model.add_coords use model.name_for

[ferrine]

when you add a coord manually into the model, you should take care to avoid name collisions with model.name_for("coord") helper. I propose to implement it in the same way we register coords, appending name prefix for convenience.

def lr(data):
    m = pm.modelcontext(None)
    m.add_coord(m.name_for("coefs"), data.columns)
    alpha = pm.Normal("alpha", sigma=1, dims=m.name_for("coefs"))
    intercept = pm.HalfCauchy("intercept", 1)
    eps = pm.HalfNormal("sigma", 1)
    y = pm.Normal("y", data.values @ alpha + intercept)
    
def lr1(data):
    pm.modelcontext(None).add_coord("coefs", data.columns)
    alpha = pm.Normal("alpha", sigma=1, dims="coefs")
    intercept = pm.HalfCauchy("intercept", 1)
    eps = pm.HalfNormal("sigma", 1)
    y = pm.Normal("y", data.values @ alpha + intercept)


with pm.Model() as model:
    with pm.Model(name="lr1") as m1:
        lr1(data)
    with pm.Model(name="lr2") as m1:
        lr1(data.iloc[:, :3])

while this is possible to correctly set all the coords

with pm.Model() as model:
    with pm.Model(name="lr1") as m1:
        lr(data)
    with pm.Model(name="lr2") as m1:
        lr(data.iloc[:, :3])

In the proposal, dims passed to the rv are used in this order:

1. check self.name_for("coord") exists, use this if true
2. use "coord"

[ferrine]

@ricardoV94 Could you please have a look on the proposal?