calour.training.plot_roc¶

calour.training.plot_roc(result, classes=None, title='ROC', cv=True, cmap=None, ax=None)[source]¶

Plot ROC curve.

Note

You may want to consider using precision-recall curve (:func:plot_prc) instead of ROC curve. If your model needs to perform equally well on the negative class as the positive class, you would use the ROC AUC. For example, for classifying images between cats and dogs, if you would like the model to perform well on the cats as well as on the dogs, then you can use ROC and ROC is more intuitive to understand. If you have imbalanced classes OR you don’t care about the negative class at all, then you should use precision-recall curve. Take cancer diagnosis as an example, you tends to have way more negative samples than positive cancer samples and want to make sure you positive predictions are correct (precision) and don’t miss any cancer (recall or aka sensitivity), you should use precision-recall curve. If the classes are balanced, ROC usually works fine even in this scenario. [1].

Parameters:	result (pandas.DataFrame) – data frame containing predictions per sample (in row). It must have a column of true class named “Y_TRUE” and multiple columns of predicted probabilities for each class. It typically takes the output of classify(). classes (list) – The list of the labels you want to include in the plot in the order specified in the list. If it is a binary classification (eg “Health” vs. “IBD”), you would want to set it to [“IBD”] and don’t plot for “Health” because it is equivalent. title (str) – plot title cv (boolean) – Whether to plot ROC curve shade by different folds of cross validation. You need to have ‘CV’ column in the input result data frame. cmap (str or matplotlib.colors.ListedColormap) – str to indicate the colormap name. Default is “Dark2” colormap. For all available colormaps in matplotlib: https://matplotlib.org/users/colormaps.html ax (matplotlib.axes.Axes or None (default), optional) – The axes where to plot. None (default) to create a new figure and axes to plot
Returns:	The axes for the ROC and the AUC value
Return type:	tuple of matplotlib.axes.Axes and float

References

[1]	Saito T and Rehmsmeier M (2015) The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets. PLoS One, 10.