from torch import rand, randint
from torchmetrics.classification import MulticlassPrecisionAtFixedRecall
metric = MulticlassPrecisionAtFixedRecall(num_classes=3, min_recall=0.5)
metric.update(rand(20, 3).softmax(dim=-1), randint(3, (20,)))
fig_, ax_ = metric.plot()  # the returned plot only shows the maximum recall value by default