Formatting fix in notebook

Author: han-ol

examples/Lotka_Volterra_point_estimation_and_expert_stats.ipynb

@@ -554,16 +554,14 @@
     "\n",
     "* To estimate **quantiles**, the following is a strictly proper scoring rule:\n",
     "$$L(\\hat \\theta, \\theta; \\tau) = (\\hat \\theta - \\theta)(\\mathbf{1}_{\\hat \\theta - \\theta > 0} - \\tau)$$\n",
-    "Here we write an indicator function as $\\mathbf{1}_{\\hat \\theta - \\theta > 0}$ to evaluate to 1 for overestimation (positive $\\hat \\theta - \\theta$) and $0$ otherwise.\n",
     "\n",
-    "    For $\\tau=\\frac 1 2$, over- or underestimating a true posterior sample $\\theta$ is weighted equally. In fact, the quantile loss with $\\tau=\\frac 1 2$ is identical to the median loss (up to a scaling of $\\frac 1 2$). For the same reasons, both estimate the median of the posterior.\n",
+    "    Here we write an indicator function as $\\mathbf{1}_{\\hat \\theta - \\theta > 0}$ to evaluate to 1 for overestimation (positive $\\hat \\theta - \\theta$) and $0$ otherwise.\n",
     "\n",
+    "    For $\\tau=\\frac 1 2$, over- or underestimating a true posterior sample $\\theta$ is weighted equally. In fact, the quantile loss with $\\tau=\\frac 1 2$ is identical to the median loss (up to a scaling of $\\frac 1 2$). For the same reasons, both estimate the median of the posterior.\n",
     "\n",
     "    More generally, $\\tau \\in (0,1)$ is the quantile level, that is the point where to evaluate the [quantile function](https://en.wikipedia.org/wiki/Quantile_function).\n",
     "\n",
     "\n",
-    "\n",
-    "\n",
     "* Note, that when approximating the full distribution in BayesFlow we score a **probability estimate** $\\hat p(\\theta|x)$ with the log-score,\n",
     "$$L(\\hat p(\\theta|x), \\theta) = \\log (\\hat p(\\theta)) $$\n",
     "which is also a strictly proper scoring rule.\n",
@@ -791,16 +789,6 @@
     "Just for fun and because we can, let us save the trained point approximator to disk."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 20,
-   "id": "0de263ba-b9a9-4aca-bf8d-0b01b18ef4e8",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "point_inference_workflow.approximator.build_from_data(adapter(training_data))"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 21,
@@ -854,27 +842,6 @@
     "Since one point estimate already summarizes many posterior samples, we only have to do one forward pass with a point inference network, where we would have to make ~100 passes with a generative, full posterior approximator."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 23,
-   "id": "2f3833f9-a155-49aa-9e0a-d1b264c72fda",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "False"
-      ]
-     },
-     "execution_count": 23,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "point_inference_workflow.approximator.built"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 24,