Remove arviz as a dependency

setup.py

@@ -17,13 +17,11 @@
 REQUIRED_MINOR = 7
 
 INSTALL_REQUIRES = [
-    "arviz>=0.12.1",
     "astor>=0.7.1",
     "black==22.3.0",
     "botorch>=0.5.1",
     "gpytorch>=1.3.0, <1.9.0",
     "graphviz>=0.17",
-    "netCDF4<=1.5.8; python_version<'3.8'",
     "numpy>=1.18.1",
     "pandas>=0.24.2",
     "plotly>=2.2.1",
@@ -36,6 +34,7 @@
 ]
 TEST_REQUIRES = ["pytest>=7.0.0", "pytest-cov"]
 TUTORIALS_REQUIRES = [
+    "arviz>=0.12.1",
     "bokeh",
     "cma",
     "ipywidgets",
@@ -44,6 +43,7 @@
     "matplotlib",
     "mdformat",
     "mdformat-myst",
+    "netCDF4<=1.5.8; python_version<'3.8'",
     "scikit-learn>=1.0.0",
     "seaborn",
     "tabulate",

src/beanmachine/ppl/inference/monte_carlo_samples.py

@@ -10,6 +10,7 @@
 import xarray as xr
 from beanmachine.ppl.inference.utils import detach_samples, merge_dicts
 from beanmachine.ppl.model.rv_identifier import RVIdentifier
+from beanmachine.ppl.diagnostics.tools.viz import _requires_dev_packages
 
 
 RVDict = Dict[RVIdentifier, torch.Tensor]
@@ -266,10 +267,12 @@ def add_groups(self, mcs: "MonteCarloSamples"):
             if n not in self.namespaces:
                 self.namespaces[n] = mcs.namespaces[n]
 
-    def to_inference_data(self, include_adapt_steps: bool = False) -> az.InferenceData:
+    @_requires_dev_packages
+    def to_inference_data(self, include_adapt_steps: bool = False):
         """
         Return an az.InferenceData from MonteCarloSamples.
         """
+        import arviz as az
 
         if "posterior" in self.namespaces:
             posterior = detach_samples(self.namespaces["posterior"].samples)