mlair.plotting.training_monitoring
¶
Plots to monitor training.
Module Contents¶
Classes¶
Plot history of all plot_metrics (default: loss) for a training event. |
|
Plot the behaviour of the learning rate in dependence of the number of epochs. |
Attributes¶
-
mlair.plotting.training_monitoring.
__date__
= 2019-12-11¶
-
mlair.plotting.training_monitoring.
history_object
¶
-
mlair.plotting.training_monitoring.
lr_object
¶
-
class
mlair.plotting.training_monitoring.
PlotModelHistory
(filename: str, history: history_object, plot_metric: str = 'loss', main_branch: bool = False, epoch_best: int = None)¶ Plot history of all plot_metrics (default: loss) for a training event.
For default plot_metric and val_plot_metric are plotted. If further metrics are provided (name must somehow include the word <plot_metric>), this additional information is added to the plot with an separate y-axis scale on the right side (shared for all additional metrics). The plot is saved locally. For a proper saving behaviour, the parameter filename must include the absolute path for the plot.
-
_get_plot_metric
(self, history, plot_metric, main_branch, correct_names=True)¶
-
_filter_columns
(self, history: Dict) → List[str]¶ Select only columns named like %<plot_metric>%.
The default metrics ‘<plot_metric>’ and ‘val_<plot_metric>’ are removed too.
- Parameters
history – a dict with at least ‘<plot_metric>’ and ‘val_<plot_metric>’ as keys (can be derived from keras History.history)
- Returns
filtered columns including all plot_metric variations except <plot_metric> and val_<plot_metric>.
-
_plot
(self, filename: str) → None¶ Create plot.
Plots <plot_metric> and val_<plot_metric> as default. If more plot_metrics are provided, they will be added with an additional yaxis on the right side. The plot is saved in filename.
- Parameters
filename – name (including total path) of the plot to save.
-
-
class
mlair.plotting.training_monitoring.
PlotModelLearningRate
(filename: str, lr_sc: lr_object)¶ Plot the behaviour of the learning rate in dependence of the number of epochs.
The plot is saved locally as pdf. For a proper saving behaviour, the parameter filename must include the absolute path for the plot.