pyanomaly.core.engine.abstract package¶

Submodules¶

pyanomaly.core.engine.abstract.abstract_engine module¶

@author: Yuhao Cheng @contact: yuhao.cheng[at]outlook.com

class pyanomaly.core.engine.abstract.abstract_engine.AbstractInference(*defaults, **kwargs)¶

Bases: object

after_inference()¶

before_inference()¶

inference()¶

run()¶: The basic loop of implement the algorithm

set_requires_grad(nets, requires_grad=False)¶

Args:: nets(list):a list of networks requores_grad(bool):whether the networks require gradients or not

class pyanomaly.core.engine.abstract.abstract_engine.AbstractTrainer(*args, **kwargs)¶

Bases: object

Abstract trainer class. The abstract defination of method and frame work during the training process. All of trainers must be the sub-class of this class.

after_step(current_step)¶: the function after step

after_train()¶: the function after train fucntion

before_step(current_step)¶: the fucntion before step

before_train()¶: the fucntion before train function

run(start_iter, max_iter)¶

Start the training process.

The basic loop of

set_requires_grad(nets, requires_grad=False)¶: Change the state of gradient. Args:

nets(list): a list of networks requores_grad(bool): whether the networks require gradients or not

train(current_step)¶: the single step of training the model

pyanomaly.core.engine.abstract.base_engine module¶

@author: Yuhao Cheng @contact: yuhao.cheng[at]outlook.com

class pyanomaly.core.engine.abstract.base_engine.BaseInference(*defaults, **kwargs)¶

Bases: pyanomaly.core.engine.abstract.abstract_engine.AbstractInference

abstract custom_setup()¶

data_parallel(model)¶: Data parallel the model

abstract inference(current_step)¶

load_pretrain()¶

set_all(is_train)¶

Set all of models in this trainer in eval or train model Args:

is_train: bool. True=train mode; False=eval mode

Returns:: None

class pyanomaly.core.engine.abstract.base_engine.BaseTrainer(*defaults, **kwargs)¶

Bases: pyanomaly.core.engine.abstract.abstract_engine.AbstractTrainer

after_step(current_step)¶: the function after step

after_train()¶: the function after train fucntion

before_step(current_step)¶: the fucntion before step

abstract custom_setup()¶

data_parallel(model)¶

Data parallel the model by using torch.nn.DataParallel Args:

model: torch.nn.Module

Returns:: model_parallel

fine_tune()¶

load_pretrain()¶

resume()¶

save(current_epoch, best=False)¶: self.saved_model: is the model or a dict of combination of models self.saved_optimizer: is the optimizer or a dict of combination of optimizers self.saved_loss: the loss or a dict of the combination of loss