Plotting API Overview#

This logic is encapsulated into a display object where the computed data isstored and the plotting is done in aplot method. The display object’s__init__ method contains only the data needed to create the visualization.Theplot method takes in parameters that only have to do with visualization,such as a matplotlib axes. Theplot method will store the matplotlib artistsas attributes allowing for style adjustments through the display object. TheDisplay class should define one or both class methods:from_estimator andfrom_predictions. These methods allow creating theDisplay object fromthe estimator and some data or from the true and predicted values. After theseclass methods create the display object with the computed values, then call thedisplay’s plot method. Note that theplot method defines attributes relatedto matplotlib, such as the line artist. This allows for customizations aftercalling theplot method.

For example, theRocCurveDisplay defines the following methods andattributes:

classRocCurveDisplay:def__init__(self,fpr,tpr,roc_auc,estimator_name):...self.fpr=fprself.tpr=tprself.roc_auc=roc_aucself.estimator_name=estimator_name@classmethoddeffrom_estimator(cls,estimator,X,y):# get the predictionsy_pred=estimator.predict_proba(X)[:,1]returncls.from_predictions(y,y_pred,estimator.__class__.__name__)@classmethoddeffrom_predictions(cls,y,y_pred,estimator_name):# do ROC computation from y and y_predfpr,tpr,roc_auc=...viz=RocCurveDisplay(fpr,tpr,roc_auc,estimator_name)returnviz.plot()defplot(self,ax=None,name=None,**kwargs):...self.line_=...self.ax_=axself.figure_=ax.figure_

Read more inROC Curve with Visualization APIand theUser Guide.

Plotting with Multiple Axes#

Some of the plotting tools likefrom_estimator andPartialDependenceDisplay support plotting onmultiple axes. Two different scenarios are supported:

1. If a list of axes is passed in,plot will check if the number of axes isconsistent with the number of axes it expects and then draws on those axes. 2.If a single axes is passed in, that axes defines a space for multiple axes tobe placed. In this case, we suggest using matplotlib’s~matplotlib.gridspec.GridSpecFromSubplotSpec to split up the space:

importmatplotlib.pyplotaspltfrommatplotlib.gridspecimportGridSpecFromSubplotSpecfig,ax=plt.subplots()gs=GridSpecFromSubplotSpec(2,2,subplot_spec=ax.get_subplotspec())ax_top_left=fig.add_subplot(gs[0,0])ax_top_right=fig.add_subplot(gs[0,1])ax_bottom=fig.add_subplot(gs[1,:])

By default, theax keyword inplot isNone. In this case, the singleaxes is created and the gridspec api is used to create the regions to plot in.

See for example,from_estimatorwhich plots multiple lines and contours using this API. The axes defining thebounding box are saved in abounding_ax_ attribute. The individual axescreated are stored in anaxes_ ndarray, corresponding to the axes position onthe grid. Positions that are not used are set toNone. Furthermore, thematplotlib Artists are stored inlines_ andcontours_ where the key is theposition on the grid. When a list of axes is passed in, theaxes_,lines_,andcontours_ are a 1d ndarray corresponding to the list of axes passed in.