Scale has no effect for Pair plots in next-gen interface

[eringrant]

The objects interface looks great, in particular the ability to have combination pair and facet plots. However, I ran into a bug when attempting to log-scale an axis for a paired plot with seaborn==0.12.0b2.

Expected behavior (no log-scaling where none applied):

import seaborn as sns
import seaborn.objects as so
diamonds = sns.load_dataset("diamonds")

(
    so.Plot(
        diamonds, 
        x="carat", 
    )
    .pair(y=(
        "price", 
        "price", 
    ))
    .add(so.Scatter())
)

Unexpected behavior (no log-scaling even when calling scale(y="log")):

import seaborn as sns
import seaborn.objects as so
diamonds = sns.load_dataset("diamonds")

(
    so.Plot(
        diamonds, 
        x="carat", 
    )
    .pair(y=(
        "price", 
        "price", 
    ))
    .scale(y="log")
    .add(so.Scatter())
)

Expected behavior (log-scaling works without pairing):

import seaborn as sns
import seaborn.objects as so
diamonds = sns.load_dataset("diamonds")

(
    so.Plot(
        diamonds, 
        x="carat", 
        y="price"
    )
    .scale(y="log")
    .add(so.Scatter())
)

[mwaskom]

Hi @eringrant thank you so much for testing out the beta!

This might not be documented (yet), but individual variables in a paired plot can be scaled (and otherwise modified in Plot.limit, Plot.label, etc.) with {var}{idx}, i.e.:

(
    so.Plot(
        diamonds, 
        x="carat", 
    )
    .pair(y=(
        "price", 
        "price", 
    ))
    .add(so.Scatter())
    .scale(y1="log")
)

Then there's a separate question of "if there are multiple y axes, should doing .scale(y="log") set a log scale on all of them?" I could imagine that being useful! Is that what you were expecting here?

[eringrant]

Thanks, @mwaskom—being able to independently scale the y-axes by indexing them in this way is great. I didn't see that in the documentation, but perhaps I missed it.

Yes, I'd expect .scale(y="log") to scale all the axes! (And for that to be incompatible with scaling them independently as y1="log", etc.)

[beyarkay]

Chiming in to say I got confused by this behaviour as well, but the solution (y1="log") works well for me