imshow: alternative to heatmap for easy facetgrid implementation?

[coxipi]

I was trying to use fg.heatmap in the same way as fg.gridmap, supplying plot_kw = dict(row="experiment", col="period"), but it seems in this case we need a two-dimensional dataset, already prepared for a sns.heatmap call, correct? Is the idea of a grid of heatmaps something that was of interest for other people? How do we proceed with figanos in this case?

An alternative can be to work: xarray.DataArray.imshow. I don't know if we lose many useful features from sns.heatmap, but with just a few lines of code I was already satisfied with the result:

# I use "step" instead of "experiment", but I believe this kind of heatmap grid 
# could be of general interest when multiple experiements SSPs are involved
fig = ds.plot.imshow(**{"col":"period", "row":"step", "cmap":"RdBu"}, figsize=(12,9)) 
axs = np.array(fig.axes).flatten()
for ax in axs: 
    ax.set(**dict(xlabel="", ylabel=""))
for ax in axs[-2:]: # bottom two axes
    ax.tick_params(axis='x', rotation=45))
    ax.set_xticklabels(ax.get_xticklabels(), ha="right")
fig.cbar.set_label('Fraction of improved grid cells')

(in the bottom plot, I am missing data, tha's why there are blank lines)

[coxipi]

I realize fg.gridmap may already cover my use case fg.gridmap(ds, plot_kw=dict(row="step", col="period")), but I'm getting

UFuncTypeError: ufunc 'subtract' did not contain a loop with signature matching types (dtype('<U15'), dtype('<U15')) -> None

I'm investigating...

EDIT: Replacing all my dimensions on ticks by integers, I can get no error and this:

ds2 = ds.copy()
for d in [ "bc", "multivar"]: 
    ds2[d] = np.arange(ds2[d].size)
fg.gridmap(ds2, plot_kw = dict(col="period", row="step",figsize=(14,6)),divergent=0.5)

So I don't think this use-case was planned for, correct?