You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The check_partials tool coupled with partial_deriv_plot has been really useful in debugging some components and learning how to set up sparse partials.
One of the plots it produces is a map of the difference between J_fd and J_fwd. Right now it uses a diverging colormap, but it's not used quite properly. This sort of diverging colormap is designed for data centered around zero, where values of zero (meaning success in this application!) are gray, low values are blue, and high values are red.
Right now, the colormap has limits (probably set with a vmin and vmax) which are something like vmin=min(data), vmax=max(data). But, if positive errors are larger than negative errors, for example, then zero values are not gray, and the whole plot is tinted one way or the other.
One solution would be to precompute z = max(abs(min(data)), abs(max(data))) and set vmin=-z, vmax=z to force a centered color bar.
Associated POEM
No response
The text was updated successfully, but these errors were encountered:
Desired capability or behavior.
The
check_partials
tool coupled withpartial_deriv_plot
has been really useful in debugging some components and learning how to set up sparse partials.One of the plots it produces is a map of the difference between
J_fd
andJ_fwd
. Right now it uses a diverging colormap, but it's not used quite properly. This sort of diverging colormap is designed for data centered around zero, where values of zero (meaning success in this application!) are gray, low values are blue, and high values are red.Right now, the colormap has limits (probably set with a
vmin
andvmax
) which are something likevmin=min(data), vmax=max(data)
. But, if positive errors are larger than negative errors, for example, then zero values are not gray, and the whole plot is tinted one way or the other.One solution would be to precompute
z = max(abs(min(data)), abs(max(data)))
and setvmin=-z, vmax=z
to force a centered color bar.Associated POEM
No response
The text was updated successfully, but these errors were encountered: