Odd looking radar plot in the interpretability notebook

[t-sanchez]

Hello everyone,

When reading through the interpretability of IQMs notebook (which is very insightful and nice to read!), I've been puzzled by the last radar plot (image below).

The red curve of the left plot looks oddly shifted to me. Shouldn't it go around/touch the center of the plot?

[neurorepro]

@t-sanchez the radar plot is a collection of (angularly) equidistant radii. Each feature is associated to a radius, with:

• the value shown along the radius (the further from the center the higher the value)
• the label shown at the end of the radius

The plot you see is from connecting all the values with a straight line, and there is no reason for them to touch the center or go around it. Just geometrically, the plot corresponding to the most features (i.e. more than half the features) will go around the center because the lines will go around more than 180 degrees.

[t-sanchez]

Hey @neurorepro,
Thanks for taking the time to reply :)

I'm not sure to understand your entire reply. What do you mean by

the plot corresponding to the most features (i.e. more than half the features) will go around the center because the lines will go around more than 180 degrees.

I'm still a bit confused, so let me specify my issue. For instance, I'm not really sure how to read the feature values for the red curve, because sometimes, it will have to intersections, and for other features, there will be no intersection with the red line. In comparison, this reads clearly on the second plot

[neurorepro]

Hello @t-sanchez , features either have a positive or a negative sign, i.e. correspond either to the blue or red curve, so in the case "no value here" it is because it actually corresponds to a feature with a negative sign, i.e. belonging to the blue curve and not the red curve. Not sure if it is clearer ?