Inverted shape in heatmap matrix visualization about fanova HOT 6 CLOSED

I think this depends on the cryptic use of this swap-dynamic (which I implemented based on how I found the code before. Basically this depends on the question whether fANOVA in any way depends on the order being based on the length of the choices. Or why this was designed this way. I suppose dropping the "swap"-block would alleviate this more in the root of the problem. @AndreBiedenkapp ?

from fanova.

I think it was designed in such a way because of axis-labels potentially becoming too squeezed.
But I agree @shukon that this could be dropped.

from fanova.

Hi @FlorentF9 . I tried to fix this in https://github.com/automl/fanova/tree/issue/96-transposed-grid-categoricals , could you maybe check with your data and give feedback?

from fanova.

Branch is merged and released as fanova 2.0.16.

from fanova.

Heatmap axes are still inverted. The reason is simple actually: the generate_pairwise_marginal method returns a matrix zz of shape (R, C) with R the number of rows (= number of values of the first param) and C the number of columns (= number of values of the second param).
Hence the matrix displayed by plt.imshow has an x-axis of size C (columns) and y-axis of size R (columns), so plt.xticks and plt.yticks are inverted in the code (https://github.com/automl/fanova/blob/master/fanova/visualizer.py#L222). Solution is to transpose zz or invert xticks and yticks.

from fanova.

🤦 I'm sorry. I think I changed most of the class in style (and actually fixed another bug for the swap-thing), but left the actual issue untouched. Thank you for your patience in pointing this out to me 😄
I decided to invert ticks and label for clarity.

from fanova.