qkeras.qtools.qgraph

Creates networkx graph from a model.

Functions

AddToNodeDict(layer_items, layer, nodes_dict)	Adds layer to a node_dict, indexed by layer.(input or output).ref
CreateGraph(model[, input_quantizers, ...])	create graph.
GenerateGraphFromModel(model, ...)	Generates single source, single sink graph from model.
GenerateInputQuantizerList(input_quantizers, ...)	Generates the list of input quantizers.
GraphAddHiddenInputLayer(model, graph, ...)	For Keras Sequential model api, input layer is hidden.
GraphAddSingleSourceSingleSink(graph)	Connects graph to source and sink nodes.
GraphGetInputs(graph)	Returns edges SOURCE->u that are inputs.
GraphGetOutputs(graph)	Returns edges u->SINK that are outputs.
GraphPropagateActivationsToEdges(graph[, debug])	Traverses graph and move activations to edges.
GraphRemoveNode(graph, v)	Removes node "v" from u -> v -> w, connecting u -> w.
GraphRemoveNodeWithNodeType(graph, node_type)	Removes node with attribute node_type, reconnecting network.
GraphUpdateEdge(graph, node_id, ...)	update the graph edges outgoing from node_id with new quantizer.
PrintGraph(graph[, msg])	Print graph structure.

Exceptions

WrongInputQuantizerError

qkeras.qtools.qgraph.AddToNodeDict(layer_items, layer, nodes_dict)

Adds layer to a node_dict, indexed by layer.(input or output).ref

qkeras.qtools.qgraph.CreateGraph(model, input_quantizers=None, default_source_quantizer='quantized_bits(8, 0, 1)', debug=False)

create graph.

qkeras.qtools.qgraph.GenerateGraphFromModel(model, input_quantizers, default_source_quantizer)

Generates single source, single sink graph from model.

qkeras.qtools.qgraph.GenerateInputQuantizerList(input_quantizers, inputs_length, default_source_quantizer)

Generates the list of input quantizers.

qkeras.qtools.qgraph.GraphAddHiddenInputLayer(model, graph, input_quantizer_map)

For Keras Sequential model api, input layer is hidden. Need to add it.

qkeras.qtools.qgraph.GraphAddSingleSourceSingleSink(graph)

Connects graph to source and sink nodes.

qkeras.qtools.qgraph.GraphGetInputs(graph)

Returns edges SOURCE->u that are inputs.

qkeras.qtools.qgraph.GraphGetOutputs(graph)

Returns edges u->SINK that are outputs.

qkeras.qtools.qgraph.GraphPropagateActivationsToEdges(graph, debug=False)

Traverses graph and move activations to edges.

1.If current dense/conv layer is specified with QA:

outgoing edge (output data type) will be QA type

2.If current dense/conv layer has no QA:

default type (float32) is used as output

3.If current layer is QA layer:

float32 is used by default as output type on the edge

Parameters:

• graph – graph to inject activations to.

• debug – debug mode

Returns:

None

qkeras.qtools.qgraph.GraphRemoveNode(graph, v)

Removes node “v” from u -> v -> w, connecting u -> w.

qkeras.qtools.qgraph.GraphRemoveNodeWithNodeType(graph, node_type)

Removes node with attribute node_type, reconnecting network.

qkeras.qtools.qgraph.GraphUpdateEdge(graph, node_id, quantizer_on_edge)

update the graph edges outgoing from node_id with new quantizer.

qkeras.qtools.qgraph.PrintGraph(graph, msg='')

Print graph structure.

exception qkeras.qtools.qgraph.WrongInputQuantizerError

Bases: ValueError