rebuilding site Fri Apr 21 22:06:37 EDT 2017

dwhitena · dwhitena · commit a8480d08cf8a · 2017-04-21T22:06:37.000-04:00
diff --git a/index.xml b/index.xml
@@ -364,7 +364,7 @@
 
 &lt;h3 id=&#34;numerical-stability&#34;&gt;Numerical Stability&lt;/h3&gt;
 
-&lt;p&gt;Consider the equation $y = log(1 + x)$. This equation is not &lt;a href=&#34;https://en.wikipedia.org/wiki/Numerical_stability&#34; target=&#34;_blank&#34;&gt;numerically stable&lt;/a&gt; - for very small values of &lt;code&gt;x&lt;/code&gt;, the answer will most likely be wrong. This is because of the way &lt;code&gt;float64&lt;/code&gt; is designed - a &lt;code&gt;float64&lt;/code&gt; does not have enough bits to be able to tell apart &lt;code&gt;1&lt;/code&gt; and &lt;code&gt;1 + 10e-16&lt;/code&gt;. In fact, the correct way to do $latex y = log(1 + x)$ is to use the built in library function &lt;code&gt;math.Log1p&lt;/code&gt;. It can be shown in this simple program:&lt;/p&gt;
+&lt;p&gt;Consider the equation $y = log(1 + x)$. This equation is not &lt;a href=&#34;https://en.wikipedia.org/wiki/Numerical_stability&#34; target=&#34;_blank&#34;&gt;numerically stable&lt;/a&gt; - for very small values of &lt;code&gt;x&lt;/code&gt;, the answer will most likely be wrong. This is because of the way &lt;code&gt;float64&lt;/code&gt; is designed - a &lt;code&gt;float64&lt;/code&gt; does not have enough bits to be able to tell apart &lt;code&gt;1&lt;/code&gt; and &lt;code&gt;1 + 10e-16&lt;/code&gt;. In fact, the correct way to do $ y = log(1 + x)$ is to use the built in library function &lt;code&gt;math.Log1p&lt;/code&gt;. It can be shown in this simple program:&lt;/p&gt;
 
 &lt;pre&gt;&lt;code class=&#34;language-go&#34;&gt;func main() {
 	fmt.Printf(&amp;quot;%v\n&amp;quot;, math.Log(1.0+10e-16))
diff --git a/post/deeplearning_in_go_part_1/index.html b/post/deeplearning_in_go_part_1/index.html
@@ -524,7 +524,7 @@ <h2 id="why-graph-objects">Why Graph Objects?</h2>
 
 <h3 id="numerical-stability">Numerical Stability</h3>
 
-<p>Consider the equation $y = log(1 + x)$. This equation is not <a href="https://en.wikipedia.org/wiki/Numerical_stability" target="_blank">numerically stable</a> - for very small values of <code>x</code>, the answer will most likely be wrong. This is because of the way <code>float64</code> is designed - a <code>float64</code> does not have enough bits to be able to tell apart <code>1</code> and <code>1 + 10e-16</code>. In fact, the correct way to do $latex y = log(1 + x)$ is to use the built in library function <code>math.Log1p</code>. It can be shown in this simple program:</p>
+<p>Consider the equation $y = log(1 + x)$. This equation is not <a href="https://en.wikipedia.org/wiki/Numerical_stability" target="_blank">numerically stable</a> - for very small values of <code>x</code>, the answer will most likely be wrong. This is because of the way <code>float64</code> is designed - a <code>float64</code> does not have enough bits to be able to tell apart <code>1</code> and <code>1 + 10e-16</code>. In fact, the correct way to do $ y = log(1 + x)$ is to use the built in library function <code>math.Log1p</code>. It can be shown in this simple program:</p>
 
 <pre><code class="language-go">func main() {
 	fmt.Printf(&quot;%v\n&quot;, math.Log(1.0+10e-16))
diff --git a/post/index.xml b/post/index.xml
@@ -354,7 +354,7 @@
 
 &lt;h3 id=&#34;numerical-stability&#34;&gt;Numerical Stability&lt;/h3&gt;
 
-&lt;p&gt;Consider the equation $y = log(1 + x)$. This equation is not &lt;a href=&#34;https://en.wikipedia.org/wiki/Numerical_stability&#34; target=&#34;_blank&#34;&gt;numerically stable&lt;/a&gt; - for very small values of &lt;code&gt;x&lt;/code&gt;, the answer will most likely be wrong. This is because of the way &lt;code&gt;float64&lt;/code&gt; is designed - a &lt;code&gt;float64&lt;/code&gt; does not have enough bits to be able to tell apart &lt;code&gt;1&lt;/code&gt; and &lt;code&gt;1 + 10e-16&lt;/code&gt;. In fact, the correct way to do $latex y = log(1 + x)$ is to use the built in library function &lt;code&gt;math.Log1p&lt;/code&gt;. It can be shown in this simple program:&lt;/p&gt;
+&lt;p&gt;Consider the equation $y = log(1 + x)$. This equation is not &lt;a href=&#34;https://en.wikipedia.org/wiki/Numerical_stability&#34; target=&#34;_blank&#34;&gt;numerically stable&lt;/a&gt; - for very small values of &lt;code&gt;x&lt;/code&gt;, the answer will most likely be wrong. This is because of the way &lt;code&gt;float64&lt;/code&gt; is designed - a &lt;code&gt;float64&lt;/code&gt; does not have enough bits to be able to tell apart &lt;code&gt;1&lt;/code&gt; and &lt;code&gt;1 + 10e-16&lt;/code&gt;. In fact, the correct way to do $ y = log(1 + x)$ is to use the built in library function &lt;code&gt;math.Log1p&lt;/code&gt;. It can be shown in this simple program:&lt;/p&gt;
 
 &lt;pre&gt;&lt;code class=&#34;language-go&#34;&gt;func main() {
 	fmt.Printf(&amp;quot;%v\n&amp;quot;, math.Log(1.0+10e-16))
