mxnet
|
The attributes of the current operation node. Usually are additional parameters like axis,. More...
#include <node.h>
Public Attributes | |
const Op * | op {nullptr} |
The operator this node uses. For place holder variable, op == nullptr. More... | |
std::string | name |
name of the node More... | |
std::unordered_map< std::string, std::string > | dict |
The dictionary representation of attributes. More... | |
any | parsed |
A parsed version of attributes, This is generated if OpProperty.attr_parser is registered. The object can be used to quickly access attributes. More... | |
std::vector< std::shared_ptr< Symbol > > | subgraphs |
Some operators take graphs as input. These operators include control flow operators and high-order functions. These graphs don't change when the operators are invoked for different mini-batches. In this sense, the subgraphs are kind of similar to the parameters and show be kept as node attributes. More... | |
The attributes of the current operation node. Usually are additional parameters like axis,.
std::unordered_map<std::string, std::string> nnvm::NodeAttrs::dict |
The dictionary representation of attributes.
std::string nnvm::NodeAttrs::name |
name of the node
const Op* nnvm::NodeAttrs::op {nullptr} |
The operator this node uses. For place holder variable, op == nullptr.
any nnvm::NodeAttrs::parsed |
A parsed version of attributes, This is generated if OpProperty.attr_parser is registered. The object can be used to quickly access attributes.
std::vector<std::shared_ptr<Symbol> > nnvm::NodeAttrs::subgraphs |
Some operators take graphs as input. These operators include control flow operators and high-order functions. These graphs don't change when the operators are invoked for different mini-batches. In this sense, the subgraphs are kind of similar to the parameters and show be kept as node attributes.
Users need to make sure the subgraphs are disjoint with the main graph. If a graph shares nodes with subgraphs, loading the graph from LoadJSON may generate a graph that has a different structure from the original graph (some of the nodes are duplicated). If nodes are shared between two graphs, shared nodes might be executed multiple times, which can be a problem for stateful operators.