Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion preliminaries/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -1280,7 +1280,7 @@ slides: true
<img src="00.Preliminaries.key-stage-0094anim0.png" data-images="00.Preliminaries.key-stage-0094anim0.png,00.Preliminaries.key-stage-0094anim1.png,00.Preliminaries.key-stage-0094anim2.png,00.Preliminaries.key-stage-0094anim3.png,00.Preliminaries.key-stage-0094anim4.png,00.Preliminaries.key-stage-0094anim5.png" class="slide-image" />

<figcaption>
<p >Here's a visualization of what it means to take the partial derivative as a point <span class="red">x</span>=0, <span class="blue">y</span>=5,<span class="red"> </span><em class="red">with respect to x</em>.<br></p><p >We let the function vary with <span class="red">x</span>, keeping <span class="blue">y</span> fixed. This gives us a kind of "slice" through the surface f. The result is that we have a one-dimensional function again, for which we can take the derivative. We do this by only caring about the way f changes if we change <span class="red">x</span>. In other words, by treating <span class="blue">y</span> as a constant.<br></p><p >Then we do the same for <span class="blue">y</span>, treating <span class="red">x</span> as a constant.<br></p><p >For both curves, the derivatives give us the slope of the tangent line. We've shown these as dotted lines here.<br></p><p >These lines cross the same point, so together, they lie in a shared <em>plane</em>. In higeher dimensions, the tangent lines of all partial derivatives lie in a shared <em>hyperplane</em>. This is the <em>tangent hyperplane</em>. The hyperplane that <em>just</em> touches the surface of f. <br></p><p >How do we describe the tangent hyperplane? Note what the tangent slopes indicate: how much the plane moves up if we take a step of 1 along the <span class="red">x</span> axis and how much the plane moves up if we take a step along the <span class="red">y</span> axis. These are exactly the roles of the constants in the function <span class="red">x</span>w<sub>1</sub> + <span class="blue">y</span>w<sub>2</sub> + b. Or, in vector notation the elements of the vector w in the function <strong>x</strong><sup>T</sup><strong>w</strong> +<span class="blue"> b</span>.<br></p><p >This tells us that if we take all of our partial derivatives and stick them in a vector <strong>w</strong>, the function that will describe our tangent hyperplane is <strong>x</strong><sup>T</sup><strong>w</strong> + <span class="blue">b</span> (for some value of <span class="blue">b</span>, which we don't usually care about).<br></p><p ></p>
<p >Here's a visualization of what it means to take the partial derivative as a point <span class="red">x</span>=0, <span class="blue">y</span>=5,<span class="red"> </span><em class="red">with respect to x</em>.<br></p><p >We let the function vary with <span class="red">x</span>, keeping <span class="blue">y</span> fixed. This gives us a kind of "slice" through the surface f. The result is that we have a one-dimensional function again, for which we can take the derivative. We do this by only caring about the way f changes if we change <span class="red">x</span>. In other words, by treating <span class="blue">y</span> as a constant.<br></p><p >Then we do the same for <span class="blue">y</span>, treating <span class="red">x</span> as a constant.<br></p><p >For both curves, the derivatives give us the slope of the tangent line. We've shown these as dotted lines here.<br></p><p >These lines cross the same point, so together, they lie in a shared <em>plane</em>. In higher dimensions, the tangent lines of all partial derivatives lie in a shared <em>hyperplane</em>. This is the <em>tangent hyperplane</em>. The hyperplane that <em>just</em> touches the surface of f. <br></p><p >How do we describe the tangent hyperplane? Note what the tangent slopes indicate: how much the plane moves up if we take a step of 1 along the <span class="red">x</span> axis and how much the plane moves up if we take a step along the <span class="red">y</span> axis. These are exactly the roles of the constants in the function <span class="red">x</span>w<sub>1</sub> + <span class="blue">y</span>w<sub>2</sub> + b. Or, in vector notation the elements of the vector w in the function <strong>x</strong><sup>T</sup><strong>w</strong> +<span class="blue"> b</span>.<br></p><p >This tells us that if we take all of our partial derivatives and stick them in a vector <strong>w</strong>, the function that will describe our tangent hyperplane is <strong>x</strong><sup>T</sup><strong>w</strong> + <span class="blue">b</span> (for some value of <span class="blue">b</span>, which we don't usually care about).<br></p><p ></p>
</figcaption>
<span class="hint">click image for animation</span>
</section>
Expand Down