Remove an edge from the graph and add annotations to both ends of the edge.
Source name.
Sink name.
Remove edges and add to annotation instead.
For root node, consider predefined types for source and sink. We do not extract predefined type from non-root so that Variables and the sgd node (op type = 'NoOp') do not get extract from inside own group.
The order of extraction is important here as swapping the order can totally screw up the graph layout.
Node to manipulate.
Extract nodes deemed to have either high in-degree or high out-degree.
Remove edges from pre-defined out-extract patterns
Remove edges from pre-defined in-extract patterns
Move nodes that are specified to be excluded out of the core graph.
Check whether the node's type is a member of the given list of types.
Node.
List of type to match.
Remove edges from a node, set its isInExtract property to true, and remove the node and move it to isolatedInExtract.
If detachAllEdgesForHighDegree or forceDetach is true, extract all of its edges. Otherwise, only remove all out-edges.
Remove edges from a node, and set its isOutExtract property to true, and remove the node and move it to isolatedOutExtract.
If detachAllEdgesForHighDegree or forceDetach is true, extract all of its edges. Otherwise, only extract all in-edges.
Given an integer, picks a hue that is far apart from other colors. The formula for picking color that avoid collision is: hue = (color range golden ratio index) % color range
Remove control edges from nodes that have too many control edges
Color parameters for node encoding.
Default fill and stroke to use when no other information is available.
Neutral color to use when the node is expanded.
Standard hue values for node color palette.
Color parameters for op nodes.
Parameters that affect how the graph is rendered on the screen.
When removing edges from a high degree node, remove all of its edges if detachAllEdgesForHighDegree is true. Otherwise remove all in-edges if the node has high in-degree, or all out-edges if the node has high out-degree.
Whether to add bridge nodes and edges to the core when building the subhierarchy of an expanded metanode. See buildSubhierarchy().
Whether to extract high degree nodes from the core part of the graph.
After extracting high in/out degree nodes and predefined source-like/sink-like, extract isolated nodes to the side if this extractIsolatedNodesWithAnnotationsOnOneSide is true.
Types patterns for predefined in-extract nodes, which are source-like nodes that will be extracted from the main graph.
Maximum number of annotations to be displayed on a node before an ellipsis is used.
Maximum in (for outbound bridge paths) or out (for inbound bridge paths) degree of a node allowed for a bridge path to be rendered to it from a subhierarchy of nodes. Having a max prevents having too many nodes emanate from a subhierarchy and crowding up.
Maximum number of control edges a node can have before they aren't displayed.
The minimum in or out degree a node must have in order to be possibly extracted.
2 colors, for the minimum and maximum value respectively, whenever we have a gradient scale.
The minimum number of nodes for a graph to have in order for high in and out degree nodes to be extracted in auxiliary. The aim here is to prevent nodes from being extracted from small graphs.
Types patterns for predefined out-extract nodes, which are sink-like nodes that will be extracted from the main graph.
Color parameters for op nodes.
Generated using TypeDoc
Package for the Render Hierarchy for TensorFlow graph.