Duncan Watson-Parris

Hi Brian, thanks for this wonderful tool and resource! I've been trying to setup a 1d model with latitudinally varying heat capacity and meridional moist diffusion but can't make it...

Sure: ``` model = climlab.EBM_seasonal() # Add a hemisphericaly asymetric heat-capacity model.Ts.domain.heat_capacity = model.Ts.domain.heat_capacity * (np.heaviside(-model.domains['Ts'].lat.points[:, np.newaxis], 0.5) + 1.) d = climlab.dynamics.MeridionalMoistDiffusion(state={'Ts': model.state['Ts']}, **model.param) model.add_subprocess('Diffusion', d) model.integrate_years(8) ``` This...

OK, I'll take a look - thanks! Presumably the diffusivity should be zero at the outermost boundaries (at the poles) so that interpolation on the internal bounds should work?

Nice, thanks! I recently factored the kernel creation in to a separate function (`_get_gpflow_kernel`), could you use that instead of replicating it here? I'm a little uncomfortable with the lack...

I like the idea but I'm not a big fan of optionally returning a different number of arguments, it can make code a bit hard to visually debug. Looking at...

Thanks! Currently there isn't a way to do this, no. It seems a very sensible thing to allow though. I think it would require each type of model (GPFlow, sckit-learn...

Ah OK, that's because of some of the `tensorflow` functions on the `Emulator`... This will need a bit more thought sorry.

Yes, you're absolutely right. It's also possible (but a little convoluted currently) to wrap the loaded model back in to ESEm: ```python # Save the sklearn model held internally in...

* How to comment code, inline and doc-strings

"Mostly good. Maybe write a program to evaluate some data that we keep returning to rather than separate examples" "Pros: * Well thought out, very useful content * Pitched at...