machin.model

nets

machin.model.nets provides implementations for various popular network architectures.

class machin.model.nets.NeuralNetworkModule

Bases: torch.nn.modules.module.Module, abc.ABC

Note: input device and output device are determined by module parameters,

your input module / output submodule should not store parameters on more than one device, and you also should not move your output to other devices other than your parameter storage device in forward().

Initializes internal Module state, shared by both nn.Module and ScriptModule.

set_input_module(input_module)

Set the input submodule of current module.

Parameters

input_module (torch.nn.modules.module.Module) –

set_output_module(output_module)

Set the output submodule of current module.

Parameters

output_module (torch.nn.modules.module.Module) –

property input_device

property output_device

static find_child(seq, is_first=True)

Find the first / last leaf child module.

forward(*_)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training = None

machin.model.nets.dynamic_module_wrapper(wrapped_module)

Wrapped module must locate on one single device, but could be moved around.

Input device and output device are automatically detected.

Parameters

wrapped_module (torch.nn.modules.module.Module) –

machin.model.nets.static_module_wrapper(wrapped_module, input_device, output_device)

Wrapped module could locate on multiple devices, but must not be moved.

Input device and output device are statically specified by user.

Parameters

• wrapped_module (torch.nn.modules.module.Module) –

• torch.device] input_device (Union[str,) –

• torch.device] output_device (Union[str,) –

class machin.model.nets.ResNet(in_planes, depth, out_planes, out_pool_size=1, 1, norm='none')

Bases: machin.model.nets.base.NeuralNetworkModule

Create a resnet of specified depth.

Parameters

• in_planes (int) – Number of input planes.

• depth (int) – Depth of resnet. Could be one of 18, 34, 50, 101, 152.

• out_planes (int) – Number of output planes.

• out_pool_size – Size of pooling output

• norm – Normalization method, could be one of “none”, “batch” or “weight”.

training = None

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.