caffe2.python.core.GradientRegistry Class Reference

Inheritance diagram for caffe2.python.core.GradientRegistry:

Public Member Functions
def	RegisterGradient (cls, op_type)
def	GetGradientForOp (cls, op, g_output)
def	GetBackwardPass (cls, operators, ys, ys_generate_gradient=False)

Static Public Attributes
dictionary	gradient_registry_ = {}

Detailed Description

GradientRegistry holds the mapping from operators to their gradients.

Definition at line 1055 of file core.py.

Member Function Documentation

def caffe2.python.core.GradientRegistry.GetBackwardPass	(		cls,
			operators,
			ys,
			ys_generate_gradient = False
	)

Gets the backward pass for the list of operators.

Args:
    operators: a list of operators constituting the forward pass.
    ys: a list or a dictionary specifying what blobs we want to compute
derivatives of. If the input is a list, we will automatically
generate their gradients with all-one values; if the input is a
dictionary, for any dictionary entries that are not None, we'll
take the corresponding blobs as their gradients; for all those
that are None, we will auto-fill them with 1.
Returns:
    gradient_ops: a list of gradient operators to run.
    all_input_to_grads: a map from input to their corresponding
gradients.

Definition at line 1133 of file core.py.

def caffe2.python.core.GradientRegistry.RegisterGradient	(		cls,
			op_type
	)

A decorator for registering gradient mappings.

Definition at line 1060 of file core.py.

---

The documentation for this class was generated from the following file:

• caffe2/python/core.py