townforge/math
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
cc
math/doc/html/math_toolkit/diff0.html
jzmaddock 72e469da0a Update version number in docs. 
[CI SKIP]
2019-10-31 17:55:35 +00:00

251 lines
24 KiB
HTML
Raw Permalink Blame History

<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=US-ASCII">
<title>Lanczos Smoothing Derivatives</title>
<link rel="stylesheet" href="../math.css" type="text/css">
<meta name="generator" content="DocBook XSL Stylesheets V1.79.1">
<link rel="home" href="../index.html" title="Math Toolkit 2.11.0">
<link rel="up" href="../quadrature.html" title="Chapter&#160;13.&#160;Quadrature and Differentiation">
<link rel="prev" href="autodiff.html" title="Automatic Differentiation">
<link rel="next" href="../inverse_complex.html" title="Chapter&#160;14.&#160;Complex Number Functions">
</head>
<body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF">
<table cellpadding="2" width="100%"><tr>
<td valign="top"><img alt="Boost C++ Libraries" width="277" height="86" src="../../../../../boost.png"></td>
<td align="center"><a href="../../../../../index.html">Home</a></td>
<td align="center"><a href="../../../../../libs/libraries.htm">Libraries</a></td>
<td align="center"><a href="http://www.boost.org/users/people.html">People</a></td>
<td align="center"><a href="http://www.boost.org/users/faq.html">FAQ</a></td>
<td align="center"><a href="../../../../../more/index.htm">More</a></td>
</tr></table>
<hr>
<div class="spirit-nav">
<a accesskey="p" href="autodiff.html"><img src="../../../../../doc/src/images/prev.png" alt="Prev"></a><a accesskey="u" href="../quadrature.html"><img src="../../../../../doc/src/images/up.png" alt="Up"></a><a accesskey="h" href="../index.html"><img src="../../../../../doc/src/images/home.png" alt="Home"></a><a accesskey="n" href="../inverse_complex.html"><img src="../../../../../doc/src/images/next.png" alt="Next"></a>
</div>
<div class="section">
<div class="titlepage"><div><div><h2 class="title" style="clear: both">
<a name="math_toolkit.diff0"></a><a class="link" href="diff0.html" title="Lanczos Smoothing Derivatives">Lanczos Smoothing Derivatives</a>
</h2></div></div></div>
<h4>
<a name="math_toolkit.diff0.h0"></a>
<span class="phrase"><a name="math_toolkit.diff0.synopsis"></a></span><a class="link" href="diff0.html#math_toolkit.diff0.synopsis">Synopsis</a>
</h4>
<pre class="programlisting"><span class="preprocessor">#include</span> <span class="special">&lt;</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">math</span><span class="special">/</span><span class="identifier">differentiation</span><span class="special">/</span><span class="identifier">lanczos_smoothing</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">&gt;</span>
<span class="keyword">namespace</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">math</span><span class="special">::</span><span class="identifier">differentiation</span> <span class="special">{</span>
<span class="keyword">template</span> <span class="special">&lt;</span><span class="keyword">class</span> <span class="identifier">Real</span><span class="special">,</span> <span class="identifier">size_t</span> <span class="identifier">order</span><span class="special">=</span><span class="number">1</span><span class="special">&gt;</span>
<span class="keyword">class</span> <span class="identifier">discrete_lanczos_derivative</span> <span class="special">{</span>
<span class="keyword">public</span><span class="special">:</span>
<span class="identifier">discrete_lanczos_derivative</span><span class="special">(</span><span class="identifier">Real</span> <span class="identifier">spacing</span><span class="special">,</span>
<span class="identifier">size_t</span> <span class="identifier">n</span> <span class="special">=</span> <span class="number">18</span><span class="special">,</span>
<span class="identifier">size_t</span> <span class="identifier">approximation_order</span> <span class="special">=</span> <span class="number">3</span><span class="special">);</span>
<span class="keyword">template</span><span class="special">&lt;</span><span class="keyword">class</span> <span class="identifier">RandomAccessContainer</span><span class="special">&gt;</span>
<span class="identifier">Real</span> <span class="keyword">operator</span><span class="special">()(</span><span class="identifier">RandomAccessContainer</span> <span class="keyword">const</span> <span class="special">&amp;</span> <span class="identifier">v</span><span class="special">,</span> <span class="identifier">size_t</span> <span class="identifier">i</span><span class="special">)</span> <span class="keyword">const</span><span class="special">;</span>
<span class="keyword">template</span><span class="special">&lt;</span><span class="keyword">class</span> <span class="identifier">RandomAccessContainer</span><span class="special">&gt;</span>
<span class="keyword">void</span> <span class="keyword">operator</span><span class="special">()(</span><span class="identifier">RandomAccessContainer</span> <span class="keyword">const</span> <span class="special">&amp;</span> <span class="identifier">v</span><span class="special">,</span> <span class="identifier">RandomAccessContainer</span> <span class="special">&amp;</span> <span class="identifier">dvdt</span><span class="special">)</span> <span class="keyword">const</span><span class="special">;</span>
<span class="keyword">template</span><span class="special">&lt;</span><span class="keyword">class</span> <span class="identifier">RandomAccessContainer</span><span class="special">&gt;</span>
<span class="identifier">RandomAccessContainer</span> <span class="keyword">operator</span><span class="special">()(</span><span class="identifier">RandomAccessContainer</span> <span class="keyword">const</span> <span class="special">&amp;</span> <span class="identifier">v</span><span class="special">)</span> <span class="keyword">const</span><span class="special">;</span>
<span class="identifier">Real</span> <span class="identifier">get_spacing</span><span class="special">()</span> <span class="keyword">const</span><span class="special">;</span>
<span class="special">};</span>
<span class="special">}</span> <span class="comment">// namespaces</span>
</pre>
<h4>
<a name="math_toolkit.diff0.h1"></a>
<span class="phrase"><a name="math_toolkit.diff0.description"></a></span><a class="link" href="diff0.html#math_toolkit.diff0.description">Description</a>
</h4>
<p>
The <code class="computeroutput"><span class="identifier">discrete_lanczos_derivative</span></code>
class calculates a finite-difference approximation to the derivative of a noisy
sequence of equally-spaced values <span class="emphasis"><em>v</em></span>. A basic usage is
</p>
<pre class="programlisting"><span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">v</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="comment">// fill v with noisy data.</span>
<span class="keyword">double</span> <span class="identifier">spacing</span> <span class="special">=</span> <span class="number">0.001</span><span class="special">;</span>
<span class="keyword">using</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">math</span><span class="special">::</span><span class="identifier">differentiation</span><span class="special">::</span><span class="identifier">discrete_lanczos_derivative</span><span class="special">;</span>
<span class="keyword">auto</span> <span class="identifier">lanczos</span> <span class="special">=</span> <span class="identifier">discrete_lanczos_derivative</span><span class="special">(</span><span class="identifier">spacing</span><span class="special">);</span>
<span class="comment">// Compute dvdt at index 30:</span>
<span class="keyword">double</span> <span class="identifier">dvdt30</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">,</span> <span class="number">30</span><span class="special">);</span>
<span class="comment">// Compute derivative of entire vector:</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">dvdt</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">);</span>
</pre>
<p>
Noise-suppressing second derivatives can also be computed. The syntax is as
follows:
</p>
<pre class="programlisting"><span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">v</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="comment">// fill v with noisy data.</span>
<span class="keyword">auto</span> <span class="identifier">lanczos</span> <span class="special">=</span> <span class="identifier">lanczos_derivative</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">,</span> <span class="number">2</span><span class="special">&gt;(</span><span class="identifier">spacing</span><span class="special">);</span>
<span class="comment">// evaluate second derivative at a point:</span>
<span class="keyword">double</span> <span class="identifier">d2vdt2</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">,</span> <span class="number">18</span><span class="special">);</span>
<span class="comment">// evaluate second derivative of entire vector:</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">d2vdt2</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">);</span>
</pre>
<p>
For memory conscious programmers, you can reuse the memory space for the derivative
as follows:
</p>
<pre class="programlisting"><span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">v</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">dvdt</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="comment">// . . . define spacing, create and instance of discrete_lanczos_derivative</span>
<span class="comment">// populate dvdt, perhaps in a loop:</span>
<span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">,</span> <span class="identifier">dvdt</span><span class="special">);</span>
</pre>
<p>
If the data has variance &#963;<sup>2</sup>, then the variance of the computed derivative
is roughly &#963;<sup>2</sup><span class="emphasis"><em>p</em></span><sup>3</sup> <span class="emphasis"><em>n</em></span><sup>-3</sup> &#916;
<span class="emphasis"><em>t</em></span><sup>-2</sup>, i.e., it increases cubically with the approximation
order <span class="emphasis"><em>p</em></span>, linearly with the data variance, and decreases
at the cube of the filter length <span class="emphasis"><em>n</em></span>. In addition, we must
not forget the discretization error which is <span class="emphasis"><em>O</em></span>(&#916;
<span class="emphasis"><em>t</em></span><sup><span class="emphasis"><em>p</em></span></sup>). You can play around with the
approximation order <span class="emphasis"><em>p</em></span> and the filter length <span class="emphasis"><em>n</em></span>:
</p>
<pre class="programlisting"><span class="identifier">size_t</span> <span class="identifier">n</span> <span class="special">=</span> <span class="number">12</span><span class="special">;</span>
<span class="identifier">size_t</span> <span class="identifier">p</span> <span class="special">=</span> <span class="number">2</span><span class="special">;</span>
<span class="keyword">auto</span> <span class="identifier">lanczos</span> <span class="special">=</span> <span class="identifier">lanczos_derivative</span><span class="special">(</span><span class="identifier">spacing</span><span class="special">,</span> <span class="identifier">n</span><span class="special">,</span> <span class="identifier">p</span><span class="special">);</span>
<span class="keyword">double</span> <span class="identifier">dvdt</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v</span><span class="special">,</span> <span class="identifier">i</span><span class="special">);</span>
</pre>
<p>
If <span class="emphasis"><em>p=2n</em></span>, then the discrete Lanczos derivative is not smoothing:
It reduces to the standard <span class="emphasis"><em>2n+1</em></span>-point finite-difference
formula. For <span class="emphasis"><em>p&gt;2n</em></span>, an assertion is hit as the filter
is undefined.
</p>
<p>
In our tests with AWGN, we have found the error decreases monotonically with
<span class="emphasis"><em>n</em></span>, as is expected from the theory discussed above. So
the choice of <span class="emphasis"><em>n</em></span> is simple: As high as possible given your
speed requirements (larger <span class="emphasis"><em>n</em></span> implies a longer filter and
hence more compute), balanced against the danger of overfitting and averaging
over non-stationarity.
</p>
<p>
The choice of approximation order <span class="emphasis"><em>p</em></span> for a given <span class="emphasis"><em>n</em></span>
is more difficult. If your signal is believed to be a polynomial, it does not
make sense to set <span class="emphasis"><em>p</em></span> to larger than the polynomial degree-
though it may be sensible to take <span class="emphasis"><em>p</em></span> less than this.
</p>
<p>
For a sinusoidal signal contaminated with AWGN, we ran a few tests showing
that for SNR = 1, p = n/8 gave the best results, for SNR = 10, p = n/7 was
the best, and for SNR = 100, p = n/6 was the most reasonable choice. For SNR
= 0.1, the method appears to be useless. The user is urged to use these results
with caution: they have no theoretical backing and are extrapolated from a
single case.
</p>
<p>
The filters are (regrettably) computed at runtime-the vast number of combinations
of approximation order and filter length makes the number of filters that must
be stored excessive for compile-time data. The constructor call computes the
filters. Since each filter has length <span class="emphasis"><em>2n+1</em></span> and there are
<span class="emphasis"><em>n</em></span> filters, whose element each consist of <span class="emphasis"><em>p</em></span>
summands, the complexity of the constructor call is O(<span class="emphasis"><em>n</em></span><sup>2</sup><span class="emphasis"><em>p</em></span>).
This is not cheap-though for most cases small <span class="emphasis"><em>p</em></span> and <span class="emphasis"><em>n</em></span>
not too large (&lt; 20) is desired. However, for concreteness, on the author's
2.7GHz Intel laptop CPU, the <span class="emphasis"><em>n=16</em></span>, <span class="emphasis"><em>p=3</em></span>
filter takes 9 microseconds to compute. This is far from negligible, and as
such the filters may be used with multiple data, and even shared between threads:
</p>
<pre class="programlisting"><span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">v1</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">v2</span><span class="special">(</span><span class="number">500</span><span class="special">);</span>
<span class="comment">// fill v1, v2 with noisy data.</span>
<span class="keyword">auto</span> <span class="identifier">lanczos</span> <span class="special">=</span> <span class="identifier">lanczos_derivative</span><span class="special">(</span><span class="identifier">spacing</span><span class="special">);</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">dv1dt</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v1</span><span class="special">);</span> <span class="comment">// threadsafe</span>
<span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special">&lt;</span><span class="keyword">double</span><span class="special">&gt;</span> <span class="identifier">dv2dt</span> <span class="special">=</span> <span class="identifier">lanczos</span><span class="special">(</span><span class="identifier">v2</span><span class="special">);</span> <span class="comment">// threadsafe</span>
<span class="comment">// need to use a different spacing?</span>
<span class="identifier">lanczos</span><span class="special">.</span><span class="identifier">reset_spacing</span><span class="special">(</span><span class="number">0.02</span><span class="special">);</span> <span class="comment">// not threadsafe</span>
</pre>
<p>
The implementation follows <a href="https://doi.org/10.1080/00207160.2012.666348" target="_top">McDevitt,
2012</a>, who vastly expanded the ideas of Lanczos to create a very general
framework for numerically differentiating noisy equispaced data.
</p>
<h4>
<a name="math_toolkit.diff0.h2"></a>
<span class="phrase"><a name="math_toolkit.diff0.example"></a></span><a class="link" href="diff0.html#math_toolkit.diff0.example">Example</a>
</h4>
<p>
We have extracted some data from the <a href="https://www.gw-openscience.org/data/" target="_top">LIGO
signal</a> and differentiated it using the (<span class="emphasis"><em>n</em></span>, <span class="emphasis"><em>p</em></span>)
= (60, 4) Lanczos smoothing derivative, as well as using the (<span class="emphasis"><em>n</em></span>,
<span class="emphasis"><em>p</em></span>) = (4, 8) (nonsmoothing) derivative.
</p>
<div class="blockquote"><blockquote class="blockquote"><p>
<span class="inlinemediaobject"><img src="../../graphs/ligo_derivative.svg" align="middle"></span>
</p></blockquote></div>
<p>
The original data is in orange, the smoothing derivative in blue, and the non-smoothing
standard finite difference formula is in gray. (Each time series has been rescaled
to fit in the same graph.) We can see that the smoothing derivative tracks
the increase and decrease in the trend well, whereas the standard finite difference
formula produces nonsense and amplifies noise.
</p>
<h4>
<a name="math_toolkit.diff0.h3"></a>
<span class="phrase"><a name="math_toolkit.diff0.caveats"></a></span><a class="link" href="diff0.html#math_toolkit.diff0.caveats">Caveats</a>
</h4>
<p>
The computation of the filters is ill-conditioned for large <span class="emphasis"><em>p</em></span>.
In order to mitigate this problem, we have computed the filters in higher precision
and cast the results to the desired type. However, this simply pushes the problem
to larger <span class="emphasis"><em>p</em></span>. In practice, this is not a problem, as large
<span class="emphasis"><em>p</em></span> corresponds to less powerful denoising, but keep it
in mind.
</p>
<p>
In addition, the <code class="computeroutput"><span class="special">-</span><span class="identifier">ffast</span><span class="special">-</span><span class="identifier">math</span></code> flag
has a very large effect on the speed of these functions. In our benchmarks,
they were 50% faster with the flag enabled, which is much larger than the usual
performance increases we see by turning on this flag. Hence, if the default
performance is not sufficient, this flag is available, though it comes with
its own problems.
</p>
<p>
This requires C++17 <code class="computeroutput"><span class="keyword">if</span> <span class="keyword">constexpr</span></code>.
</p>
<h4>
<a name="math_toolkit.diff0.h4"></a>
<span class="phrase"><a name="math_toolkit.diff0.references"></a></span><a class="link" href="diff0.html#math_toolkit.diff0.references">References</a>
</h4>
<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
<li class="listitem">
Corless, Robert M., and Nicolas Fillion. <span class="emphasis"><em>A graduate introduction
to numerical methods.</em></span> AMC 10 (2013): 12.
</li>
<li class="listitem">
Lanczos, Cornelius. <span class="emphasis"><em>Applied analysis.</em></span> Courier Corporation,
1988.
</li>
<li class="listitem">
Timothy J. McDevitt (2012): <span class="emphasis"><em>Discrete Lanczos derivatives of noisy
data</em></span>, International Journal of Computer Mathematics, 89:7, 916-931
</li>
</ul></div>
</div>
<table xmlns:rev="http://www.cs.rpi.edu/~gregod/boost/tools/doc/revision" width="100%"><tr>
<td align="left"></td>
<td align="right"><div class="copyright-footer">Copyright &#169; 2006-2019 Nikhar
Agrawal, Anton Bikineev, Paul A. Bristow, Marco Guazzone, Christopher Kormanyos,
Hubert Holin, Bruno Lalande, John Maddock, Jeremy Murphy, Matthew Pulver, Johan
R&#229;de, Gautam Sewani, Benjamin Sobotta, Nicholas Thompson, Thijs van den Berg,
Daryle Walker and Xiaogang Zhang<p>
Distributed under the Boost Software License, Version 1.0. (See accompanying
file LICENSE_1_0.txt or copy at <a href="http://www.boost.org/LICENSE_1_0.txt" target="_top">http://www.boost.org/LICENSE_1_0.txt</a>)
</p>
</div></td>
</tr></table>
<hr>
<div class="spirit-nav">
<a accesskey="p" href="autodiff.html"><img src="../../../../../doc/src/images/prev.png" alt="Prev"></a><a accesskey="u" href="../quadrature.html"><img src="../../../../../doc/src/images/up.png" alt="Up"></a><a accesskey="h" href="../index.html"><img src="../../../../../doc/src/images/home.png" alt="Home"></a><a accesskey="n" href="../inverse_complex.html"><img src="../../../../../doc/src/images/next.png" alt="Next"></a>
</div>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink