hierarchical forecasting

categories/bayesian/index.html

@@ -84,6 +84,13 @@
 
     <h2 class="archive-title">2024</h2>
 
+    <article class="archive-item">
+      <a href="../../numpyro_hierarchical_forecasting_1/" class="archive-item-link">From Pyro to NumPyro: Forecasting Hierarchical Models - Part I</a>
+      <span class="archive-item-date">
+        2024-10-03
+      </span>
+    </article>
+
     <article class="archive-item">
       <a href="../../numpyro_forecasting-univariate/" class="archive-item-link">From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series</a>
       <span class="archive-item-date">

categories/bayesian/index.xml

@@ -6,7 +6,17 @@
     <description>Recent content in Bayesian on Dr. Juan Camilo Orduz</description>
     <generator>Hugo -- gohugo.io</generator>
     <language>en-us</language>
-    <lastBuildDate>Tue, 01 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/categories/bayesian/index.xml" rel="self" type="application/rss+xml" />
+    <lastBuildDate>Thu, 03 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/categories/bayesian/index.xml" rel="self" type="application/rss+xml" />
+    <item>
+      <title>From Pyro to NumPyro: Forecasting Hierarchical Models - Part I</title>
+      <link>/numpyro_hierarchical_forecasting_1/</link>
+      <pubDate>Thu, 03 Oct 2024 00:00:00 +0000</pubDate>
+
+      <guid>/numpyro_hierarchical_forecasting_1/</guid>
+      <description>In this notebook we provide a NumPyro implementation of the first model presented in the Pyro forecasting documentation: Forecasting III: hierarchical models. This model generalizes the local level model with seasonality presented in the univariate example Forecasting I: univariate, heavy tailed (see From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series for the corresponding NumPyro implementation).
+In this example, we continue working with the BART train ridership dataset.</description>
+    </item>
+
     <item>
       <title>From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series</title>
       <link>/numpyro_forecasting-univariate/</link>

categories/index.html

@@ -87,7 +87,7 @@ <h2 class="archive-title">2024</h2>
     <article class="archive-item">
       <a href="../categories/bayesian/" class="archive-item-link">Bayesian</a>
       <span class="archive-item-date">
-        2024-10-01
+        2024-10-03
       </span>
     </article>
 

categories/index.xml

@@ -6,11 +6,11 @@
     <description>Recent content in Categories on Dr. Juan Camilo Orduz</description>
     <generator>Hugo -- gohugo.io</generator>
     <language>en-us</language>
-    <lastBuildDate>Tue, 01 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/categories/index.xml" rel="self" type="application/rss+xml" />
+    <lastBuildDate>Thu, 03 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/categories/index.xml" rel="self" type="application/rss+xml" />
     <item>
       <title>Bayesian</title>
       <link>/categories/bayesian/</link>
-      <pubDate>Tue, 01 Oct 2024 00:00:00 +0000</pubDate>
+      <pubDate>Thu, 03 Oct 2024 00:00:00 +0000</pubDate>
 
       <guid>/categories/bayesian/</guid>
       <description></description>

index.html

@@ -87,6 +87,13 @@
 
     <h2 class="archive-title">2024</h2>
 
+    <article class="archive-item">
+      <a href="./numpyro_hierarchical_forecasting_1/" class="archive-item-link">From Pyro to NumPyro: Forecasting Hierarchical Models - Part I</a>
+      <span class="archive-item-date">
+        2024-10-03
+      </span>
+    </article>
+
     <article class="archive-item">
       <a href="./numpyro_forecasting-univariate/" class="archive-item-link">From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series</a>
       <span class="archive-item-date">

index.xml

@@ -6,7 +6,17 @@
     <description>Recent content on Dr. Juan Camilo Orduz</description>
     <generator>Hugo -- gohugo.io</generator>
     <language>en-us</language>
-    <lastBuildDate>Tue, 01 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/index.xml" rel="self" type="application/rss+xml" />
+    <lastBuildDate>Thu, 03 Oct 2024 00:00:00 +0000</lastBuildDate><atom:link href="/index.xml" rel="self" type="application/rss+xml" />
+    <item>
+      <title>From Pyro to NumPyro: Forecasting Hierarchical Models - Part I</title>
+      <link>/numpyro_hierarchical_forecasting_1/</link>
+      <pubDate>Thu, 03 Oct 2024 00:00:00 +0000</pubDate>
+
+      <guid>/numpyro_hierarchical_forecasting_1/</guid>
+      <description>In this notebook we provide a NumPyro implementation of the first model presented in the Pyro forecasting documentation: Forecasting III: hierarchical models. This model generalizes the local level model with seasonality presented in the univariate example Forecasting I: univariate, heavy tailed (see From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series for the corresponding NumPyro implementation).
+In this example, we continue working with the BART train ridership dataset.</description>
+    </item>
+
     <item>
       <title>From Pyro to NumPyro: Forecasting a univariate, heavy tailed time series</title>
       <link>/numpyro_forecasting-univariate/</link>

numpyro_forecasting-univariate/index.html

@@ -252,6 +252,9 @@ <h2>Seasonal Features</h2>
 fourier_modes_jax = periodic_features_jax(T2, 365.25 / 7)
 # Verify that the Fourier features are the same as the Pyro code
 assert jnp.allclose(jnp.array(fourier_modes_torch), fourier_modes_jax)</code></pre>
+</div>
+<div id="train---test-split" class="section level2">
+<h2>Train - Test Split</h2>
 <p>Next, we split these features into train and test.</p>
 <pre class="python"><code>time = jnp.arange(T2)
 time_train = time[:T1]
@@ -435,7 +438,7 @@ <h1>Inference with SVI</h1>
 <center>
 <img src="../images/numpyro_forecasting_univariate_files/numpyro_forecasting_univariate_33_2.png" style="width: 900px;"/>
 </center>
-<p>The elbo loss is decreasing as expected.</p>
+<p>The ELBO loss is decreasing as expected.</p>
 <div id="posterior-predictive-check" class="section level2">
 <h2>Posterior Predictive Check</h2>
 <p>We can use the fitted model to sample from the posterior predictive distribution and generate posterior samples for the train and test data.</p>