README
¶
Operators
This directory holds the operators that can be promoted and used in gorgonnx.
All the operators fulfils the Operator interface{}
Add a new operator
If you want to add an operator, be sure to read its documentation from the ONNX web page.
Then generate a skeleton of implementation and the unit tests with
./op.sh operator # operator is lowercase
This will generate two files:
operator_autogenerated_tests.gooperator_todo.go
Tests
The file operator_autogenerated_tests.go is a pure go representation of the onnx tests related to the operator.
if you run
go test -run=Operator
you should see that all the tests are skipped because by default the Init and Apply methods return an ErrNotImplemented error.
Implementation
a good practise is to rename the file from _todo.go to _wip.go until the implementation is ok.
Init
You should start by implementing the Init method.
You can rely on the UnmarshalAttributes method of the onnx-go package.
Read the onnx documentation of the package and create an inner-type that will hold the values.
For example, if Operator expects an attribute named myattribute of type int with a defaut value of 42, just create
type attributes struct {
MyAttr int64 `attributeName:"myattribute"`
}
// Create an instance with default values
attr := attributes{
MyAttr: 42,
}
err := onnx.UnmarshalAttributes(attrs, &attr)
Apply
The apply method is straightforward. See the Add Operator for an example.
Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Add ¶
type Add struct {
// contains filtered or unexported fields
}
Add operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Add Warning this operation is broadcastable See https://github.com/onnx/onnx/blob/master/docs/Broadcasting.md
type Batchnorm ¶
type Batchnorm struct {
Epsilon float64
Momentum float64
Spatial int
// contains filtered or unexported fields
}
Batchnorm operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#BatchNormalization
type Concat ¶
type Concat struct {
Axis int
// contains filtered or unexported fields
}
Concat operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Concat
type Constant ¶
Constant operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Constant
type Conv ¶
type Conv struct {
AutoPad string
Pads []int
Dilations []int
Group int
KernelShape tensor.Shape
Strides []int
// contains filtered or unexported fields
}
Conv operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Conv
For more information about convolution, please visit https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md
type Div ¶
type Div struct {
// contains filtered or unexported fields
}
Div operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Div
type Dropout ¶
type Dropout struct {
// contains filtered or unexported fields
}
Dropout operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Dropout
type ErrBadArity ¶
type ErrBadArity struct {
Operator string
ExpectedInput int
ActualInput int
ExpectedOutput int
ActualOutput int
}
ErrBadArity is raised when an operator do not have the correct amount of input or output
func (*ErrBadArity) Error ¶
func (e *ErrBadArity) Error() string
type Matmul ¶
type Matmul struct {
// contains filtered or unexported fields
}
Matmul operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Matmul
type Maxpool ¶
type Maxpool struct {
Pads []int
AutoPad string
StorageOrder int
KernelShape tensor.Shape
Strides []int
// contains filtered or unexported fields
}
Maxpool operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#MaxPool
type Mul ¶
type Mul struct {
// contains filtered or unexported fields
}
Mul operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Mul
type Relu ¶
type Relu struct {
// contains filtered or unexported fields
}
Relu operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Relu
type Reshape ¶
type Reshape struct {
// contains filtered or unexported fields
}
Reshape operator https://github.com/onnx/onnx/blob/master/docs/Operators.md#Reshape
Source Files