read_tensorboard_graph

read_tensorboard_graph(logdir, *, runs=None, expand_nested=False)

Read the Keras model graphs of a TensorBoard log directory.

One chart per run that logged a model, as TensorBoard’s Graphs dashboard draws it under the Keras tag: one box per layer, top to bottom from the inputs, with an arrow from each layer to the layers called on its output.

Parameters

Name	Type	Description	Default
logdir	str or os.PathLike	The directory TensorBoard would be started with (--logdir), such as the log_dir of Keras’s TensorBoard callback with write_graph=True, its default.	required
runs	iterable of str	Only these runs. Keras logs the model in its train run.	None
expand_nested	bool	Whether a model used as a layer is drawn as its own layers, in a scope named for it, rather than as one box.	False

Returns

Name	Type	Description
	list of TensorBoardChart	One per run, tagged keras. Pass one to :func:maidr.show, :func:maidr.render or :func:maidr.save_html.

Raises

Name	Type	Description
	FileNotFoundError	If logdir is not a directory.

Notes

The chart is a directed_graph, an experimental type that needs a maidr.js release carrying it. A run that logged its model more than once is drawn from the last. A model logged as something other than a Keras config, such as a TensorFlow op graph, is not read.

Examples

>>> import maidr
>>> (graph,) = maidr.read_tensorboard_graph("logs/fit")
>>> maidr.save_html(graph, "model.html")