Within these groups, confirmation bias runs rampant. People surround themselves with like-minded peers, further pushing productive discourse into a neglected and forgotten corner. "You're either with us, or you're against us! And there's no in-between," we say.

I was having a typical conversation at work, discussing life after the next apocalyptic event. It occurred to me as we discussed our plans that the first few days after rule of law falls would be the scariest. People rioting through the streets, looking everything in sight, killing on a whim to get a pretzel or something; every man for himself. During this time, I plan to just hunker down with some PB&J and wait out the craziness. But after a few days, somewhat structured neighborhood alliances would form, enabling me to have a bit of trust in those around me. Instead of every man for himself, it's now us vs. them. And in that situation, I'm feeling much safer. I can cooperate with people again and defend against the neighboring block when they decide to invade again, and so on. Hooray.

Snapping out of this dark fantasy, and thinking instead about our less dire identity complexes, I came up with the following hypothesis.

Humans need safety to survive, by definition. Community groups have historically offered a bit of safety against the roving hoards. As old structures (such as neighborhoods, small towns, and churches) decrease in dominance in the modern US, there is an equal and opposite force diminishing community. Naturally, we have a subconscious urge to regain some community, so we do it the old way, but in lieu of "real" danger, we fabricate enemies through fairly trivial mechanisms (think sports) and in doing so, recreate the good old feeling of being part of "us" where they, those bastards, are part of "them."

If this is the case, then the natural antidote to cure our fabricated identity wars is to bring community back. Things like Facebook show how much we want it, but they are still pretty empty in terms of how well they satisfy our instinctual quest for safety. Everybody knows there's something ridiculous about how pigeon-holed we have made ourselves, but we're all in a kind of daze as to figuring out what to do with our ceaseless and unproductive arguing.

Others have argued similar points, but not quite for this reason. A book review I read in New Scientist of this book explained that this author claims that secular society needs religious institutions to be happy. There may be some truth to that, but trying to promote these ideas coming from an atheist group is just fanning the flames. It'd be nice to get some easy community going that is inclusive to both religious people and secular people. Maybe after cookies-and-punch hour we could all get together on Sunday afternoons and have community members give little talks about things they're good at, or a news story they've been following closely. There could be like 10 rooms, and you can go from one to the other depending on your interests. The mechanics could host a "learn to change your own tire day." If we could somehow institutionalize these kinds of gatherings, thereby giving people a little community, I think the ridiculous arguing would slow down, and we might feel better about ourselves.

Thankfully, there's a very simple solution (assuming you made my same mistake). Rather than setting the console tab to C:\python27\scripts\ipython.exe, instead set it to C:\python27\python.exe C:\python27\scripts\ipython-script.py and Ctrl+C will give a KeyboardInterrupt and leave you in ipython, just like the good old days!

• Epetra
• AztecOO
• Anasazi
• Teuchos
• epetraext
• Amesos

To do this, first run the trilinos cmake system without shared libraries. Then build. This builds things like epetra.lib. Now, go back and turn on shared libraries and build again, this time with shared libraries turned on. This builds the dlls, which I believe are required to build a python extension. If one of the ones you needs fails, just go in to Visual Studio 2008 and open up the Trilinos solution. Right click the project. Change from a static library to a dynamic librray. Change Use of MFC to "Use MFC if a Shared DLL." Change Use of ATL to "Dynamic Link to ATL." The rebuild the package. The DLL should be created.

VS 2008 options to build shared libraries

Once all the static and shared libraries are built, I didn't even mess with environmental variables (being a grad student and all) so I just copied them all into the build directory of my Python extension. On import, it worked just fine. Amazing!

I have yet to try linking to MSMPI and running large reactor problems on Microsoft HPC but that will be coming someday.

Thousands of mourners chanted “Death to Israel” and “Death to America” on Friday during the funeral of a slain nuclear expert whom Iranian officials accuse the two nations of killing in a bomb blast this week as part of a secret operation to stop Iran’s nuclear program.

- The Washington Post

Even some western experts have claimed that it must be Israel:

"If you look at the choice of target, it really could only be Israel,'' says Robert Baer, a former CIA agent in the Middle East.

- The Sydney Morning Herald

On the other hand, some suspect that certain factions of the Iranian state could have done it due to their suspicious of disloyalty and involvement with western institutions. And in that case, the following message is basically irrelevant.

Perspective: I'm a run-of-the-mill US nuclear scientist working on advanced nuclear reactors to help mankind get energy to improve quality of life.

If Israel and/or other western-like states did this, shame on them. Covertly killing scientists is nothing but counter-productive towards any imaginable goal. The enraged Iranians will likely place a more pressing national priority on developing nuclear weapons, and make it even more secret. In this case, you're not scaring people from going into the industry, but instead you're inspiring them to become national heroes. You're not causing meaningful set-backs in terms of Iranian expertise, the industrial knowledge to manufacture weapons is somewhat well-known. All you're doing is justifying their quest for nuclear weapons (who would risk losing a major city for one botched assassination attempt?) and making them more angry with the oppressively powerful west.

Countries with nuclear weapons are acting like overgrown bullies on the playground who got held back a few grades. When some less powerful country tries to pick up a stick, the bully smashes it out of his hands and says: only bullies can have sticks. The non-proliferation treaty was meant to prevent the spread of nuclear weapons AND to reduce the nuclear capabilities of the nuclear states to minimal levels. The latter has not happened (as is thoroughly detailed in this wonderful book). I fully understand and support non-proliferation, and accordingly I would like to see more reasonable attempts at stopping the spread. Until we live in a world where all nations can be comfortable in their sovereignty, the non-nuclear states will try to get nuclear weapons. The posturing we're seeing now, and this and other similar assassinations will only lead to more people trying to get nuclear weapons, in any possible aftermath. These are the most horribly destructive weapons known to man, and indirectly promoting their spread by bickering about minutia and/or invading countries for minor reasons is an extreme disservice to mankind. You're nourishing seeds that will last for generations telling people that they aren't safe until they have nuclear weapons.

Awesome bike. It even had the dinger!

Here's a really dark video that kind of shows how much fun it was

• What's a pirate's favorite programming language?
• What's a miner's favorite programming language?
• What's a hispanic's favorite programming language?
• What's a diver's favorite programming language?
• What's a mongoose's favorite programming language?
• What's a Seattleite's favorite programming language?
• What's the army's favorite scripting language?
• What's the little mermaid's favorite scripting language?
• What's Beethoven's favorite programming language?
• What's a Frenchman's favorite programming language?

Hilarious. Answers below.

• What's a pirate's favorite programming language?
  R
• What's a miner's favorite programming language?
  Ruby
• What's a hispanic's favorite programming language?
  C
• What's a diver's favorite programming language?
  Perl
• What's a mongoose's favorite programming language?
  Python
• What's a Seattleite's favorite programming language?
  Java
• What's the army's favorite scripting language?
  bash
• What's the little mermaid's favorite scripting language?
  C-shell
• What's Beethoven's favorite programming language?
  C#
• What's a Frenchman's favorite programming language?
  Pascal

I started with Linux, because I figured it would be easier. I learned a few things along the way. First, you have to have built Trilinos with shared libraries enabled. This is required for how Python handles things, I guess, and is made clear under the heading "Shared Libraries" in this file. The PyTrilinos package (which is effectively a professional set of python-trilinos extensions, btw. I needed a custom one!) needs them as well. So run your cmake-gui and check the shared libraries button. Besides the libepetra.a in the build folder, you'll now also find a libepetra.so. In Windows, you'll get >500 errors if you compile all of trilinos with shared libraries, but Epetra at least worked.

Next I needed an example Python extension file to work with. I just took my skeletal Python-C API test code and combined it with one of the simple Epetra examples from the Trilinos Wiki.

In my setup.py file for distutils, I studied the Makefile.export.Epetra to figure out which directories and libraries were needed. I then translated it over to the distutils format and got the resulting file.

from distutils.core import setup, Extension

module1 = Extension('cdifnt', sources = ['src/powerIterationExample.cpp'],
  include_dirs=['/home/user/Desktop/trilinos-10.8.3/packages/epetra/src',
                '/home/user/Desktop/trilinos-10.8.3/build',
                '/home/user/Desktop/trilinos-10.8.3/build/packages/epetra/src'],
  library_dirs=['/home/user/Desktop/trilinos-10.8.3/build/packages/epetra/src'],
                    libraries=['lapack', 'blas','epetra'])

setup (name = 'Diffusion for Neutron Transport',
        version = '1.0',
        description = 'This does diffusion and burnup.',
        ext_modules = [module1]
        )

Then just compile with the standard

python setup.py build

In the build directory (./build/lib.linux-x86_64-2.7 or something), there will now be a shared library *.so file that was built. To get it to work properly, you'll have to tell the code where the Trilinos libraries are. On Linux, this is done by setting the LD_LIBRARY_PATH environmental variable. You can do it the easy way and just copy the libepetra.so to the build directory and execute:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:.

This will add the current working directory to the path. There are better, more permanent ways of doing this.

Windows Specific:

To get it working under windows, first try building with shared libraries. In this case, you'll now find .lib and .dll files for each successful package (including Epetra) in their build/src/Resource/...etc... folders. In this case, I just changed all the paths in the setup.py file to point to the corresponding locations on the Windows machine. In this case, I used raw strings such as r'C:\Programs\trilinos\' to get it right. I have MinGW installed but I wanted it to compile with MSVC so I used the command

python setup.py build --compiler=msvc

Then, I copied the libepetra.dll file into the extension build directory and went to import it and bam! It worked. Values matched Linux values.

In [1]: import cdifnt

In [2]: cdifnt.buildMesh(1,2,3,4)
(1, 1, 1)
<nil>
Epetra in Trilinos 10.8.3

Epetra::Comm::Processor 0 of 1 total processors.
Iter = 0  Lambda = 1.73973  Residual of A*q - lambda*q = 1.32293
Iter = 100  Lambda = 3.98831  Residual of A*q - lambda*q = 0.0152304
Iter = 200  Lambda = 3.9955  Residual of A*q - lambda*q = 0.00873993
Total Mflop/s for first solve = 281.802
Increasing magnitude of first diagonal term, solving again
Iter = 0  Lambda = 1.93001  Residual of A*q - lambda*q = 1.57278
Iter = 100  Lambda = 20.0556  Residual of A*q - lambda*q = 1.69407e-21
Total Mflop/s for second solve = 324.116

Boom. Easy. Although it took like, all day.

Errors I encountered along the way:

relocation R_X86_64_32S against `vtable for Epetra_SerialComm' can
not be used when making a shared object; recompile with -fPIC
/home/user/Desktop/trilinos-10.8.3/build/packages/epetra/src/
libepetra.a: could not read symbols: Bad value

This means that you haven't compiled with shared libraries. Go back through the trilinos build process and select the option about shared libraries and rebuild.

ImportError: ./cdifnt.so: undefined symbol:
_ZNK17Epetra_SerialComm7ScanSumEPdS0_i

This means that while your module compiled fine, it did not link properly. Check your setup.py and make sure all the required libraries are available. The error you get will help figure out which one is missing. For instance, in this case, libepetra.so was not found.

Exception TypeError: 'bad argument type for built-in operation' in
<module 'threading' from 'C:\Python27\lib\threading.pyc'> ignored

This error would come after my program (not the one listed below but a more involved one) had finished executing. I narrowed it down by commenting out various parts of the code until it stopped coming. The culprit was using a PyFloat_AsDouble() call on a Python object that was actually an Int. Solution: Use PyInt_AsDouble().

Here's all the code of my mini toy python extension package, based on the Trilinos example:

#include
#include
#include
#include
#include "Python.h"
#include "Epetra_CrsMatrix.h"
#include "Epetra_MultiVector.h"
#include "Epetra_Time.h"
#include "Epetra_Vector.h"
//#include "Epetra_BlockMap.h"
#include "Epetra_Map.h"

#include "Epetra_Version.h"
#include "Epetra_SerialComm.h"

//
// Compute the eigenvalue of maximum magnitude of the given matrix A,
// using the power method.
//
// Input arguments:
//
// A: The matrix to which to apply the power method.  It's not const
//   because this method sets its flop counter.
// niters: Number of iterations of the power method.
// tolerance: Iterate until the (absolute) residual is strictly less
//   than tolerance.
// verbose: Whether or not to print status output to stdout during
//   iterations.
//
// Output argument:
//
// lambda: The eigenvalue of maximum magnitude of the matrix A.
//
// Return value: An integer error code.  Zero if no error occured,
//   else nonzero.
//
int
powerMethod (double & lambda,
             Epetra_CrsMatrix& A,
             const int niters,
             const double tolerance,
             const bool verbose);

int
runPow (void)
{
  // These "using" statements make the code a bit more concise.
  using std::cout;
  using std::endl;

  int ierr = 0, i;

  // If Trilinos was built with MPI, initialize MPI, otherwise
  // initialize the serial "communicator" that stands in for MPI.

  Epetra_SerialComm Comm;

  const int MyPID = Comm.MyPID();
  const int NumProc = Comm.NumProc();
  // We only allow (MPI) Process 0 to write to stdout.
  const bool verbose = (MyPID == 0);
  const int NumGlobalElements = 100;

  if (verbose)
    cout << Epetra_Version() << endl << endl;

  // Asking the Epetra_Comm to print itself is a good test for whether
  // you are running in an MPI environment.  However, it will print
  // something on all MPI processes, so you should remove it for a
  // large-scale parallel run.
  cout << Comm << endl;

  if (NumGlobalElements < NumProc)
    {
      if (verbose)
        cout << "numGlobalBlocks = " << NumGlobalElements
             << " cannot be < number of processors = " << NumProc << endl;
      std::exit (EXIT_FAILURE);
    }

  // Construct a Map that puts approximately the same number of rows
  // of the matrix A on each processor.
  Epetra_Map Map (NumGlobalElements, 0, Comm);

  // Get update list and number of local equations from newly created Map.
  int NumMyElements = Map.NumMyElements();

  std::vector MyGlobalElements(NumMyElements);
  Map.MyGlobalElements(&MyGlobalElements[0]);

  // NumNz[i] is the number of nonzero elements in row i of the sparse
  // matrix on this MPI process.  Epetra_CrsMatrix uses this to figure
  // out how much space to allocate.
  std::vector NumNz (NumMyElements);

  // We are building a tridiagonal matrix where each row contains the
  // nonzero elements (-1 2 -1).  Thus, we need 2 off-diagonal terms,
  // except for the first and last row of the matrix.
  for (int i = 0; i < NumMyElements; ++i)
    if (MyGlobalElements[i] == 0 || MyGlobalElements[i] == NumGlobalElements-1)
      NumNz[i] = 2; // First or last row
    else
      NumNz[i] = 3; // Not the (first or last row)

  // Create the Epetra_CrsMatrix.
  Epetra_CrsMatrix A (Copy, Map, &NumNz[0]);

  //
  // Add rows to the sparse matrix one at a time.
  //
  std::vector Values(2);
  Values[0] = -1.0; Values[1] = -1.0;
  std::vector Indices(2);
  const double two = 2.0;
  int NumEntries;

  for (int i = 0; i < NumMyElements; ++i)
    {
      if (MyGlobalElements[i] == 0)
        { // The first row of the matrix.
          Indices[0] = 1;
          NumEntries = 1;
        }
      else if (MyGlobalElements[i] == NumGlobalElements - 1)
        { // The last row of the matrix.
          Indices[0] = NumGlobalElements-2;
          NumEntries = 1;
        }
      else
        { // Any row of the matrix other than the first or last.
          Indices[0] = MyGlobalElements[i]-1;
          Indices[1] = MyGlobalElements[i]+1;
          NumEntries = 2;
        }
      ierr = A.InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
      assert (ierr==0);
      // Insert the diagonal entry.
      ierr = A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i]);
      assert(ierr==0);
    }

  // Finish up.  We can call FillComplete() with no arguments, because
  // the matrix is square.
  ierr = A.FillComplete ();
  assert (ierr==0);

  // Parameters for the power method.
  const int niters = NumGlobalElements*10;
  const double tolerance = 1.0e-2;

  //
  // Run the power method.  Keep track of the flop count and the total
  // elapsed time.
  //
  Epetra_Flops counter;
  A.SetFlopCounter(counter);
  Epetra_Time timer(Comm);
  double lambda = 0.0;
  ierr += powerMethod (lambda, A, niters, tolerance, verbose);
  double elapsedTime = timer.ElapsedTime ();
  double totalFlops =counter.Flops ();
  // Mflop/s: Million floating-point arithmetic operations per second.
  double Mflop_per_s = totalFlops / elapsedTime / 1000000.0;

  if (verbose)
    cout << endl << endl << "Total Mflop/s for first solve = "
         << Mflop_per_s << endl<< endl;

  // Increase the first (0,0) diagonal entry of the matrix.
  if (verbose)
    cout << endl << "Increasing magnitude of first diagonal term, solving again"
         << endl << endl << endl;

  if (A.MyGlobalRow (0)) {
    int numvals = A.NumGlobalEntries (0);
    std::vector Rowvals (numvals);
    std::vector Rowinds (numvals);
    A.ExtractGlobalRowCopy (0, numvals, numvals, &Rowvals[0], &Rowinds[0]); // Get A(0,0)
    for (int i = 0; i < numvals; ++i)
      if (Rowinds[i] == 0)
        Rowvals[i] *= 10.0;

    A.ReplaceGlobalValues (0, numvals, &Rowvals[0], &Rowinds[0]);
  }

  //
  // Run the power method again.  Keep track of the flop count and the
  // total elapsed time.
  //
  lambda = 0.0;
  timer.ResetStartTime();
  counter.ResetFlops();
  ierr += powerMethod (lambda, A, niters, tolerance, verbose);
  elapsedTime = timer.ElapsedTime();
  totalFlops = counter.Flops();
  Mflop_per_s = totalFlops / elapsedTime / 1000000.0;

  if (verbose)
    cout << endl << endl << "Total Mflop/s for second solve = "
         << Mflop_per_s << endl << endl;

#ifdef EPETRA_MPI
  MPI_Finalize() ;
#endif

  return ierr;
}

int
powerMethod (double & lambda,
             Epetra_CrsMatrix& A,
             const int niters,
             const double tolerance,
             const bool verbose)
{
  // In the power iteration, z = A*q.  Thus, q must be in the domain
  // of A, and z must be in the range of A.  The residual vector is of
  // course in the range of A.
  Epetra_Vector q (A.OperatorDomainMap ());
  Epetra_Vector z (A.OperatorRangeMap ());
  Epetra_Vector resid (A.OperatorRangeMap ());

  Epetra_Flops* counter = A.GetFlopCounter();
  if (counter != 0) {
    q.SetFlopCounter(A);
    z.SetFlopCounter(A);
    resid.SetFlopCounter(A);
  }

  // Initialize the starting vector z with random data.
  z.Random();

  double normz, residual;
  int ierr = 1;
  for (int iter = 0; iter < niters; ++iter)
    {
      z.Norm2 (&normz);        // normz  := ||z||_2
      q.Scale (1.0/normz, z);  // q      := z / normz
      A.Multiply(false, q, z); // z      := A * q
      q.Dot(z, &lambda);       // lambda := dot (q, z)

      // Compute the residual vector and display status output every
      // 100 iterations, or if we have reached the maximum number of
      // iterations.
      if (iter % 100 == 0 || iter + 1 == niters)
        {
          resid.Update (1.0, z, -lambda, q, 0.0); // resid := A*q - lambda*q
          resid.Norm2 (&residual);                // residual := ||resid||_2
          if (verbose)
            cout << "Iter = " << iter << "  Lambda = " << lambda
                 << "  Residual of A*q - lambda*q = " << residual << endl;
        }
      if (residual < tolerance) { // We've converged!
        ierr = 0;
        break;
      }
    }
  return ierr;
}

static PyObject * buildMesh(PyObject *difntInterface, PyObject *args)
{
	PyObject *lookupTable;
	int *iMax;
	int *jMax;
	int *kMax;
	PyObject *key;
	PyObject *block = Py_None;

	if (!PyArg_ParseTuple(args, "Oiii", &lookupTable,&iMax,&jMax,&kMax))
	        return NULL;

	// Demonstrate that the lookupTable is accessible.
	key = Py_BuildValue("iii", 1, 1, 1);
	//Py_INCREF(key);
	PyObject_Print(key,stdout,Py_PRINT_RAW);
	printf("\n");
	block = PyDict_GetItem(lookupTable,key);
	//Py_INCREF(block);
	PyObject_Print(block,stdout,Py_PRINT_RAW);
	printf("\n");

	runPow();

	Py_RETURN_NONE;
}

static PyMethodDef DifntSolverMethods[] =
{
     {"buildMesh", buildMesh, METH_VARARGS, "Builds a mesh."},
     {NULL, NULL, 0, NULL}
};

PyMODINIT_FUNC

initcdifnt(void)
{
     (void) Py_InitModule("cdifnt", DifntSolverMethods);
}

Hope this helps someone!

Trilinos 10.4 is available in the repositories for Ubuntu (11.10 confirmed). But Trilinos 10.8 is out now, so I had to build my own. Their instructions with CMAKE were quite good and the build process didn't have too many hiccups. I found that the cmake gui is the way to go. Make sure you enable Didasko (the tutorial) in the cmake gui. When you do this, a few more Didasko options show up so you have to go enable those too.

What, in retrospect, should have been obvious from the documentation, the way to build your own programs against Trilinos is to use the Makefile.export files. Once built, there will be a Makefile.export.[PackageName] for each packages in the /build/packages directories. That file has all the variables needed to tell your custom Makefile how to include Trilinos. So, you take your makefile and just say "include Makefile.export.Epetra" for example. I used the makefile at the bottom of the previous link as a skeleton for my first sample program. One problem I ran into with the skeleton is that it includes the Makefile.export.Trilinos and my build of Trilinos was in a long-named folder and has all the packages enabled, so I kept getting

c++: error trying to exec '/usr/lib/gcc/x86_64-linux-gnu/4.6.1/collect2': execv: Argument list too long

This is because the argument list generated in my Trilinos makefile was literally 133469 characters long, and according to

xargs --show-limits

my limit is 131072. Too bad. I haven't found out how to increase this limit other than recompiling the kernel, which I'm not about to do. Solution? Forget about the full Makefile and just add the packages I'm using. In this sample case, I'm compiling the first example from Didasko, ex1.cpp.

Another error I had was

ex1.cpp:35:32: fatal error: Didasko_ConfigDefs.h: No such file or directory

To get around that, I edited ex1.cpp and removed the outer #ifdef loop that provided the error message. This included removing calls to Didasko_ConfigDefs.h.

Final trick: In the skeletal Makefile, all the variables are defined as Trilinos_CXX, etc. When you include a specific package's makefile, you have to change them to Epetra_CXX, etc. So I did a find/replace Trilinos->Epetra and hit make. Check it out:

user@computer:~/Desktop/trilinos-10.8.3/build/Testing/Temporary$ make
Found Epetra!  Here are the details:
 Epetra_VERSION =
 Epetra_PACKAGE_LIST = Epetra
 Epetra_LIBRARIES = -lepetra
 Epetra_INCLUDE_DIRS = -I~/Desktop/trilinos-10.8.3/build -I~/Desktop/trilinos-10.8.3/build/packages/epetra/src -I~/Desktop/trilinos-10.8.3/packages/epetra/src
 Epetra_LIBRARY_DIRS = -L~/Desktop/trilinos-10.8.3/build/packages/epetra/src
 Epetra_TPL_LIST = LAPACK BLAS
 Epetra_TPL_INCLUDE_DIRS =
 Epetra_TPL_LIBRARIES = /usr/lib/liblapack.so /usr/lib/libblas.so
 Epetra_TPL_LIBRARY_DIRS =
 Epetra_BUILD_SHARED_LIBS = OFF
 End of Epetra details
/usr/bin/c++   ex1.o  -o ex1  -I~/Desktop/trilinos-10.8.3/build -I~/Desktop/trilinos-10.8.3/build/packages/epetra/src -I~/Desktop/trilinos-10.8.3/packages/epetra/src   -DMYAPP_EPETRA -L~/Desktop/trilinos-10.8.3/build/packages/epetra/src  -lepetra /usr/lib/liblapack.so /usr/lib/libblas.so

Compiled! Sweet. More on this as it progresses.

I've been interested in accessing very large dictionaries very quickly for a nuclear reactor simulation code I'm writing for my thesis. For a large core (3000+ regions, 6 mesh points/region, 33 groups), the scatter matrices easily make dictionaries with 16 million entries or so. Since scatter matrices are fairly sparse, I've saved memory by only entering non-zero values. Then, looping back through, I just used get. It was slow though. I wondered whether saving the non-zero keys and accessing the dictionary directly would be much faster. So I ran a test with various amounts of data density. Low density means there are more zero entries.

Dictionary access timing results for various amounts of data sparsity

In conclusion, directly accessing remembered keys converges to not quite twice as fast for full dictionaries. Good to know.

More info on speed in Python here.

the code:

import random
import time
import pylab

build=[]
get=[]
direct=[]

# build a big dict
densities=[1,2,10,20,30,40]
for density in densities:
    data={}
    keys=[]
    t0=time.time()
    for block in xrange(3000):
        for i in xrange(33):
            for j in xrange(33):
                key = (block,i,j)
                if not (block+j+i)%density:
                    # sparse data.
                    pass
                else:
                    data[key] = block+i+j
                    keys.append(key)

    # try it out with gets
    t1=time.time()
    for block in range(3000):
        for i in xrange(33):
            for j in xrange(33):
                val = data.get((block,i,j),0.0)
    t2=time.time()

    # try it with keys
    for key in keys:
        val = data[key]

    t3=time.time()

    print '{0:17s} {1:10.3f}'.format('Building the data',t1-t0)
    print '{0:17s} {1:10.3f}'.format('Gets',t2-t1)
    print '{0:17s} {1:10.3f}'.format('Directs',t3-t2)

    build.append(t1-t0)
    get.append(t2-t1)
    direct.append(t3-t2)

pylab.figure()
pylab.plot(densities,get,label='dict.get(key,0.0)')
pylab.plot(densities,direct,label='dict[key]')
pylab.grid()
pylab.legend(loc='lower right')
pylab.title('Speed of dictionary methods')
pylab.xlabel('Data density')
pylab.ylabel('Time (s)')
pylab.savefig('dictTimingResults.png')

Hi! I'm Chester. How about a nice game of chess?

Sure you're playing better but your taking more time than me.

Now I've got you!

How about a draw?

Many good times hanging out with old Chester. They did a great job putting in tonality into his commentary. It's really like playing with another person. I only wish the random number generator had him do other openings more often. I've never really used the tournament openings mode, but I'm kind of curious. I wish I still had the manual. Anyway, the best part is that I know that my 4th-grade self is better at chess than my current self because I used to be able to easily beat Chester on a higher level than I can now.