Read a declared precision-recall figure: one row of points per curve.
The curve is read exactly as a line is, with the recall on x and the
precision on y; what a precision-recall figure carries beyond one comes off
each curve's own rows (the threshold) and off its own block (the
prevalence and the average precision, on its first point). A curve merged
in without a block of its own says neither, rather than borrowing the
declaring curve's, which may have been scored on different data.
The curves' points, the live marks behind them, and the series
drawing each row -- a curve with no readable mark is left out, so a row
is not always the series at the same position.
Read a declared precision-recall figure: one row of points per curve.
The curve is read exactly as a line is, with the recall on x and the precision on y; what a precision-recall figure carries beyond one comes off each curve's own rows (the threshold) and off its own block (the prevalence and the average precision, on its first point). A curve merged in without a block of its own says neither, rather than borrowing the declaring curve's, which may have been scored on different data.