolution(self.config)
self.dataTransformer = DataTransformer(self.config)
self.recorder = MonitoringMultiAgentPerformance(self.config)
self.model = DecentralPlannerNet(self.config)
self.logger.info("Model: \n".format(print(self.model)))
# define data_loader
self.data_loader = DecentralPlannerDataLoader(config=config)
# define loss
self.loss = CrossEntropyLoss()
self.l1_reg = L1Regularizer(self.model)
self.l2_reg = L2Regularizer(self.model)
# define optimizers
self.optimizer = optim.Adam(self.model.parameters(),
lr=self.config.learning_rate,
weight_decay=self.config.weight_decay)
print(self.config.weight_decay)
self.scheduler = optim.lr_scheduler.CosineAnnealingLR(
self.optimizer, T_max=self.config.max_epoch, eta_min=1e-6)
# for param in self.model.parameters():
# print(param)
# for name, param in self.model.state_dict().items():
# print(name, param)
# initialize counter
self.current_epoch = 0
self.current_iteration = 0
self.current_iteration_validStep = 0
self.rateReachGoal = 0.0
# set cuda flag
self.is_cuda = torch.cuda.is_available()
if self.is_cuda and not self.config.cuda:
self.logger.info(
"WARNING: You have a CUDA device, so you should probably enable CUDA"
)
self.cuda = self.is_cuda & self.config.cuda
# set the manual seed for torch
self.manual_seed = self.config.seed
if self.cuda:
torch.cuda.manual_seed_all(self.manual_seed)
self.config.device = torch.device("cuda")
torch.cuda.set_device(self.config.gpu_device)
self.model = self.model.to(self.config.device)
self.loss = self.loss.to(self.config.device)
self.logger.info("Program will run on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.config.device = torch.device("cpu")
torch.manual_seed(self.manual_seed)
self.logger.info("Program will run on *****CPU*****\n")
# Model Loading from the latest checkpoint if not found start from scratch.
if self.config.train_TL or self.config.test_general:
self.load_pretrained_checkpoint(self.config.test_epoch,
lastest=self.config.lastest_epoch,
best=self.config.best_epoch)
else:
self.load_checkpoint(self.config.test_epoch,
lastest=self.config.lastest_epoch,
best=self.config.best_epoch)
# Summary Writer
self.robot = multiRobotSim(self.config)
self.switch_toOnlineExpert = False
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='NerualMAPP')
self.plot_graph = True
self.save_dump_input = False
self.dummy_input = None
self.dummy_gso = None
self.time_record = None
# dummy_input = (torch.zeros(self.config.map_w,self.config.map_w, 3),)
# self.summary_writer.add_graph(self.model, dummy_input)
def save_checkpoint(self, epoch, is_best=0, lastest=True):
"""
Checkpoint saver
:param file_name: name of the checkpoint file
:param is_best: boolean flag to indicate whether current checkpoint's accuracy is the best so far
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
state = {
'epoch': self.current_epoch + 1,
'iteration': self.current_iteration,
'state_dict': self.model.state_dict(),
'optimizer': self.optimizer.state_dict(),
'scheduler_state_dict': self.scheduler.state_dict(),
}
# Save the state
torch.save(state, os.path.join(self.config.checkpoint_dir, file_name))
# If it is the best copy it to another file 'model_best.pth.tar'
if is_best:
shutil.copyfile(
os.path.join(self.config.checkpoint_dir, file_name),
os.path.join(self.config.checkpoint_dir, 'model_best.pth.tar'))
def load_pretrained_checkpoint(self, epoch, lastest=True, best=False):
"""
Latest checkpoint loader
:param file_name: name of the checkpoint file
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
elif best:
file_name = "model_best.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
filename = os.path.join(self.config.checkpoint_dir_load, file_name)
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
# checkpoint = torch.load(filename)
checkpoint = torch.load(filename,
map_location='cuda:{}'.format(
self.config.gpu_device))
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
self.logger.info(
"Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir_load, checkpoint['epoch'],
checkpoint['iteration']))
if self.config.train_TL:
param_name_GFL = '*GFL*'
param_name_action = '*actions*'
assert param_name_GFL != '', 'you must specified the name of the parameters to be re-trained'
for model_param_name, model_param_value in self.model.named_parameters(
):
# print("---All layers -- \n", model_param_name)
if fnmatch(model_param_name, param_name_GFL) or fnmatch(
model_param_name, param_name_action
): # and model_param_name.endswith('weight'):
# print("---retrain layers -- \n", model_param_name)
model_param_value.requires_grad = True
else:
# print("---freezed layers -- \n", model_param_name)
model_param_value.requires_grad = False
except OSError as e:
self.logger.info(
"No checkpoint exists from '{}'. Skipping...".format(
self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def load_checkpoint(self, epoch, lastest=True, best=False):
"""
Latest checkpoint loader
:param file_name: name of the checkpoint file
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
elif best:
file_name = "model_best.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
filename = os.path.join(self.config.checkpoint_dir, file_name)
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
# checkpoint = torch.load(filename)
checkpoint = torch.load(filename,
map_location='cuda:{}'.format(
self.config.gpu_device))
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
self.logger.info(
"Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir, checkpoint['epoch'],
checkpoint['iteration']))
except OSError as e:
self.logger.info(
"No checkpoint exists from '{}'. Skipping...".format(
self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def run(self):
"""
The main operator
:return:
"""
assert self.config.mode in ['train', 'test']
try:
if self.config.mode == 'test':
print("-------test------------")
start = time.process_time()
self.test('test')
self.time_record = time.process_time() - start
# self.test('test_trainingSet')
# self.pipeline_onlineExpert(self.current_epoch)
else:
self.train()
except KeyboardInterrupt:
self.logger.info("You have entered CTRL+C.. Wait to finalize")
def train(self):
"""
Main training loop
:return:
"""
for epoch in range(self.current_epoch, self.config.max_epoch + 1):
# for epoch in range(1, self.config.max_epoch + 1):
self.current_epoch = epoch
# TODO: Optional 1: del dataloader before train
self.train_one_epoch()
self.logger.info('Train {} on Epoch {}: Learning Rate: {}]'.format(
self.config.exp_name, self.current_epoch,
self.scheduler.get_lr()))
print('Train {} on Epoch {} Learning Rate: {}'.format(
self.config.exp_name, self.current_epoch,
self.scheduler.get_lr()))
rateReachGoal = 0.0
if self.config.num_agents >= 10:
if epoch % self.config.validate_every == 0:
rateReachGoal = self.test(self.config.mode)
self.switch_toOnlineExpert = True
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
else:
if epoch <= 4:
rateReachGoal = self.test(self.config.mode)
self.switch_toOnlineExpert = True
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
elif epoch % self.config.validate_every == 0:
rateReachGoal = self.test(self.config.mode)
self.switch_toOnlineExpert = True
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
# pass
is_best = rateReachGoal > self.rateReachGoal
if is_best:
self.rateReachGoal = rateReachGoal
self.save_checkpoint(epoch, is_best=is_best, lastest=True)
self.scheduler.step()
# TODO: Optional 2: del dataloader after train
self.excuation_onlineExport(epoch)
def excuation_onlineExport(self, epoch):
if epoch >= self.config.Start_onlineExpert:
if self.config.num_agents >= 10:
if epoch % self.config.validate_every == 0:
self.pipeline_onlineExpert(epoch)
else:
if epoch <= 4:
self.pipeline_onlineExpert(epoch)
elif epoch % self.config.validate_every == 0:
self.pipeline_onlineExpert(epoch)
def pipeline_onlineExpert(self, epoch):
# TODO: del dataloader
# create dataloader
self.onlineExpert.set_up()
self.onlineExpert.computeSolution()
self.dataTransformer.set_up(epoch)
self.dataTransformer.solutionTransformer()
del self.data_loader
self.data_loader = DecentralPlannerDataLoader(config=self.config)
def train_one_epoch(self):
"""
One epoch of training
:return:
"""
# Set the model to be in training mode
self.model.train()
# for param in self.model.parameters():
# print(param.requires_grad)
# for batch_idx, (input, target, GSO) in enumerate(self.data_loader.train_loader):
for batch_idx, (batch_input, batch_target, _, batch_GSO,
_) in enumerate(self.data_loader.train_loader):
inputGPU = batch_input.to(self.config.device)
gsoGPU = batch_GSO.to(self.config.device)
# gsoGPU = gsoGPU.unsqueeze(0)
targetGPU = batch_target.to(self.config.device)
batch_targetGPU = targetGPU.permute(1, 0, 2)
self.optimizer.zero_grad()
# loss
loss = 0
# model
self.model.addGSO(gsoGPU)
predict = self.model(inputGPU)
for id_agent in range(self.config.num_agents):
# for output, target in zip(predict, target):
batch_predict_currentAgent = predict[id_agent][:]
batch_target_currentAgent = batch_targetGPU[id_agent][:][:]
loss = loss + self.loss(
batch_predict_currentAgent,
torch.max(batch_target_currentAgent, 1)[1])
# print(loss)
loss = loss / self.config.num_agents
loss.backward()
# for param in self.model.parameters():
# print(param.grad)
self.optimizer.step()
if batch_idx % self.config.log_interval == 0:
self.logger.info(
'Train {} on Epoch {}: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.
format(
self.config.exp_name, self.current_epoch,
batch_idx * len(inputGPU),
len(self.data_loader.train_loader.dataset),
100. * batch_idx / len(self.data_loader.train_loader),
loss.item()))
self.current_iteration += 1
# print(loss)
log_loss = loss.item()
self.summary_writer.add_scalar("iteration/loss", log_loss,
self.current_iteration)
def test_step(self):
"""
One epoch of testing the accuracy of decision-making of each step
:return:
"""
# Set the model to be in training mode
self.model.eval()
log_loss_validStep = []
for batch_idx, (batch_input, batch_target, _, batch_GSO,
_) in enumerate(self.data_loader.validStep_loader):
inputGPU = batch_input.to(self.config.device)
gsoGPU = batch_GSO.to(self.config.device)
# gsoGPU = gsoGPU.unsqueeze(0)
targetGPU = batch_target.to(self.config.device)
batch_targetGPU = targetGPU.permute(1, 0, 2)
self.optimizer.zero_grad()
# loss
loss_validStep = 0
# model
self.model.addGSO(gsoGPU)
predict = self.model(inputGPU)
for id_agent in range(self.config.num_agents):
# for output, target in zip(predict, target):
batch_predict_currentAgent = predict[id_agent][:]
batch_target_currentAgent = batch_targetGPU[id_agent][:][:]
loss_validStep = loss_validStep + self.loss(
batch_predict_currentAgent,
torch.max(batch_target_currentAgent, 1)[1])
# print(loss)
loss_validStep = loss_validStep / self.config.num_agents
if batch_idx % self.config.log_interval == 0:
self.logger.info(
'ValidStep {} on Epoch {}: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'
.format(
self.config.exp_name, self.current_epoch,
batch_idx * len(inputGPU),
len(self.data_loader.validStep_loader.dataset), 100. *
batch_idx / len(self.data_loader.validStep_loader),
loss_validStep.item()))
log_loss_validStep.append(loss_validStep.item())
# self.current_iteration_validStep += 1
# self.summary_writer.add_scalar("iteration/loss_validStep", loss_validStep.item(), self.current_iteration_validStep)
# print(loss)
avg_loss = sum(log_loss_validStep) / len(log_loss_validStep)
self.summary_writer.add_scalar("epoch/loss_validStep", avg_loss,
self.current_epoch)
def test(self, mode):
"""
One cycle of model validation
:return:
"""
self.model.eval()
if mode == 'test':
dataloader = self.data_loader.test_loader
label = 'test'
elif mode == 'test_trainingSet':
dataloader = self.data_loader.test_trainingSet_loader
label = 'test_training'
if self.switch_toOnlineExpert:
self.robot.createfolder_failure_cases()
else:
dataloader = self.data_loader.valid_loader
label = 'valid'
size_dataset = dataloader.dataset.data_size
self.logger.info('\n{} set on {} in {} testing set \n'.format(
label, self.config.exp_name, size_dataset))
self.recorder.reset()
# maxstep = self.robot.getMaxstep()
with torch.no_grad():
for input, target, makespan, _, tensor_map in dataloader:
inputGPU = input.to(self.config.device)
targetGPU = target.to(self.config.device)
log_result = self.mutliAgent_ActionPolicy(
inputGPU, targetGPU, makespan, tensor_map,
self.recorder.count_validset, mode)
self.recorder.update(self.robot.getMaxstep(), log_result)
self.summary_writer = self.recorder.summary(label, self.summary_writer,
self.current_epoch)
self.logger.info(
'Accurracy(reachGoalnoCollision): {} \n '
'DeteriorationRate(MakeSpan): {} \n '
'DeteriorationRate(FlowTime): {} \n '
'Rate(collisionPredictedinLoop): {} \n '
'Rate(FailedReachGoalbyCollisionShielding): {} \n '.format(
round(self.recorder.rateReachGoal, 4),
round(self.recorder.avg_rate_deltaMP, 4),
round(self.recorder.avg_rate_deltaFT, 4),
round(self.recorder.rateCollisionPredictedinLoop, 4),
round(self.recorder.rateFailedReachGoalSH, 4),
))
# if self.config.mode == 'train' and self.plot_graph:
# self.summary_writer.add_graph(self.model,None)
# self.plot_graph = False
return self.recorder.rateReachGoal
def mutliAgent_ActionPolicy(self, input, load_target, makespanTarget,
tensor_map, ID_dataset, mode):
self.robot.setup(input, load_target, makespanTarget, tensor_map,
ID_dataset)
maxstep = self.robot.getMaxstep()
allReachGoal = False
noReachGoalbyCollsionShielding = False
check_collisionFreeSol = False
check_CollisionHappenedinLoop = False
check_CollisionPredictedinLoop = False
findOptimalSolution = False
compare_makespan, compare_flowtime = self.robot.getOptimalityMetrics()
currentStep = 0
Case_start = time.process_time()
Time_cases_ForwardPass = []
for step in range(maxstep):
currentStep = step + 1
currentState = self.robot.getCurrentState()
currentStateGPU = currentState.to(self.config.device)
gso = self.robot.getGSO(step)
gsoGPU = gso.to(self.config.device)
self.model.addGSO(gsoGPU)
# self.model.addGSO(gsoGPU.unsqueeze(0))
step_start = time.process_time()
actionVec_predict = self.model(currentStateGPU)
time_ForwardPass = time.process_time() - step_start
Time_cases_ForwardPass.append(time_ForwardPass)
allReachGoal, check_moveCollision, check_predictCollision = self.robot.move(
actionVec_predict, currentStep)
if check_moveCollision:
check_CollisionHappenedinLoop = True
if check_predictCollision:
check_CollisionPredictedinLoop = True
if allReachGoal:
# findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality()
# print("### Case - {} within maxstep - RealGoal: {} ~~~~~~~~~~~~~~~~~~~~~~".format(ID_dataset, allReachGoal))
break
elif currentStep >= (maxstep):
# print("### Case - {} exceed maxstep - RealGoal: {} - check_moveCollision: {} - check_predictCollision: {}".format(ID_dataset, allReachGoal, check_CollisionHappenedinLoop, check_CollisionPredictedinLoop))
break
num_agents_reachgoal = self.robot.count_numAgents_ReachGoal()
store_GSO, store_communication_radius = self.robot.count_GSO_communcationRadius(
currentStep)
if allReachGoal and not check_CollisionHappenedinLoop:
check_collisionFreeSol = True
noReachGoalbyCollsionShielding = False
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(
True)
if self.config.log_anime and self.config.mode == 'test':
self.robot.save_success_cases('success')
if currentStep >= (maxstep):
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(
False)
if mode == 'test_trainingSet' and self.switch_toOnlineExpert:
self.robot.save_failure_cases()
if currentStep >= (
maxstep
) and not allReachGoal and check_CollisionPredictedinLoop and not check_CollisionHappenedinLoop:
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(
False)
# print("### Case - {} -Step{} exceed maxstep({})- ReachGoal: {} due to CollsionShielding \n".format(ID_dataset,currentStep,maxstep, allReachGoal))
noReachGoalbyCollsionShielding = True
if self.config.log_anime and self.config.mode == 'test':
self.robot.save_success_cases('failure')
time_record = time.process_time() - Case_start
if self.config.mode == 'test':
exp_status = "################## {} - End of loop ################## ".format(
self.config.exp_name)
case_status = "####### Case{} \t Computation time:{} \t Step{}/{}\t- AllReachGoal-{}\n".format(
ID_dataset, time_record, currentStep, maxstep, allReachGoal)
self.logger.info('{} \n {}'.format(exp_status, case_status))
# if self.config.mode == 'test':
# self.robot.draw(ID_dataset)
# return [allReachGoal, noReachGoalbyCollsionShielding, findOptimalSolution, check_collisionFreeSol, check_CollisionPredictedinLoop, makespanPredict, makespanTarget, flowtimePredict,flowtimeTarget,num_agents_reachgoal]
return allReachGoal, noReachGoalbyCollsionShielding, findOptimalSolution, check_collisionFreeSol, check_CollisionPredictedinLoop, compare_makespan, compare_flowtime, num_agents_reachgoal, store_GSO, store_communication_radius, time_record, Time_cases_ForwardPass
def finalize(self):
"""
Finalizes all the operations of the 2 Main classes of the process, the operator and the data loader
:return:
"""
if self.config.mode == 'train':
print(self.model)
print("Experiment on {} finished.".format(self.config.exp_name))
print("Please wait while finalizing the operation.. Thank you")
# self.save_checkpoint()
self.summary_writer.export_scalars_to_json("{}all_scalars.json".format(
self.config.summary_dir))
self.summary_writer.close()
self.data_loader.finalize()
if self.config.mode == 'test':
print("################## End of testing ################## ")
print("Computation time:\t{} ".format(self.time_record))
class DecentralPlannerAgentLocal(BaseAgent):
def __init__(self, config):
super().__init__(config)
self.config = config
self.recorder = MonitoringMultiAgentPerformance(self.config)
self.model = DecentralPlannerNet(self.config, config.feature_noise_std, config.sybil_attack_count)
self.logger.info("Model: \n".format(print(self.model)))
# Add additional noise model parameters to config
self.map_noise_prob = config.map_noise_prob
self.map_shift_units = config.map_shift_units
self.move_noise_std = config.move_noise_std
self.comm_dropout_param = config.comm_dropout_param
# Add additonal attack model parameters to config
self.rogue_agent_count = config.rogue_agent_count
# define data_loader
self.data_loader = DecentralPlannerDataLoader(config=config)
# define loss
self.loss = CrossEntropyLoss()
self.l1_reg = L1Regularizer(self.model)
self.l2_reg = L2Regularizer(self.model)
# define optimizers
self.optimizer = optim.Adam(self.model.parameters(), lr=self.config.learning_rate, weight_decay=self.config.weight_decay)
print(self.config.weight_decay)
self.scheduler = optim.lr_scheduler.CosineAnnealingLR(self.optimizer, T_max=self.config.max_epoch, eta_min=1e-6)
# for param in self.model.parameters():
# print(param)
# for name, param in self.model.state_dict().items():
# print(name, param)
# initialize counter
self.current_epoch = 0
self.current_iteration = 0
self.current_iteration_validStep = 0
self.rateReachGoal = 0.0
# set cuda flag
self.is_cuda = torch.cuda.is_available()
if self.is_cuda and not self.config.cuda:
self.logger.info("WARNING: You have a CUDA device, so you should probably enable CUDA")
self.cuda = self.is_cuda & self.config.cuda
# set the manual seed for torch
self.manual_seed = self.config.seed
if self.cuda:
torch.cuda.manual_seed_all(self.manual_seed)
self.config.device = torch.device("cuda")
torch.cuda.set_device(self.config.gpu_device)
self.model = self.model.to(self.config.device)
self.loss = self.loss.to(self.config.device)
self.logger.info("Program will run on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.config.device = torch.device("cpu")
torch.manual_seed(self.manual_seed)
self.logger.info("Program will run on *****CPU*****\n")
# Model Loading from the latest checkpoint if not found start from scratch.
if self.config.train_TL or self.config.test_general:
self.load_pretrained_checkpoint(self.config.test_epoch, lastest=self.config.lastest_epoch, best=self.config.best_epoch)
else:
self.load_checkpoint(self.config.test_epoch, lastest=self.config.lastest_epoch, best=self.config.best_epoch)
# Summary Writer
self.robot = multiRobotSim(self.config)
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir, comment='NerualMAPP')
self.plot_graph = True
self.save_dump_input = False
self.dummy_input = None
self.dummy_gso = None
self.time_record = None
# dummy_input = (torch.zeros(self.config.map_w,self.config.map_w, 3),)
# self.summary_writer.add_graph(self.model, dummy_input)
self.results_file = open(self.config.data_root + '/results.txt', 'a+')
def save_checkpoint(self, epoch, is_best=0, lastest=True):
"""
Checkpoint saver
:param file_name: name of the checkpoint file
:param is_best: boolean flag to indicate whether current checkpoint's accuracy is the best so far
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
state = {
'epoch': self.current_epoch + 1,
'iteration': self.current_iteration,
'state_dict': self.model.state_dict(),
'optimizer': self.optimizer.state_dict(),
'scheduler_state_dict': self.scheduler.state_dict(),
}
# Save the state
torch.save(state, os.path.join(self.config.checkpoint_dir, file_name))
# If it is the best copy it to another file 'model_best.pth.tar'
if is_best:
shutil.copyfile(os.path.join(self.config.checkpoint_dir, file_name),
os.path.join(self.config.checkpoint_dir, 'model_best.pth.tar'))
def load_pretrained_checkpoint(self, epoch, lastest=True, best=False):
"""
Latest checkpoint loader
:param file_name: name of the checkpoint file
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
elif best:
file_name = "model_best.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
filename = os.path.join(self.config.checkpoint_dir_load, file_name)
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
checkpoint = torch.load(filename, map_location=torch.device('cpu'))
#checkpoint = torch.load(filename, map_location='cuda:{}'.format(self.config.gpu_device))
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
self.logger.info("Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir_load, checkpoint['epoch'], checkpoint['iteration']))
if self.config.train_TL:
param_name_GFL = '*GFL*'
param_name_action = '*actions*'
assert param_name_GFL != '', 'you must specified the name of the parameters to be re-trained'
for model_param_name, model_param_value in self.model.named_parameters():
# print("---All layers -- \n", model_param_name)
if fnmatch(model_param_name, param_name_GFL) or fnmatch(model_param_name, param_name_action): # and model_param_name.endswith('weight'):
# print("---retrain layers -- \n", model_param_name)
model_param_value.requires_grad = True
else:
# print("---freezed layers -- \n", model_param_name)
model_param_value.requires_grad = False
except OSError as e:
self.logger.info("No checkpoint exists from '{}'. Skipping...".format(self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def load_checkpoint(self, epoch, lastest=True, best=False):
"""
Latest checkpoint loader
:param file_name: name of the checkpoint file
:return:
"""
if lastest:
file_name = "checkpoint.pth.tar"
elif best:
file_name = "model_best.pth.tar"
else:
file_name = "checkpoint_{:03d}.pth.tar".format(epoch)
filename = os.path.join(self.config.checkpoint_dir, file_name)
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
checkpoint = torch.load(filename, map_location=torch.device('cpu'))
#checkpoint = torch.load(filename, map_location='cuda:{}'.format(self.config.gpu_device))
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
self.logger.info("Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir, checkpoint['epoch'], checkpoint['iteration']))
except OSError as e:
self.logger.info("No checkpoint exists from '{}'. Skipping...".format(self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def run(self):
"""
The main operator
:return:
"""
assert self.config.mode in ['train', 'test']
try:
if self.config.mode == 'test':
print("-------test------------")
start = time.process_time()
self.test('test')
self.time_record = time.process_time()-start
# self.test('test_trainingSet')
else:
self.train()
except KeyboardInterrupt:
self.logger.info("You have entered CTRL+C.. Wait to finalize")
def train(self):
"""
Main training loop
:return:
"""
for epoch in range(self.current_epoch, self.config.max_epoch + 1):
# for epoch in range(1, self.config.max_epoch + 1):
self.current_epoch = epoch
self.train_one_epoch()
# self.train_one_epoch_BPTT()
self.logger.info('Train {} on Epoch {}: Learning Rate: {}]'.format(self.config.exp_name, self.current_epoch, self.scheduler.get_lr()))
print('Train {} on Epoch {} Learning Rate: {}'.format(self.config.exp_name, self.current_epoch, self.scheduler.get_lr()))
rateReachGoal = 0.0
if self.config.num_agents >= 10:
if epoch % self.config.validate_every == 0:
rateReachGoal = self.test(self.config.mode)
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
else:
if epoch <= 4:
rateReachGoal = self.test(self.config.mode)
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
elif epoch % self.config.validate_every == 0:
rateReachGoal = self.test(self.config.mode)
self.test('test_trainingSet')
# self.test_step()
self.save_checkpoint(epoch, lastest=False)
# pass
is_best = rateReachGoal > self.rateReachGoal
if is_best:
self.rateReachGoal = rateReachGoal
self.save_checkpoint(epoch, is_best=is_best, lastest=True)
self.scheduler.step()
def train_one_epoch(self):
"""
One epoch of training
:return:
"""
# Set the model to be in training mode
self.model.train()
# for param in self.model.parameters():
# print(param.requires_grad)
# for batch_idx, (input, target, GSO) in enumerate(self.data_loader.train_loader):
for batch_idx, (batch_input, batch_target, _, batch_GSO, _) in enumerate(self.data_loader.train_loader):
inputGPU = batch_input.to(self.config.device)
gsoGPU = batch_GSO.to(self.config.device)
# gsoGPU = gsoGPU.unsqueeze(0)
targetGPU = batch_target.to(self.config.device)
batch_targetGPU = targetGPU.permute(1,0,2)
self.optimizer.zero_grad()
# loss
loss = 0
# model
self.model.addGSO(gsoGPU)
predict = self.model(inputGPU)
for id_agent in range(self.config.num_agents):
# for output, target in zip(predict, target):
batch_predict_currentAgent = predict[id_agent][:]
batch_target_currentAgent = batch_targetGPU[id_agent][:][:]
loss = loss + self.loss(batch_predict_currentAgent, torch.max(batch_target_currentAgent, 1)[1])
# print(loss)
loss = loss/self.config.num_agents
loss.backward()
# for param in self.model.parameters():
# print(param.grad)
self.optimizer.step()
if batch_idx % self.config.log_interval == 0:
self.logger.info('Train {} on Epoch {}: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(self.config.exp_name,
self.current_epoch, batch_idx * len(inputGPU), len(self.data_loader.train_loader.dataset),
100. * batch_idx / len(self.data_loader.train_loader), loss.item()))
self.current_iteration += 1
# print(loss)
log_loss = loss.item()
self.summary_writer.add_scalar("iteration/loss", log_loss, self.current_iteration)
def train_one_epoch_BPTT(self):
"""
One epoch of training
:return:
"""
# Set the model to be in training mode
self.model.train()
# seq_length = 5
seq_length = 10
# for batch_idx, (input, target, GSO) in enumerate(self.data_loader.train_loader):
# for batch_idx, (batch_input, batch_GSO, batch_target) in enumerate(self.data_loader.train_loader):
for batch_idx, (batch_input, batch_target, list_makespan, batch_GSO, _) in enumerate(self.data_loader.train_loader):
batch_makespan = max(list_makespan)
batch_size = batch_input.shape[1]
# print(mask_makespan)
inputGPU = batch_input.to(self.config.device)
gsoGPU = batch_GSO.to(self.config.device)
targetGPU = batch_target.to(self.config.device)
# for step in range(batch_makespan):
# self.model.initialize_hidden(batch_size)
log_loss = []
for id_seq in range(0, batch_makespan, seq_length):
# solution # 2
if id_seq == 0:
self.model.initialize_hidden(batch_size)
else:
self.model.detach_hidden()
if id_seq + seq_length + 1 >= batch_makespan:
id_seq_end = batch_makespan
self.retain_graph = True
else:
id_seq_end = id_seq + seq_length
self.retain_graph = True
# loss
loss = 0
# backpropagate after aggregate loss within certain number of step (5) instead of full makespan
for step in range(id_seq, id_seq_end):
# Back Propagation through time (BPTT)
step_inputGPU = inputGPU[step][:]
step_targetGPU = targetGPU[step][:]
step_gsoGPU = gsoGPU[step][:]
step_targetGPU = step_targetGPU.permute(1, 0, 2)
self.optimizer.zero_grad()
# model
self.model.addGSO(step_gsoGPU)
step_predict = self.model(step_inputGPU)
for id_agent in range(self.config.num_agents):
# for output, target in zip(predict, target):
batch_predict_currentAgent = step_predict[id_agent][:]
batch_target_currentAgent = step_targetGPU[id_agent][:]
loss = loss + self.loss(batch_predict_currentAgent,
torch.max(batch_target_currentAgent, 1)[1]) / self.config.num_agents
# print(loss)
# optimizer
loss.backward()
torch.nn.utils.clip_grad_norm_(self.model.parameters(), 5)
# solution # 1
# loss.backward(retain_graph=self.retain_graph)
#https://github.com/pytorch/examples/blob/e11e0796fc02cc2cd5b6ec2ad7cea21f77e25402/word_language_model/main.py#L155
# torch.nn.utils.clip_grad_norm(model.parameters(), 0.25)#args.clip)
# for p in model.parameters():
# p.data.add_(-lr, p.grad.data)
# for param in self.model.parameters():
# print(param.grad)
self.optimizer.step()
log_loss.append(loss.item())
avg_loss = sum(log_loss) / len(log_loss)
if batch_idx % self.config.log_interval == 0:
self.logger.info('Train {} on Epoch {}: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(self.config.exp_name,
self.current_epoch,
batch_idx * batch_size,
len(self.data_loader.train_loader.dataset),
100. * batch_idx / len(self.data_loader.train_loader),
avg_loss))
self.current_iteration += 1
# print(loss)
# log_loss = loss.item()
self.summary_writer.add_scalar("iteration/loss", avg_loss, self.current_iteration)
def test_step(self):
"""
One epoch of testing the accuracy of decision-making of each step
:return:
"""
# Set the model to be in training mode
self.model.eval()
log_loss_validStep = []
for batch_idx, (batch_input, batch_target, _, batch_GSO, _) in enumerate(self.data_loader.validStep_loader):
inputGPU = batch_input.to(self.config.device)
gsoGPU = batch_GSO.to(self.config.device)
# gsoGPU = gsoGPU.unsqueeze(0)
targetGPU = batch_target.to(self.config.device)
batch_targetGPU = targetGPU.permute(1, 0, 2)
self.optimizer.zero_grad()
# loss
loss_validStep = 0
# model
self.model.addGSO(gsoGPU)
predict = self.model(inputGPU)
for id_agent in range(self.config.num_agents):
# for output, target in zip(predict, target):
batch_predict_currentAgent = predict[id_agent][:]
batch_target_currentAgent = batch_targetGPU[id_agent][:][:]
loss_validStep = loss_validStep + self.loss(batch_predict_currentAgent, torch.max(batch_target_currentAgent, 1)[1])
# print(loss)
loss_validStep = loss_validStep/self.config.num_agents
if batch_idx % self.config.log_interval == 0:
self.logger.info('ValidStep {} on Epoch {}: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(self.config.exp_name,
self.current_epoch,
batch_idx * len(inputGPU),
len(self.data_loader.validStep_loader.dataset),
100. * batch_idx / len(self.data_loader.validStep_loader),
loss_validStep.item()))
log_loss_validStep.append(loss_validStep.item())
# self.current_iteration_validStep += 1
# self.summary_writer.add_scalar("iteration/loss_validStep", loss_validStep.item(), self.current_iteration_validStep)
# print(loss)
avg_loss = sum(log_loss_validStep)/len(log_loss_validStep)
self.summary_writer.add_scalar("epoch/loss_validStep", avg_loss, self.current_epoch)
def test(self, mode):
"""
One cycle of model validation
:return:
"""
self.model.eval()
if mode == 'test':
dataloader = self.data_loader.test_loader
label = 'test'
elif mode == 'test_trainingSet':
dataloader = self.data_loader.test_trainingSet_loader
label = 'test_training'
else:
dataloader = self.data_loader.valid_loader
label = 'valid'
self.logger.info('\n{} set on {} \n'.format(label, self.config.exp_name))
self.recorder.reset()
# maxstep = self.robot.getMaxstep()
with torch.no_grad():
for input, target, makespan, _, tensor_map in dataloader:
inputGPU = input.to(self.config.device)
targetGPU = target.to(self.config.device)
log_result = self.mutliAgent_ActionPolicy(inputGPU, targetGPU, makespan, tensor_map, self.recorder.count_validset)
self.recorder.update(self.robot.getMaxstep(), log_result)
self.summary_writer = self.recorder.summary(label, self.summary_writer, self.current_epoch)
results = ('Accurracy(reachGoalnoCollision): {} \n '
'DeteriorationRate(MakeSpan): {} \n '
'DeteriorationRate(FlowTime): {} \n '
'Rate(collisionPredictedinLoop): {} \n '
'Rate(FailedReachGoalbyCollisionShielding): {} \n '.format(
round(self.recorder.rateReachGoal, 4),
round(self.recorder.avg_rate_deltaMP, 4),
round(self.recorder.avg_rate_deltaFT, 4),
round(self.recorder.rateCollisionPredictedinLoop, 4),
round(self.recorder.rateFailedReachGoalSH, 4),
))
self.logger.info(results)
self.results_file.write('K={}, no OE\n'.format(self.config.nGraphFilterTaps))
self.results_file.write(results)
if self.recorder.avg_NonRogueFT:
nonRogueOut = 'NonRogueFT: {}\n'.format(round(self.recorder.avg_NonRogueFT,4))
self.results_file.write(nonRogueOut)
self.logger.info(nonRogueOut)
# if self.config.mode == 'train' and self.plot_graph:
# self.summary_writer.add_graph(self.model,None)
# self.plot_graph = False
return self.recorder.rateReachGoal
def mutliAgent_ActionPolicy(self, input, load_target, makespanTarget, tensor_map, ID_dataset):
t0_setup = time.process_time()
self.robot.setup(input, load_target, makespanTarget, tensor_map, ID_dataset)
deltaT_setup = time.process_time()-t0_setup
# print(" Computation time \t-[Step up]-\t\t :{} ".format(deltaT_setup))
maxstep = self.robot.getMaxstep()
allReachGoal = False
noReachGoalbyCollsionShielding = False
check_collisionFreeSol = False
check_CollisionHappenedinLoop = False
check_CollisionPredictedinLoop = False
findOptimalSolution = False
compare_makespan, compare_flowtime = self.robot.getOptimalityMetrics()
currentStep = 0
Case_start = time.process_time()
Time_cases_ForwardPass = []
for step in range(maxstep):
currentStep = step + 1
t0_getState = time.process_time()
currentState = self.robot.getCurrentState()
currentStateGPU = currentState.to(self.config.device)
deltaT_getState = time.process_time() - t0_getState
#print(" Computation time \t-[getState]-\t\t :{} ".format(deltaT_getState))
t0_getGSO = time.process_time()
gso = self.robot.getGSO(step)
deltaT_getGSO = time.process_time() - t0_getGSO
#print(" Computation time \t-[getGSO]-\t\t :{} ".format(deltaT_getGSO))
gsoGPU = gso.to(self.config.device)
pos_agents = self.robot.get_PosAgents()
self.model.addGSO(gsoGPU)
# model sensor failure by randomly flipping bits of
# map (which consists of 0s to indicate no object and 1s to indicate object)
if self.map_noise_prob:
bit_flip_mask = (torch.rand(currentStateGPU.shape) < self.map_noise_prob).to(self.config.device)
# bits in mask each = 1 with probability specified
# so xoring with this mask flips bits in map with probability specified
currentStateGPU = torch.logical_xor(currentStateGPU, bit_flip_mask).float().to(self.config.device)
# model miscalibration of sensor - shift map bits up by some number of units
if self.map_shift_units:
shifted_maps = currentStateGPU[:,:,:,-(currentStateGPU.shape[3] - self.map_shift_units):,:]
# zero pad the missing rows of the field of view maps
zero_pad_size = list(currentStateGPU.shape)
zero_pad_size[3] = self.map_shift_units
currentStateGPU = torch.cat((shifted_maps, torch.zeros(tuple(zero_pad_size), dtype = torch.float)), dim=3).to(self.config.device)
if self.comm_dropout_param:
# noise model:
# create mask to drop out each message between robots i,j with probability
# max(1,theta * distance(i,j)/(communication radius))
distances = squareform(pdist(self.robot.get_PosAgents()[0]))
loss_prob = torch.from_numpy(self.comm_dropout_param/self.robot.communicationRadius * distances).to(self.config.device)
loss_prob = torch.reshape(loss_prob, (1,1,loss_prob.shape[0], loss_prob.shape[1]))
comm_loss_mask = (torch.rand(loss_prob.shape) > loss_prob).to(self.config.device)
else:
comm_loss_mask = None
step_start = time.process_time()
actionVec_predict = self.model(currentStateGPU, comm_loss_mask)
# softmax of actionVec_predict is used to determine probabilities of each of the 5 moves
# (so at test time, the argmax of actionVec_predict is taken as the move).
# To simulate control errors (e.g. motors not properly responding to commands, breaking, wheels slipping, etc)
# we add gaussian noise to actionVec_predict
if self.move_noise_std:
for av in actionVec_predict:
av += torch.normal(0.0,self.move_noise_std,list(av.shape))
time_ForwardPass = time.process_time() - step_start
step_move = time.process_time()
allReachGoal, check_moveCollision, check_predictCollision = self.robot.move(actionVec_predict, currentStep, self.rogue_agent_count)
deltaT_move = time.process_time() - step_move
#print(" Computation time \t-[move]-\t\t :{} ".format(deltaT_move))
#print(" Computation time \t-[loopStep]-\t\t :{}\n ".format(time.process_time() - t0_getState))
Time_cases_ForwardPass.append([deltaT_setup, deltaT_getState, deltaT_getGSO, time_ForwardPass, deltaT_move])
if check_moveCollision:
check_CollisionHappenedinLoop = True
if check_predictCollision:
check_CollisionPredictedinLoop = True
if allReachGoal:
# findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality()
# print("### Case - {} within maxstep - RealGoal: {} ~~~~~~~~~~~~~~~~~~~~~~".format(ID_dataset, allReachGoal))
break
elif currentStep >= (maxstep):
# print("### Case - {} exceed maxstep - RealGoal: {} - check_moveCollision: {} - check_predictCollision: {}".format(ID_dataset, allReachGoal, check_CollisionHappenedinLoop, check_CollisionPredictedinLoop))
break
num_agents_reachgoal = self.robot.count_numAgents_ReachGoal()
store_GSO, store_communication_radius = self.robot.count_GSO_communcationRadius(currentStep)
if allReachGoal and not check_CollisionHappenedinLoop:
check_collisionFreeSol = True
noReachGoalbyCollsionShielding = False
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(True)
if self.config.log_anime and self.config.mode == 'test':
self.robot.save_success_cases('success')
if currentStep >= (maxstep):
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(False)
if currentStep >= (maxstep) and not allReachGoal and check_CollisionPredictedinLoop and not check_CollisionHappenedinLoop:
findOptimalSolution, compare_makespan, compare_flowtime = self.robot.checkOptimality(False)
# print("### Case - {} -Step{} exceed maxstep({})- ReachGoal: {} due to CollsionShielding \n".format(ID_dataset,currentStep,maxstep, allReachGoal))
noReachGoalbyCollsionShielding = True
if self.config.log_anime and self.config.mode == 'test':
self.robot.save_success_cases('failure')
time_record = time.process_time() - Case_start
if self.config.mode == 'test':
exp_status = "################## {} - End of loop ################## ".format(self.config.exp_name)
case_status = "####### Case{} \t Computation time:{} \t Step{}/{}\t- AllReachGoal-{}\n".format(ID_dataset, time_record,
currentStep,
maxstep, allReachGoal)
self.logger.info('{} \n {}'.format(exp_status, case_status))
# if self.config.mode == 'test':
# self.robot.draw(ID_dataset)
# elif self.config.mode == 'train' and self.current_epoch == self.config.max_epoch:
# # self.robot.draw(ID_dataset)
# pass
# return [allReachGoal, noReachGoalbyCollsionShielding, findOptimalSolution, check_collisionFreeSol, check_CollisionPredictedinLoop, makespanPredict, makespanTarget, flowtimePredict,flowtimeTarget,num_agents_reachgoal]
return allReachGoal, noReachGoalbyCollsionShielding, findOptimalSolution, check_collisionFreeSol, check_CollisionPredictedinLoop, compare_makespan, compare_flowtime, num_agents_reachgoal, store_GSO, store_communication_radius, time_record,Time_cases_ForwardPass, self.robot.nonRogueFlowtimePredict
def finalize(self):
"""
Finalizes all the operations of the 2 Main classes of the process, the operator and the data loader
:return:
"""
if self.config.mode == 'train':
print(self.model)
print("Experiment on {} finished.".format(self.config.exp_name))
print("Please wait while finalizing the operation.. Thank you")
# self.save_checkpoint()
self.summary_writer.export_scalars_to_json("{}all_scalars.json".format(self.config.summary_dir))
self.summary_writer.close()
self.data_loader.finalize()
if self.config.mode == 'test':
print("################## End of testing ################## ")
time = "Computation time:{}\n".format(self.time_record)
print(time)
self.results_file.write(time)
self.results_file.close()
class ERFNetAgent(BaseAgent):
"""
This class will be responsible for handling the whole process of our architecture.
"""
def __init__(self, config):
super().__init__(config)
# Create an instance from the Model
self.logger.info("Loading encoder pretrained in imagenet...")
if self.config.pretrained_encoder:
pretrained_enc = torch.nn.DataParallel(
ERFNet(self.config.imagenet_nclasses)).cuda()
pretrained_enc.load_state_dict(
torch.load(self.config.pretrained_model_path)['state_dict'])
pretrained_enc = next(pretrained_enc.children()).features.encoder
else:
pretrained_enc = None
# define erfNet model
self.model = ERF(self.config, pretrained_enc)
# Create an instance from the data loader
#self.data_loader = VOCDataLoader(self.config)
self.data_loader = CityscapesDataLoader(self.config)
'''
net_h, net_w = 448, 896
augment = Compose([RandomHorizontallyFlip(), RandomSized((0.625, 0.75)),
RandomRotate(6), RandomCrop((net_h, net_w))])
local_path = "./data/Cityscapes"
self.data_loader = CityscapesLoader(local_path, split="test", is_transform=True, augmentations=None, gt="gtFine")
'''
########################################
self.color_transform = Colorize(self.config.num_classes)
self.image_transform = ToPILImage()
# Create instance from the loss
self.loss = CrossEntropyLoss(self.config)
# Create instance from the optimizer
self.optimizer = torch.optim.Adam(
self.model.parameters(),
lr=self.config.learning_rate,
betas=(self.config.betas[0], self.config.betas[1]),
eps=self.config.eps,
weight_decay=self.config.weight_decay)
# Define Scheduler
lambda1 = lambda epoch: pow(
(1 - ((epoch - 1) / self.config.max_epoch)), 0.9)
self.scheduler = lr_scheduler.LambdaLR(self.optimizer,
lr_lambda=lambda1)
# initialize my counters
self.current_epoch = 0
self.current_iteration = 0
self.best_valid_mean_iou = 0
# Check is cuda is available or not
self.is_cuda = torch.cuda.is_available()
# Construct the flag and make sure that cuda is available
self.cuda = self.is_cuda & self.config.cuda
if self.cuda:
torch.cuda.manual_seed_all(self.config.seed)
self.device = torch.device("cuda")
torch.cuda.set_device(self.config.gpu_device)
self.logger.info("Operation will be on *****GPU-CUDA***** ")
print_cuda_statistics()
else:
self.device = torch.device("cpu")
torch.manual_seed(self.config.seed)
self.logger.info("Operation will be on *****CPU***** ")
self.model = self.model.to(self.device)
self.loss = self.loss.to(self.device)
# Model Loading from the latest checkpoint if not found start from scratch.
self.load_checkpoint(self.config.checkpoint_file)
# Tensorboard Writer
self.summary_writer = SummaryWriter(log_dir=self.config.summary_dir,
comment='FCN8s')
# # scheduler for the optimizer
# self.scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(self.optimizer,
# 'min', patience=self.config.learning_rate_patience,
# min_lr=1e-10, verbose=True)
def save_checkpoint(self, filename='checkpoint.pth.tar', is_best=0):
"""
Saving the latest checkpoint of the training
:param filename: filename which will contain the state
:param is_best: flag is it is the best model
:return:
"""
state = {
'epoch': self.current_epoch + 1,
'iteration': self.current_iteration,
'state_dict': self.model.state_dict(),
'optimizer': self.optimizer.state_dict(),
}
# Save the state
torch.save(state, self.config.checkpoint_dir + filename)
# If it is the best copy it to another file 'model_best.pth.tar'
if is_best:
shutil.copyfile(self.config.checkpoint_dir + filename,
self.config.checkpoint_dir + 'model_best.pth.tar')
def load_checkpoint(self, filename):
filename = self.config.checkpoint_dir + filename
try:
self.logger.info("Loading checkpoint '{}'".format(filename))
checkpoint = torch.load(filename)
self.current_epoch = checkpoint['epoch']
self.current_iteration = checkpoint['iteration']
self.model.load_state_dict(checkpoint['state_dict'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.logger.info(
"Checkpoint loaded successfully from '{}' at (epoch {}) at (iteration {})\n"
.format(self.config.checkpoint_dir, checkpoint['epoch'],
checkpoint['iteration']))
except OSError as e:
self.logger.info(
"No checkpoint exists from '{}'. Skipping...".format(
self.config.checkpoint_dir))
self.logger.info("**First time to train**")
def run(self):
"""
This function will the operator
:return:
"""
assert self.config.mode in ['train', 'test', 'random']
try:
if self.config.mode == 'test':
self.test()
else:
self.train()
except KeyboardInterrupt:
self.logger.info("You have entered CTRL+C.. Wait to finalize")
def train(self):
"""
Main training function, with per-epoch model saving
"""
for epoch in range(self.current_epoch, self.config.max_epoch):
self.current_epoch = epoch
self.scheduler.step(epoch)
self.train_one_epoch()
valid_mean_iou, valid_loss = self.validate()
self.scheduler.step(valid_loss)
is_best = valid_mean_iou > self.best_valid_mean_iou
if is_best:
self.best_valid_mean_iou = valid_mean_iou
self.save_checkpoint(is_best=is_best)
def train_one_epoch(self):
"""
One epoch training function
"""
# Initialize tqdm
tqdm_batch = tqdm(self.data_loader.train_loader,
total=self.data_loader.train_iterations,
desc="Epoch-{}-".format(self.current_epoch))
# Set the model to be in training mode (for batchnorm)
self.model.train()
# Initialize your average meters
epoch_loss = AverageMeter()
metrics = IOUMetric(self.config.num_classes)
for x, y in tqdm_batch:
if self.cuda:
x, y = x.pin_memory().cuda(
non_blocking=self.config.async_loading), y.cuda(
non_blocking=self.config.async_loading)
x, y = Variable(x), Variable(y)
# model
pred = self.model(x)
# loss
cur_loss = self.loss(pred, y)
if np.isnan(float(cur_loss.item())):
raise ValueError('Loss is nan during training...')
# optimizer
self.optimizer.zero_grad()
cur_loss.backward()
self.optimizer.step()
epoch_loss.update(cur_loss.item())
_, pred_max = torch.max(pred, 1)
metrics.add_batch(pred_max.data.cpu().numpy(),
y.data.cpu().numpy())
self.current_iteration += 1
# exit(0)
epoch_acc, _, epoch_iou_class, epoch_mean_iou, _ = metrics.evaluate()
self.summary_writer.add_scalar("epoch-training/loss", epoch_loss.val,
self.current_iteration)
self.summary_writer.add_scalar("epoch_training/mean_iou",
epoch_mean_iou, self.current_iteration)
tqdm_batch.close()
print("Training Results at epoch-" + str(self.current_epoch) + " | " +
"loss: " + str(epoch_loss.val) + " - acc-: " + str(epoch_acc) +
"- mean_iou: " + str(epoch_mean_iou) + "\n iou per class: \n" +
str(epoch_iou_class))
def validate(self):
"""
One epoch validation
:return:
"""
tqdm_batch = tqdm(self.data_loader.valid_loader,
total=self.data_loader.valid_iterations,
desc="Valiation at -{}-".format(self.current_epoch))
# set the model in training mode
self.model.eval()
epoch_loss = AverageMeter()
metrics = IOUMetric(self.config.num_classes)
for x, y in tqdm_batch:
if self.cuda:
x, y = x.pin_memory().cuda(
non_blocking=self.config.async_loading), y.cuda(
non_blocking=self.config.async_loading)
x, y = Variable(x), Variable(y)
# model
pred = self.model(x)
# loss
cur_loss = self.loss(pred, y)
if np.isnan(float(cur_loss.item())):
#print("error")
raise ValueError('Loss is nan during Validation.')
_, pred_max = torch.max(pred, 1)
metrics.add_batch(pred_max.data.cpu().numpy(),
y.data.cpu().numpy())
epoch_loss.update(cur_loss.item())
epoch_acc, _, epoch_iou_class, epoch_mean_iou, _ = metrics.evaluate()
self.summary_writer.add_scalar("epoch_validation/loss", epoch_loss.val,
self.current_iteration)
self.summary_writer.add_scalar("epoch_validation/mean_iou",
epoch_mean_iou, self.current_iteration)
print("Validation Results at epoch-" + str(self.current_epoch) +
" | " + "loss: " + str(epoch_loss.val) + " - acc-: " +
str(epoch_acc) + "- mean_iou: " + str(epoch_mean_iou) +
"\n iou per class: \n" + str(epoch_iou_class))
tqdm_batch.close()
return epoch_mean_iou, epoch_loss.val
def test(self):
'''
test_loader = torch.utils.data.DataLoader(self.data_loader, batch_size = self.config.batch_size
,num_workers = self.config.data_loader_workers,
pin_memory=self.config.pin_memory, shuffle = False)
test_iterations = (len(self.data_loader) + self.config.batch_size) // self.config.batch_size
'''
tqdm_batch = tqdm(self.data_loader.test_loader,
total=self.data_loader.test_iterations,
desc="Test at -{}-".format(self.current_epoch))
#tqdm_batch = tqdm(test_loader, total = test_iterations, desc = "Test at -{}-".format(self.current_epoch))
# set the model in training mode
self.model.eval()
'''
for x, y in tqdm_batch:
if self.cuda:
x, y = x.pin_memory().cuda(non_blocking=self.config.async_loading), y.cuda(non_blocking=self.config.async_loading)
x, y = Variable(x), Variable(y)
# model
pred = self.model(x)
'''
i = 0
for x_name, x in tqdm_batch:
if self.cuda:
x = x.pin_memory().cuda(non_blocking=self.config.async_loading)
x = Variable(x)
x = x.unsqueeze(0)
pred = self.model(x)
segmented_img = self.image_transform(
self.color_transform(
pred[0].cpu().max(0)[1].data.unsqueeze(0)))
j = str(i)
imageio.imsave(j + ".png", segmented_img)
i += 1
#imageio.imsave(i+".png" , segmented_img)
tqdm_batch.close()
return
def finalize(self):
"""
Finalize all the operations of the 2 Main classes of the process the operator and the data loader
:return:
"""
print("Please wait while finalizing the operation.. Thank you")
self.save_checkpoint()
self.summary_writer.export_scalars_to_json("{}all_scalars.json".format(
self.config.summary_dir))
self.summary_writer.close()
self.data_loader.finalize()