🚚 Move diagnostics into separate files

Author: francois-rozet

docs/tutorials/coverage.ipynb

@@ -170,7 +170,7 @@
    "source": [
     "## Expected coverage\n",
     "\n",
-    "Using the [`expected_coverage_mc`](lampe.diagnostics.expected_coverage_mc) and [`expected_coverage_ni`](lampe.diagnostics.expected_coverage_ni) functions provided by the [`lampe.diagnostics`](lampe.diagnostics) module, we estimate the respective expected coverage of the three estimators. The expected coverage at credible level $1 - \\alpha$ is the probability of a set of parameters $\\theta^*$ to fall inside the highest posterior density \n",
+    "Using the [`expected_coverage_mc`](lampe.diagnostics.coverage.expected_coverage_mc) and [`expected_coverage_ni`](lampe.diagnostics.coverage.expected_coverage_ni) functions provided by the [`lampe.diagnostics`](lampe.diagnostics) module, we estimate the respective expected coverage of the three estimators. The expected coverage at credible level $1 - \\alpha$ is the probability of a set of parameters $\\theta^*$ to fall inside the highest posterior density \n",
     "\n",
     "The expected coverage of a posterior estimator $q(\\theta | x)$ at credible level $1 - \\alpha$ is the probability of a set of parameters $\\theta^*$ to be included in the highest density region of total probability $1 - \\alpha$ of the posterior estimator, given an observation $x^* \\sim p(x | \\theta^*)$. If the estimator is callibrated, the expected coverage at credible level $1 - \\alpha$ is exactly $1 - \\alpha$. If it is overconfident, the highest density regions are smaller than they should be, which translates into lower coverages. If it is conservative/underconfident, the highest density regions are larger than they should be and the expected coverages are larger than their nominal level."
    ]

lampe/diagnostics/__init__.py

@@ -0,0 +1,3 @@
+r"""Diagnostics and reliability assessment."""
+
+from .coverage import *

lampe/diagnostics.py renamed to lampe/diagnostics/coverage.py

@@ -1,4 +1,4 @@
-r"""Diagnostics and reliability assessment."""
+r"""Coverage diagnostics."""
 
 import torch
 
@@ -8,7 +8,7 @@
 from typing import Callable, Iterable, Tuple
 
 # isort: local
-from .utils import gridapply
+from ..utils import gridapply
 
 
 def expected_coverage_mc(