Method _create_conv

python/singa/sonnx.py:1638–1675

get the clip operator from onnx node Args: onnx_node (OnnxNode): a given onnx node operator (Operator Class): a singa operator class opset_version (int): the opset version Returns: singa operator instance

(cls, onnx_node, operator, opset_version=_opset_version)

    @classmethod
    def _create_conv(cls, onnx_node, operator, opset_version=_opset_version):
        """
        get the clip operator from onnx node
        Args:
            onnx_node (OnnxNode): a given onnx node
            operator (Operator Class): a singa operator class
            opset_version (int): the opset version
        Returns: 
            singa operator instance
        """
        kernel_size = tuple(onnx_node.getattr('kernel_shape'))
        padding = tuple(onnx_node.getattr('pads', (0, 0)))
        stride = tuple(onnx_node.getattr('strides', (1, 1)))
        auto_pad = utils.force_unicode(onnx_node.getattr('auto_pad', 'NOTSET'))

        # not support dilation
        dilation = onnx_node.getattr('dilations', 1)
        if dilation != 1 and list(dilation) != [1, 1]:
            raise ValueError("Not implemented yet for dilation")
        group = onnx_node.getattr('group', 1)

        # only support 1d or 2d
        if len(kernel_size) > 2:
            raise ValueError("Only implemented for 1d or 2d")

        onnx_node.set_weight_inputs(onnx_node.inputs[1], 'W')
        bias = False
        if len(onnx_node.inputs) == 3:
            onnx_node.set_weight_inputs(onnx_node.inputs[2], 'b')
            bias = True
        return operator(None,
                        kernel_size,
                        stride=stride,
                        padding=padding,
                        dilation=dilation,
                        group=group,
                        bias=bias,
                        pad_mode=auto_pad)

    @classmethod
    def _create_max_avg_pool(cls,