<?xml version="1.0" encoding="iso-8859-1"?>
<!-- name="generator" content="pyblosxom/1.3.2 2/13/2006" -->
<!DOCTYPE rss PUBLIC "-//Netscape Communications//DTD RSS 0.91//EN" "http://my.netscape.com/publish/formats/rss-0.91.dtd">

<rss version="0.91">
<channel>
<title>Diane's World 18 Aug 2006</title>
<link>http://woldlab.caltech.edu/~diane/index.cgi</link>
<description>I take the 5th</description>
<language>en</language>
<item>
  <title>SciPy 2006 Day 2</title>
  <link>http://woldlab.caltech.edu/~diane/index.cgi/geek/scipy2006-08-18-18-00.html</link>
  <description><![CDATA[

<div class="document">
<blockquote>
<ul>
<li><p class="first">AutoLigand AutoDock, Rodney Harris</p>
<ul class="simple">
<li>Tool for &quot;rational&quot; ligand design</li>
<li>computes affinity</li>
</ul>
</li>
<li><p class="first">GpuPy, Benjamin Eitzen, WashU</p>
<ul class="simple">
<li>strengths<ul>
<li>currently improving faster than traditional cpus</li>
</ul>
</li>
<li>weaknesses<ul>
<li>single precision</li>
<li>data must be copied to gpu</li>
</ul>
</li>
<li>programmability<ul>
<li>newer gpus can execute programs in &quot;high level&quot; languages
Cg GLSL HLSL</li>
<li>programs are called shaders</li>
</ul>
</li>
<li>how it works<ul>
<li>GPU executes program once for each pixel that is drawn to screen</li>
<li>overrrides default numpy functions, can  be transparent</li>
<li>some functions are slower on gpu (like simple add or multiply)</li>
<li>some get faster with larger N (pow, arccosh)</li>
<li>some are just faster (cos, sin)</li>
<li>the more complex you can pass over the faster, correlation</li>
<li>not hooked up yet, but have edge detection filter</li>
</ul>
</li>
<li>possible future work,<ul>
<li>autoconfig</li>
</ul>
</li>
<li>simulation of double  with singles hal.ccssd.cnrs.fr/ccsd-00021443</li>
<li>htpp://eecs.wsu.edu/~eitzenb/gpupy</li>
</ul>
</li>
<li><p class="first">Boost Graph Library, Douglas Gregor</p>
<ul>
<li><p class="first">path finding</p>
</li>
<li><p class="first">connectivity &amp; clustering</p>
</li>
<li><p class="first">network flow</p>
</li>
<li><p class="first">coloring</p>
</li>
<li><p class="first">layout &amp; visualization</p>
</li>
<li><p class="first">motivation</p>
<ul class="simple">
<li>python is easier to teach</li>
<li>dynamic features of python</li>
<li>usabiliy (e.g. c++ error messages)</li>
</ul>
</li>
<li><p class="first">Task: network tolerance</p>
<p>from boost.graph import *
g = read_graphviz(&quot;network.dot&quot;)</p>
<p>bicomponent = g.add_edge_property(name='label', type='integer)
art_points = biconnected_components(g, bicomponent)</p>
<dl class="docutils">
<dt>for v in g.vertices:</dt>
<dd><p class="first last">if v in art_points:</p>
</dd>
</dl>
</li>
<li><p class="first">graph types in python</p>
<ul class="simple">
<li>Graph</li>
<li>Digraph</li>
<li>Graph exploration interface</li>
</ul>
</li>
<li><p class="first">vertex and edge properties</p>
<ul class="simple">
<li>attach additional model domain knowledge<ul>
<li>computer net, vertex(host, ip_addr), edge(bandwidth, latency)</li>
</ul>
</li>
</ul>
</li>
<li><p class="first">Properties in the BGL</p>
<ul class="simple">
<li>property map<ul>
<li>implicity, just access property</li>
<li>explicity, allows more specifity, 2 params optional
weight = g.add_edge_property(name='weight', type='float')</li>
</ul>
</li>
<li>using property maps<ul>
<li>similar to dictionaries (weight[e], =0.75</li>
<li>pass to algorithms
dijkstra_shortest_path(g, start_vertex, weight)</li>
</ul>
</li>
<li>has visitor model to do something when you hit interesting
points in the algorithm. (e.g. debugging, new algorithms)</li>
</ul>
</li>
<li><p class="first"><a class="reference" href="http://www.generic-programming.org/~dgregor/bgl-python/">http://www.generic-programming.org/~dgregor/bgl-python/</a></p>
</li>
</ul>
</li>
<li><p class="first">summary of research issues with/Fast multipole Algorithm, Idesbald van den Bosch</p>
<ul class="simple">
<li>ship signature reduction</li>
<li>lots of details about problem and need for optimization</li>
<li>gmsh (mesh generation)</li>
</ul>
</li>
<li><p class="first">Object Oriented Finite Elements, Andrew Reid</p>
<ul class="simple">
<li>OOF2,<ul>
<li>rapid construction of meshes adapted to irregular microstructural
geometriesp</li>
</ul>
</li>
</ul>
</li>
<li><p class="first">Parallel PDE solvers in python, Bill Spotz</p>
<ul class="simple">
<li>trilinos project<ul>
<li>provide central repository for sandia's solver</li>
<li><a class="reference" href="http://softwrae.sandia.gov/trilinos">http://softwrae.sandia.gov/trilinos</a></li>
</ul>
</li>
</ul>
</li>
<li><p class="first">Python imaging tools for reconstructing magnetic resonance images, make trumpis</p>
<ul class="simple">
<li><a class="reference" href="https://cirl.berkeley.edu/view/BIC/ImagingTools">https://cirl.berkeley.edu/view/BIC/ImagingTools</a></li>
</ul>
</li>
<li><p class="first">PyRoot, Wim Lavrijsen (pronounced more like vim)</p>
<ul class="simple">
<li><a class="reference" href="http://cern.ch/wlav/pyroot/">http://cern.ch/wlav/pyroot/</a></li>
<li>ROOT root.cern.ch</li>
<li>reflection for C++ (reflex, <a class="reference" href="http://cern.ch/seal-reflex">http://cern.ch/seal-reflex</a>)</li>
<li>they combine reflection based meta-information to generate python
wrappers on the fly.</li>
<li>also can call python from C++/CINT</li>
<li>very impressive looking</li>
</ul>
</li>
<li><p class="first">QScimpl, Eric Dobbs</p>
<ul class="simple">
<li>personal, dobbse.net</li>
<li>Tech-X Corporation, txcorp (scientific physics stuff)</li>
<li>mostly decided to advertise the advantage of simplicity</li>
</ul>
</li>
<li><p class="first">Genesis SciFlo, Scientific Knowledge Creation on the Grid, Brian Wilson</p>
<ul class="simple">
<li>Problem, terabytes of data (years) distributed at several data centers</li>
<li>iEarth</li>
<li>showed an updating svg diagram showing the execution of their workflow.<ul>
<li>distributed across many machines via web services</li>
</ul>
</li>
<li>data access protocol, retrieve chunks of data over web.</li>
<li>used xpath to extract results from amazon soap results</li>
<li>software shipped<ul>
<li>lxml (XPath 1.0)</li>
<li>twisted, pyldap, openldap,</li>
<li>sleepycat dblxml has xquery and xpath</li>
<li>dojo AJAX library</li>
<li>openDAP, extract cdf, hdf file</li>
<li>globas v4 grid</li>
</ul>
</li>
<li>globus v4 is supposed to be soap webservice</li>
<li>WSRF (web services resource framework)</li>
<li>UDDI universal description discovery and integration, find service</li>
<li>grid workflow,</li>
<li>rest vs soap, each have advantages/disadvantages</li>
<li>UDDI4py from IBM, SOAPpy</li>
<li>recommends more XML microformats for &quot;mash ups&quot;</li>
<li>publish science algorithms as reusable web services</li>
<li>www.opendap.org</li>
</ul>
</li>
<li><p class="first">Python Web &amp; Grid Service Tools, Keith R Jackson</p>
<ul class="simple">
<li>Globus toolkit</li>
<li>decided to simplify system by standardized into webservices</li>
<li>helps organize coordinated usage of resources</li>
<li>his mac crashed</li>
<li>SOAPpy is officially unsported</li>
<li>ZSI is hopefully the &quot;new&quot; better way, though they're more complicated than SOAPpy</li>
<li>XML schema scary complicated</li>
<li>Mapping tool from WSDL to python (wsdl2py)</li>
<li>using m2crypto (they thought it was being supported, unlike pySSL)</li>
<li>LIGO used it to ship 50 terabytes of data through pyglobus</li>
<li>WSRF supposed to define way to hand stateful connections</li>
<li>lifecycle, things will die at some point, though can be refreshed.</li>
<li>WS-N another webservices spec</li>
<li>seems to actually work, java, C, python call all talk.</li>
<li>PyCLST</li>
<li>GridFTP, implement extra features to handle transfer over highperformance network</li>
<li>Visual programming environment, useful for viewing workflow</li>
<li>Userinterface</li>
<li>planner, condor, dagman</li>
<li>exec manager (CBEI)</li>
<li>exec context</li>
<li>physical resources</li>
<li>pyGlobus, <a class="reference" href="http://dsd.lbl.gov/gtg/projects/pyGlobus/">http://dsd.lbl.gov/gtg/projects/pyGlobus/</a></li>
<li>ZSI, <a class="reference" href="http://pywebsvcs.sourceforge.net/zsi.html">http://pywebsvcs.sourceforge.net/zsi.html</a></li>
<li>pyGridWare, <a class="reference" href="http://dsd.lbl.gov/gtg/projects/pyGridWare">http://dsd.lbl.gov/gtg/projects/pyGridWare</a></li>
<li>ViCE, <a class="reference" href="http://dsd.lbl.gov/gtg/projects/vice/">http://dsd.lbl.gov/gtg/projects/vice/</a></li>
<li>PythonCLServiceTool, <a class="reference" href="http://dsd.lbl.gov/gtg/projects/PythonCLServiceTrool">http://dsd.lbl.gov/gtg/projects/PythonCLServiceTrool</a></li>
<li>everything BSD license</li>
<li>eclipse has a nice WSDL interface generator</li>
<li>python to wsdl would require static type annotation.</li>
<li>webservices, text to do control, then binary for shipping data</li>
<li>amazon, uses a very efficient xml binary encoding</li>
</ul>
</li>
<li><p class="first">Seeing through the MIST, Tripp Lilley</p>
<ul class="simple">
<li>interesting point about maliable UIs<ul>
<li>how do you let a user, know theyre changing their world, and revert.</li>
</ul>
</li>
<li>super early project, not even screenshots</li>
</ul>
</li>
</ul>
</blockquote>
</div>

]]></description>
</item>

</channel>
</rss>
