<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> <html> <head> <link rel="stylesheet" href="style.css" type="text/css"> <meta content="text/html; charset=iso-8859-1" http-equiv="Content-Type"> <link rel="Start" href="index.html"> <link rel="previous" href="Gpr_cov_se_iso.html"> <link rel="next" href="Gpr_fitc_gp.html"> <link rel="Up" href="index.html"> <link title="Index of types" rel=Appendix href="index_types.html"> <link title="Index of exceptions" rel=Appendix href="index_exceptions.html"> <link title="Index of values" rel=Appendix href="index_values.html"> <link title="Index of modules" rel=Appendix href="index_modules.html"> <link title="Index of module types" rel=Appendix href="index_module_types.html"> <link title="Gpr" rel="Chapter" href="Gpr.html"> <link title="Gpr_interfaces" rel="Chapter" href="Gpr_interfaces.html"> <link title="Gpr_utils" rel="Chapter" href="Gpr_utils.html"> <link title="Gpr_block_diag" rel="Chapter" href="Gpr_block_diag.html"> <link title="Gpr_cov_const" rel="Chapter" href="Gpr_cov_const.html"> <link title="Gpr_cov_lin_ard" rel="Chapter" href="Gpr_cov_lin_ard.html"> <link title="Gpr_cov_lin_one" rel="Chapter" href="Gpr_cov_lin_one.html"> <link title="Gpr_cov_se_iso" rel="Chapter" href="Gpr_cov_se_iso.html"> <link title="Gpr_cov_se_fat" rel="Chapter" href="Gpr_cov_se_fat.html"> <link title="Gpr_fitc_gp" rel="Chapter" href="Gpr_fitc_gp.html"> <link title="Gpr_version" rel="Chapter" href="Gpr_version.html"><title>Gpr_cov_se_fat</title> </head> <body> <div class="navbar"><a class="pre" href="Gpr_cov_se_iso.html" title="Gpr_cov_se_iso">Previous</a> <a class="up" href="index.html" title="Index">Up</a> <a class="post" href="Gpr_fitc_gp.html" title="Gpr_fitc_gp">Next</a> </div> <h1>Module <a href="type_Gpr_cov_se_fat.html">Gpr_cov_se_fat</a></h1> <pre><span class="keyword">module</span> Gpr_cov_se_fat: <code class="code">sig</code> <a href="Gpr_cov_se_fat.html">..</a> <code class="code">end</code></pre><div class="info module top"> <h6 id="6_Featurerichfatsquaredexponentialcovariance">Feature-rich ("fat") squared exponential covariance</h6><br> </div> <hr width="100%"> <br> The covariance is defined as: <p> <code class="code">k(x, y) = sf^2 * exp(-1/2 * |(Q_i*P*(x-y))|^2)</code> <p> where <code class="code">sf^2</code> is the amplitude, <code class="code">P</code> a general <code class="code">d*D</code> dimensionality reduction matrix (<code class="code">d << D</code>), and <code class="code">Q_i</code> is a <code class="code">d*d</code> diagonal matrix containing all multiscales for inducing input number <code class="code">i</code>. <p> Note that multiscales must not get smaller than <code class="code">0.5</code> in this framework, because the overall length scale is considered to be equal to <code class="code">1</code>, which imposes this mathematical constraint for positive-definiteness. There is no need for a variable global length scale, because the dimensionality reduction matrix already generalizes this feature anyway. Hence an unconstrained multiscale parameter <code class="code">q</code> is stored as <code class="code">log(q - 0.5)</code>. <p> If <code class="code">x</code> and <code class="code">y</code> are the same inducing input, then and only then extra noise (a different noise level for each inducing input) will be added for heteroskedasticity. <p> Dimensionality reduction, heteroskedasticity, and multiscales are optional features and can be easily turned off by setting the parameters to <code class="code">None</code>.<br> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Params.html">Params</a>: <code class="code">sig</code> <a href="Gpr_cov_se_fat.Params.html">..</a> <code class="code">end</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Eval.html">Eval</a>: <code class="type">Eval</code><code class="type"> with type Kernel.params = Params.t</code><code class="type"> with type Inducing.t = mat</code><code class="type"> with type Input.t = vec</code><code class="type"> with type Inputs.t = mat</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Proj_hyper.html">Proj_hyper</a>: <code class="code">sig</code> <a href="Gpr_cov_se_fat.Proj_hyper.html">..</a> <code class="code">end</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Dim_hyper.html">Dim_hyper</a>: <code class="code">sig</code> <a href="Gpr_cov_se_fat.Dim_hyper.html">..</a> <code class="code">end</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Inducing_hyper.html">Inducing_hyper</a>: <code class="code">sig</code> <a href="Gpr_cov_se_fat.Inducing_hyper.html">..</a> <code class="code">end</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Hyper_repr.html">Hyper_repr</a>: <code class="code">sig</code> <a href="Gpr_cov_se_fat.Hyper_repr.html">..</a> <code class="code">end</code></pre> <pre><span class="keyword">module</span> <a href="Gpr_cov_se_fat.Deriv.html">Deriv</a>: <code class="type">Deriv</code><code class="type"> with module Eval = Eval</code><code class="type"> with type Hyper.t = Hyper_repr.t</code></pre></body></html>