2.5.8. Graph “CobWeb”

For multidimensional problem, the drawPairs method is limited to spot correlation because of the way the output looks. For instance in a problem with 8 uniformly-distributed inputs and one output, one can get a graphic as the one shown in Figure 2.50 (obtained with the code below). No special trend can be seen here.

tdsCobweb = DataServer.TDataServer("tdscobweb", "Database of the cobweb")
tdsCobweb.fileDataRead("cobwebdata.dat") # read data file
tdsCobweb.drawPairs() # do the drawPairs graph

../../_images/flowrate_cobweb_pairs.png — Figure 2.50 Graphs of “drawPairs” type between the 8 uniformly-distributed inputs and the output of a given problem.

The “CobWeb” multidimensional graph, on the other hand, consists in plotting every dimension on a vertical axis and connecting all the points for a single event with a straight line. It is particularly useful to spot correlation in high-dimension problems as it is simple to highlight a certain region (by changing the colour for instance) to see if the rest of the variable are randomly distributed or not.

# Draw the cobweb plot
tdsCobweb.drawCobWeb("x0:x1:x2:x3:x4:x5:x6:x7:out") # Draw the cobweb

# Get the parallel coordination part to perform modification
para = ROOT.gPad.GetListOfPrimitives().FindObject("__tdspara__0")
# Get the output axis
axis = para.GetVarList().FindObject("out")

# Create a range for 0.97 < out < 1.0 and display it in blue
Range  = ROOT.TParallelCoordRange(axis, 0.97, 1.0)
axis.AddRange(Range)
para.AddSelection("blue")
Range.Draw()

This code for instance specifies a region-of-interest in the output, when the value are greater than 0.97. A correlation is highlighted between these values and a region, for which the fourth input variable has high value while the sixth input variable values lie between 0.2 and 0.4. The result of this code is shown in Figure 2.51.