Error computing posterior predictive check calibration_error

[jpintar]

Using scvi.criticism, I can successfully perform all available posterior predictive checks on my scANVI model (coefficient_of_variation, zero_fraction, differential_expression) except for calibration_error.

When I try

from scvi.criticism import PosteriorPredictiveCheck
ppc = PosteriorPredictiveCheck(
    adata,
    models_dict={'scANVI': scanvi_model},
    count_layer_key='raw',
    n_samples=10,
)
ppc.calibration_error()

I get the following error:

AttributeError                            Traceback (most recent call last)
Cell In[90], line 1
----> 1 ppc.calibration_error()

File [/<envpath>/lib/python3.10/site-packages/scvi/criticism/_ppc.py:240](http://localhost:9998/<envpath>/python3.10/site-packages/scvi/criticism/_ppc.py#line=239), in PosteriorPredictiveCheck.calibration_error(self, confidence_intervals)
    238 end = interval[1]
    239 true_width = ps[end] - ps[start]
--> 240 greater_than = (quants[DATA_VAR_RAW] >= quants.model1.isel(quantile=start)).data
    241 less_than = (quants[DATA_VAR_RAW] <= quants.model1.isel(quantile=end)).data
    242 # Logical and

File [/<envpath>/lib/python3.10/site-packages/xarray/core/common.py:286](http://localhost:9998/<envpath>/lib/python3.10/site-packages/xarray/core/common.py#line=285), in AttrAccessMixin.__getattr__(self, name)
    284         with suppress(KeyError):
    285             return source[name]
--> 286 raise AttributeError(
    287     f"{type(self).__name__!r} object has no attribute {name!r}"
    288 )

AttributeError: 'Dataset' object has no attribute 'model1'

Versions:

scvi-tools 1.1.1
xarray 2024.2.0

[canergen]

This function is currently not implemented. It exists in the code base but was never used. We are currently reworking the criticism module and will likely remove this metric in the future package.