plot_history

keras.plot_history(history, *, metrics=None)

Draw the training curves of a Keras model.

One chart per metric, stacked top to bottom, with the epoch along the x axis and two lines, named training and validation: the metric on the training data, and on the validation data when model.fit was given some. A metric with no validation values, such as the learning rate, is one line named for the metric. The loss comes first.

As in Keras, a metric whose name starts with val_ is read as the validation curve of the metric named without it.

Parameters

Name	Type	Description	Default
history	keras.callbacks.History or dict	What model.fit returns, or its .history dictionary, which maps each metric, such as "loss" or "val_accuracy", to its value at each epoch.	required
metrics	iterable of str	Only these metrics, in this order, named without the val_ prefix, such as ["loss", "accuracy"]. A metric not in history is warned about.	None

Returns

Name	Type	Description
	matplotlib.figure.Figure	The charts, ready for :func:maidr.show, :func:maidr.render or :func:maidr.save_html. The figure is not managed by pyplot, so plt.show() does not show it.

Raises

Name	Type	Description
	TypeError	If history is neither a History nor a dictionary.
	ValueError	If history holds no metric that can be drawn.

Notes

Epochs are counted from 1, as Keras counts them when it trains. A value that is not finite, such as a loss that became NaN, is a gap in the line. A metric whose value is not a single number at each epoch, such as a per-class score, is left out with a warning.

With validation_freq above 1, Keras records the validation values on fewer epochs than the training ones without saying which, so plot_history leaves them out with a warning. :class:MaidrCallback records the epoch of every value and draws them.

Examples

>>> import maidr
>>> from maidr.keras import plot_history
>>> history = {"loss": [0.9, 0.6, 0.5], "val_loss": [1.0, 0.7, 0.65]}
>>> maidr.save_html(plot_history(history), "training.html")