Old Change Logs

Renamed the two parameters in the probabilistic_decision_function from a and b
to alpha and beta.  This is to avoid any possibility of a probabilistic_decision_function
being accidentally passed into templated code that is looking for the bias (the b element)
term of a decision_function.

Added a spinner to part of a test.

Removed two try/catch blocks from the threading code.  One was replaced with
the RAII equivalent and the other was the final catch block to log uncaught exceptions
from the user and then abort the program.  The code now lets the exception escape
to the operating system so that it is easier to debug uncaught exceptions.  

Added the gradient_norm_stop_strategy

Fixed compiler warning from gcc

Added a missing #include.

Changed the code a little in an attempt to make it more robust against Windows header
files that try and #define the words "min" and "max"

Fixed typo in spec.

Added comments regarding the paired calling order of functions and their
derivatives.

Fixed spelling and grammar.

Refined the spec for some functions and also added some missing asserts.

Split the code up into multiple files and setup the abstracts for each of them.

Removed the old find_min* functions, updated the regression tests, and
modified any code that used the old functions.

Added negate_function and find_max versions of the find_min functions.

Added the objective_delta_stop_strategy and generally cleaned up the code more.
Also optimized the derivative() function a bit.

Added an L-BFGS search strategy.

Changed code slightly to avoid a warning from gcc.

This is the first checkin of a bunch of code refactoring I have been doing to this component.
Most of these changes are just rearrangements of the old code.  However, I also changed
the starting value of a line search and also removed some unneeded calls to the objective
function and its derivative in the course of doing the refactoring.  This has all resulted
in a significant reduction in calls to the objective function.

Changed the image_window so that it doesn't resize itself if you keep
giving it images that are the same size.  This way if you are sending
video data to it will allow the user to change the size of the window
without having it snap right back.

Fixed grammar

Fixed a bug in the get_filesystem_roots() roots function that
prevented it from compiling.

Changed code to avoid compiler warning.

Minor changes to avoid compiler warnings in mingw.

In addition to minor code cleanup I also changed the following:
   - The zoomable_region widget now uses exponential rather than linear 
     zoom scaling since this is much more pleasing to use.  There is now 
     a new requirement on the zoom increment that it must be between 0 
     and 1.
   - I added a few missing requires clauses.  
   - I changed the set_zoom_scale() function so that it actually changes
     the zoom scale even if the user requests a zoom value outside the 
     allowed range.  It just uses the max or min allowed value rather than 
     doing nothing.

Added a check to the deserialization code for vector_normalizer objects that
will detect if the user is trying to deserialize a vector_normalizer_pca object
into a vector_normalizer.  If so it throws serialization_error.

Added a get_rect() for array2d objects.

Added the bgr_pixel and cv_image objects.

Added some typedefs for signed fixed width integers.

Switched the batch mode kernel cache to use whatever floating point
type the rest of the code is using because otherwise it can negatively
impact the results of training.

Added a version of clamp that can take runtime determined values.

Fixed a minor bug in the zoomable_region::set_min_zoom_scale() function.
It didn't always end up zooming in a smooth sensible manner after this
function was called.

Added a dot() function for matrix objects.

Added the reciprocal_max() function.

Split the vector_normalizer into a normal normalizer and a pca normalizer
version.  I did this mostly so that the normal version can return
matrix_type objects and thus avoid any reallocation of matrix objects
in client code.

Fixed incorrect comments.

Fixed the serialization functions for the normalized_function object.  They will
now work with custom normalizer function objects.

Fixed an inclusion order bug between the shared_ptr_thread_safe and thread_pool files.

Removed cruft from spec.

Modified the svm_pegasos class so that the user can set independent lambda
parameters for each class.  This also breaks backwards compatibility with 
the previous interface slightly and changes the serialization format
of this class.

Fixed some compiler warnings and also a bug in the cached version of the
batch_trainer.

Added the ability to use a kernel cache to the batch_trainer object.  I also
changed it so that it always calls clear() on the trainer it uses before it
begins training.  This way it always forgets the results of previous trainings.

Fixed incorrect ensures clause in the pegasos clear() function.

Changed the thread_pool and future classes so that any mixture of
destruction orders between the two is legal.

Fixed a bug in the cross_validate_trainer_threaded() function.  It could deadlock if
more than about 10 folds were requested.

Fixed grammar

Upgraded the thread_function object so that it is more general.  It
can now call arbitrary function objects with up to 4 arguments.

Added some comments to clarify why I put a COMPILE_TIME_ASSERT
on function references.

Changed code to compile in visual studio 2003

Minor change to avoid compiler warnings.

I just fixed some warnings and compiler errors you get if you try to compile dlib 
with the newest version of mingw.

Changed code to avoid compiler warning.

Fixed a warning you get in mingw.

Upgraded the smart_ptr so that it can handle array pointers as well
as customer deleter functions.

Just retabbed this file.

Changed the names of the variables in the TIME_THIS macro to avoid
any likely name collisions with user's code.

This commit adds the requirement that matrices being multiplied together
aren't empty.  Previously this was technically allowed and worked but was
weird.  The optimization I checked a few hours ago also breaks when empty
matrices are multiplied together so I'm just adding this new requirement.

I also had to fix a part of the LU decomposition because it was doing an
empty matrix multiplication at one point.

Minor changes to test case.

Made the normalized_function object fully generic by making the
normalizer a template argument (with its default set to the vector_normalizer
so this change doesn't break backwards compatibility).

Changed the default matrix multiplication expression object slightly
to optimize it a little more.

Changed a few things to avoid bugs in visual studio 7.1

Changed the matrix_assign() functions a bit to force the compiler to unroll the 
assignment loop if it is small and known at compile time.

Refactored the + and - operators for the dlib::vector object so that they always compile
when used with mixed matrix and vector objects.

Changed code slightly to avoid a warning in gcc when -W -Wall is given.

Fixed typo in spec

Added a missing rethrow statement.  The xml parser would
eat exceptions thrown by event handlers rather than letting
them propagate out as documented in the specification.

Added a hook to the logger object that lets you set a different kind of
output logging destination (in addition to the std::ostream supported already).

Renamed unroll_to_column_vector() to reshape_to_column_vector()

Added the unroll_to_column_vector() function.

Made the spec more clear.

Fixed grammar

Reworked the config_reader interface a little to make it easier to use.  In 
particular, I removed the enumerator over blocks in favor of a simple 
get_blocks() function that just returns a std::vector of all the blocks.  
I also removed the requires clauses on the block and key accessor functions 
and instead made a request for a non-existent key/block result in a non-fatal 
exception.  This way users can let the config reader perform a more natural 
role in config file validation (by catching this exception and acting accordingly).

Updated the equal() function so that it can compare complex matrices.
I also changed a matrix test case to be more robust.

Made the clp checker spec more clear.

Simplified the matrix code slightly.   This also fixed a strange compile time bug
that you can get from gcc in certain cases.

Fixed the gaussian() function used by the SURF code.  It wasn't 
computing a properly weighted Gaussian function.

Changed the reduced_decision_function_trainer2 object so that it performs a final 
optimization of the weights before it finishes.  This ensures that we always
end up with the best weights for the support vectors we obtain after the
conjugate gradient optimization step.

Cleanup test code and fixed compiler warning.

Added overloads of the kernel_derivative object for all the kernels in dlib.

Slightly optimized how overlays are added to the image_display widget.

Fixed compiler error I just introduced a moment ago.

Merged in Miguel Grinberg's patch to add a non-blocking read() function to the
connection object.

Added some stuff to test the integral_image class.

Added more tests

Removed clutter from spec file and also fixed a bug in the assign_border_pixels() and
zero_border_pixels() functions.  Their contracts said there was no upper limit on the
size of the border that could be assigned/zeroed but the implementations failed to 
handle the case where the border was bigger than the image.

Changed code to avoid compiler warnings.

Made the spec more clear

Fixed some typos

Added missing comments and asserts.

Changed code to avoid false test failures in some compilers.

Made this code slightly more robust.

Made the spec more clear

Added the roc_trainer_type object.

Changed the definition of the is_binary_classification_problem() function
so that it is a little more reasonable and also easier to understand.

General code cleanup.  Most of the changes are due to me simplifying
the way a bunch of requirements were written.  I also added missing
assert statements to the randomize_samples() functions.

Made the specs more clear

Added the null_trainer_type object.

I just did some code cleanup.  In particular, I added a typedef to the decision function
objects to allow you to get the kernel type out of them.  I also removed some extraneous
get_kernel() functions from some of the trainer adapter classes since they really aren't
needed.

Made the pipe spec more clear.

Fixed spelling error in spec.

Added the get_files_in_directory_tree() function.

Changed code slightly to avoid warnings from gcc when -W -Wall is given.

Changed the pinv() function so that it doesn't get slow when operating on 
matrices with many more columns than rows.

Added functions to grow and shrink the border of rectangle objects.

Fixed the load_bmp function with respect to another weird BMP format I found.

Added missing & symbols to the return types of the matrix cast operators. 
Now vector_to_matrix() and array_to_matrix() produce matrix expressions
that return their elements by reference rather than by value.

Added a missing type cast to the reciprocal() function to fix a compile
time error you get when you use it with complex<float> type matrices.

Added serialization support to the type_safe_union object.

Added the zeros_matrix(), ones_matrix() functions.  I also changed the pointwise_multiply()
and complex_matrix() functions so that they are a little more convenient when dealing
with complex types.

Updated the load_bmp() function so that it is capable of reading BMP
files that have been compressed with the RLE compression sometimes
used for 8bit BMP files.

Changed the code so that gcc doesn't give warnings when supplied with the "-Wall -W"
combined warning option.

Changed the typedefs for the xml_parser object so that the checked version
results in a checked attribute list being passed back to the user when
the parser is used.  

Added a default constructor to the kcentroid object since it can be irritating in 
some cases without one.

Fixed typo in spec.

Fixed typo in spec file.

Changed the socket read/write code so that it can handle a large number ( > 2 billion)
of bytes in I/O calls.

Changed the code a little bit so that it should be more portable and
robust to variations in how std::streamsize is defined.

Fixed spelling error

Applied Steven Van Ingelgem's http server patch to trim inappropriate whitespace
from the headers field.

Added in Steven Van Ingelgem's patch to add an HTTP status string to the HTTP server.

Fixed a bug in how the response headers were set.

added missing spec

updated spec slightly

Added an initial cut of the SURF and image keypoint finding code

Fixed compile error in gcc-3.3

Changed code to avoid compiler error in visual studio 7.1

Added the point_rotator object.

added some more tests

Added the rotate_point() function.

Made the centered_rect() spec more clear.

Switched the svm_nu_trainer object's kernel cache from using 
the type of scalar used by the user supplied kernel to just always
use float.

Switched all the test statements in the regression test suite from
DLIB_CASSERT to DLIB_TEST and DLIB_TEST_MSG.

Added DLIB_TEST and DLIB_TEST_MSG macros to the testing suite

Changed code a bit to avoid false test failure

Modified the thread_pool and bound_function_pointer so that they can 
easily deal with functions that take no arguments.

Added a missing typedef to the pipe object.

Improved the image_window a bit.

Added the image_window and image_display objects.

Added implicit conversions of std_vector_c back to std::vector so that
it can be used with functions that expect std::vector objects.

Fixed some typos in the spec files.

Fixed the code so that it doesn't assume sparse vectors are of std::map type.

Tweaked epsilon values to avoid false test failures.

Fixed the dictionary_size() member functions in the kcentroid overloads so
that they always return the correct numbers.

Added a lot more tests for the kcentroid object.

cleaned up the code

Added optimized overloads of the kcentroid object for various 
linear kernels.

Added some more stuff to the sparse vector functions.

Added a missing requirement and ASSERT.

Fixed a compile time bug in the offset_kernel.  Also fixed some
bugs in the sparse vector operations and added a few more new
operations as well.

updated comments.

Added a set of kernels that can operate on sparse vectors. 

Made spec more clear

Added Steven Van Ingelgem's patch to add the body of data posted back to the
server into the incoming data object given to the server_http::on_request() handler.

Made the specs more clear.

Changed the kcentroid object so that its default is to keep more linearly
independent vectors rather than newer vectors.

Broke the matrix_la.cpp file into four separate files.

Removed the pixel.cpp file and moved the few things that were
in it into pixel.h.  I did this to make sure that all the things
that #include pixel.h are always "header-only" since most of 
dlib is header-only already.

Merged in Steven Van Ingelgem's patch to cleanup the HTTP server and
add new functionality.  This breaks backwards compatibility with the
previous on_request() interface but it is easy to update old code and
it is now much cleaner.  

Added the strings_equal_ignore_case() functions.

Changed the kcentroid so that you can tell it to keep the most linearly 
independent vectors rather than the newest vectors.  I then changed the
svm_pegasos object so that it has a max number of support vector setting
so that the user can supply an upper limit on the number of support 
vectors to use.  Note that these changes broke backwards compatibility 
with the previous serialized format of these objects.

This object was throwing away the wrong vector the first time
it started to prune the set of independent vectors.  All subsequent
prunings were right, just not the first one.

Fixed a spelling error

Fixed incorrect comment.

Changed code to avoid a compiler warning in mingw

Made the specs more clear.

Made the spec more clear

Added some tests for the svm_pegasos object.

Optimized the kernel cache usage slightly.

Changed #includes to link to the abstract docs rather than
implementation files.

Cleaned up the pegasos code and filled out the spec file.

Added some missing requirements to a few functions.

Fixed compiler error in visual studio 7.1

Added a get_socket_descriptor() function to the connection class.

Made the std_vector_c able to copy directly from std::vector objects.

Fixed the code so that it works properly with the new row vector
range() rather than the previous column vector range().

Changed the cross_validate_trainer() function so that it catches
invalid_svm_nu_error exceptions and just gives them a bad score rather
than letting them get out and annoy the user.

  - Added some tests for the new linspace, logspace, and cartesian_product functions.
  - Updated old tests now that range() gives row vectors instead of column vectors.
  - Fixed a bug in subm() that showed up when statically dimensioned row vectors 
    were used to select a sub matrix.

  - Added the randm(), linspace(), logspace() and cartesian_product() 
    functions
  - Added an overload of pow() for pow(scalar, matrix)
  - Changed the range() function so that it returns row vectors
    instead of column vectors.  

Fixed invalid message in an assert statement.

Templated the matrix_range_exp so that it can use types other than long.

Added a try catch block to catch invalid_svm_nu_error exceptions.

Added missing std:: qualifier.

Added an initial cut of the pegasos stuff.

Removed unnecessary #include.

Modified the svm_nu_trainer so that it uses its kernel
matrix cache more efficiently.

Added some functions to the kcentroid to allow the user to 
compute the inner_product() of kcentroids as well as a few
other useful things.

Added the offset_kernel

Made the spec more clear.

Added a trace() function.

cleanup spec

Added some tests for the new min/max functions.

Added functions to find the index of the min and max element of a vector.

Fixed incorrect message in an assert.

Added the find_min_and_max() function.  I also changed the code in the
lu_decomposition::is_singular() function to avoid what appears to be a
bug in gcc 4.2.4 that caused one of the regression tests to segfault
when built with -O3.  

Added some functions to the rectangle to make it easy
to get the corner points.

Fixed a bug in the way Content-type was handled in HTTP posts.  
(See bug #2600378)

Fixed a bug in an assert statement and also changed the code to avoid
potential compiler warnings.

Adjusted the epsilon to avoid false positives.

Changed cmake file so that it can find the 64bit Intel MKL when 
appropriate.

Changed code to avoid a potential race condition during program termination.

Renamed cdiv to cdiv_ to avoid conflicts with code that #defines cdiv
to some macro.

  - Added a bool to protect the unregister_thread_end_handler() function from 
    being called from the destructor of a global object when the state object
    the unregister_thread_end_handler() function uses has already been destroyed.
  - Changed the thread pool so that it only gets destroyed if there aren't any
    outstanding threads when the program ends.  If there are threads still
    running user tasks then the thread pool never attempts to destruct.

Fixed a race condition in the logger object.  It was missing a needed call
to unregister_thread_end_handler().  What could happen in some scenarios is,
during program termination, a global part of the logger object could be destructed
when it still had outstanding thread end handlers registered to it.  

Cleaned up the way the BLAS prototypes were declared.

Added some missing comments and asserts.

Added the real_eigenvalues() function and also cleaned up a few things.

Moved all the linear algebra code into the matrix_la.h file.

fixed some compiler warnings and errors in visual studio

Fixed dlib::abs() so that it works right on complex matrices.

Added QR, LU, cholesky, and eigenvalue decomposition classes.

Fixed some errors in the requirements for calling the new rowm() and colm() 
functions.

Changed the reciprocal() function so that it can work with
complex numbers.

Fixed an incorrect comment.

Added an option to control linking with libpng

Made the thread destruction a little more robust.

   - Added overloads for rowm() and colm() that allow you to pick out
     less than an entire vector
   - Added the lowerm() and upperm() functions
   - Added the const_temp_matrix class.

Removed another errant cout.

Removed errant cout statements.

Checking in changes to the matrix object that allow it to
factor expressions containing trans() operators. 

Added a missing &

Made the specs more clear.

Fixed typo in the vector spec.

Made widget_group objects apply fit_to_contents() recursively when
widget_group objects are nested.  Also fixed some spelling errors in the
spec file.

Added some functions to the text_field to control input focus and
select text.

Removed confusing options.

Renamed the array::expand() function to array::resize().

Clarified the specs.

Fixed some confusing bits of the vector spec and also 
added a unary - operator.

Changed code to avoid a bug in visual studio 7.1

Added some comments

Added more type_safe_union tests.

Added the type_safe_union object.

Added some bogus header files to catch when XCode messes up
the project include paths.

Fixed typo in spec.

Added the text_box gui widget.

code cleanup

Changed the BLAS cmake file stuff.

Made a few of the tests more robust.

Changed code to avoid another visual studio compiler bug.

Uncommented some stuff.

Tweaked a few things.

Changed code so that it compiles in visual studio.

Changed a few things to avoid compiler warnings and errors in visual studio 7.1

Fixed the remaining known issues with the BLAS bindings and added
a bunch of related tests to the regression test suite.

Added the last few BLAS bindings and adjusted a few of the 
constants related to matrix expression cost and when the BLAS
bindings can become active.

Added more BLAS bindings.

Added some assignment overloads that loop over matrices in either row-major or
column-major order depending on the layout of the given matrix.

Added stuff to the BLAS bindings so that they can now bind
to statements with matrix sub expressions as the destination
matrix.

Added the layout_type into the matrix_type typedef.

Fixed some bugs and cleaned up the code.

Changed the code so that the implicit conversion of a matrix to 
a scalar can trigger any appropriate BLAS calls.

Commented out an unnecessary using namespace statement.

Added stuff to make matrix sub-expressions get bound into BLAS
calls.

Simplified the BLAS binding macro.

Added GEMV BLAS bindings.

Changed to avoid warning in visual studio.

Improved the BLAS binding system.  It should now break expressions up into
their proper basic BLAS function calls.

Setup the cmake files so that they can find and link to BLAS libraries
when available (on UNIX, no windows CMake BLAS stuff yet).

Cleaned up the BLAS bindings more.

   - Added overloads to cause scalar multiplications to combine and percolate out
     of matrix multiplications.
   - Worked more on the optimized overloads that call BLAS functions.
   - Changed the code inside the matrix assignment overloads so that it works
     better with GCC's optimizer.

Added a few more things to the svn:ignore list.

Added some more tests

Cleaned up a few things and also fixed the compiler error in the 
tensor_product() function.

Clarified some comments.

Added a check to prevent the user from pasting a multi-line
string into a single line text_field.

Fixing more issues with visual studio 2003

Reorganized the matrix_assign code a little and also added in
some stuff to allow me to bind whatever expressions I feel like
to optimized BLAS libraries.

Moved the matrix subexpression stuff into its own file.

Added a column_major_layout for the matrix.

updated spec

More code cleanup.

More code cleanup.

General cleanup (also added all the needed typedefs for layout_type).

Optimized the matrix multiply a little more.

Updated the assign_matrix stuff so that it works again.

Switched the gui stuff over to the new 2D dlib::vector object.

Clarified the vector spec and added some more tests.

Fixed compile error in gcc

Added a bunch of voodoo to appease the bugs in visual studio.  Also added
promotions to the global operator* overloads for the vector.

  - Made the vector class inherit from matrix
  - Added code to make sure the vector class does the 
    appropriate type promotion when vector objects
    with different types are used together.

Adding Steven Van Ingelgem's cmake patch to clean things up and add
shared builds.

Added more constructors to the matrix expression objects so that various potential
errors are avoided entirely.

Added something to suppress warnings from visual studio.

Refactored a bunch of the matrix code into a cleaner form.  I also added a
cost member to all the matrix expressions so that the complexity of an 
expression can be measured and temporaries introduced when appropriate.

Added the is_convertible template.

Added copy and paste support to the text_field and also cleaned up some 
of the popup_menu stuff.

updated makefile

Changed the vector code so that it compiles in cygwin and visual studio.

Cleaned up the vector and point classes.  Now there is only one class, the vector,
class and it is capable of representing everything the old vector and point
class could. 

Made the matrix rounding functions a little more general.  They
are now able to be called on integral types and just do nothing
in that case.

Cleaned up the PCA code slightly.

  - Changed the range() stuff so that it works with ranges of the form
    range(1,5) as well as range(5,1)
  - Added overloads of colm(), rowm(), set_colm(), and set_rowm() that
    can work with ranges.

Cleaned up the matrix_assign() stuff a little more.  Also added some stuff
particular to gcc to control how things get inlined.

  - Moved the matrix_assign() totally into the matrix_assign.h file
  - Updated matrix regression tests.

Updated the diag() spec.

Changed the comma based matrix assignment to use operator= instead
of operator<<.

Split the matrix tests into two files to speed up compilation of the regression tests.

Fixed compiler warning

Added a more convenient way to initialize a matrix using the operator<<.

Added another matrix test.

Simplified the matrix_assign() function a bit.

Fixed typo in spec

Added some code to cause the user to get a compiler error if they accidentally
try to combine a quantum gate with an integer.

Fixed a bug in the apply_gate_to() function.  Also clarified the spec.

Changed code so that it compiles in gcc 4.3

Changed the code slightly to avoid warnings from gcc 4.3.

Changed the matrix so that it isn't forced to use the one matrix_data object.  Now
it has a template argument that lets you supply any kind of matrix layout object
you want.

Changed the code to avoid a compiler warning from the Intel compiler.

Added an overload of set_subm() that can operate on ranges.

Fixed some comments.

   - Added the tmax, tmin, and tabs templates
   - Changed the diag() function so that it is allowed to take
     non-square matrices.

Added the range() function and an overload of subm() that allows you to
pick out slices of matrices like in Matlab.

Added some more things to the svn:ignore list.

Fixed a problem with draw_line where it didn't always redraw the line properly.

Added a function to the gate_exp() to return a matrix object for the gate.

Fixed the deadlock I just introduced a few minutes ago.

Added missing mutex lock

Fixed typo in spec

Changed the future object so that its destructor waits 
for any tasks in a thread_pool that use the future to stop 
before letting a future go out of scope.

Fixed typos in spec

Fixed a bug in the measurement functions and also cleaned up the spec.

More code cleanup

Cleaned up the code more and fully filled out the spec.

Cleaned up this code a little.

Adding a rough version of the quantum computing simulation code.

Clarified the spec for the thread_pool.

Added some more tests for the thread_pool object.

Added the ability to use function objects as tasks and also the ability to
set the number of threads in a thread pool to 0.

Added the ability to bind function objects as well as normal C style functions 
and member function pointers.

changed code so that it compiles in visual studio

Added tests for the thread_pool and future objects.

Changed code to avoid compiler warning.

Added futures support to the thread_pool object.

Cleaned up a few things

Added the bound_function_pointer object.

Fixed grammar in comments.

Fixed a minor bug in how the zoomable_region widget 
drew itself after a resize in some cases.

Added comments to clarify when exceptions are thrown since I just changed 
the definition of a missing throws block in the library introduction.

Fixed incorrect #include

Changed the fatal_error exception class so that it aborts your program
and prints a message if you try to construct it more than once since 
doing so indicates that you ignored the first fatal error.

Fixed compile time warning.

Changed the code to suppress a potential compiler warning.

made the specs more clear

Fixed grammar

   - Changed cmake file so that it won't print messages about looking for libpng
   - Added some missing #includes to the thread_pool header

Fixed some typos in the svd specs

Added overloads for listener::accept() that can take scoped_ptr 
objects as well as normal connection pointers.  Also reconciled
the type used for timeouts between the win32 and posix version
of the code.

Added a missing text_field::set_style() to the spec file.

Changed a few things to avoid warnings in visual studio

Fixed a typo in the spec for the std_vector_c object.

fixed compiler warning in visual studio

Cleaned up the code for converting between hostnames and IP addresses 
a little bit.  

Finished integrating the new svd from Jack Riddle.

Added missing requires clause.

Changed code slightly to avoid a bunch of compiler warnings from gcc 4.3

Added missing assert to the thread_pool object

Made the spec more clear

Added the thread_pool object

Changed code so that it compiles in visual studio

Made the member_function_pointer more robust to the amount of memory in its stack
based block of memory not being big enough.  We should now be guaranteed that it
will either fit or it won't compile.

Added more matrix tests

Added the diagm(), svd2() and svd3() functions.

Added functions to the vector_normalizer object to allow you to 
inspect the mean, stddev, and pca matrices it contains.  Also 
changed it so that, if you ask it to train pca, then it always
uses the pca matrix even if it doesn't result in a smaller
output vector.

Removed all the makedepend stuff from the makefile since it just
ends up confusing people and you have to regenerate it for different
machines most of the time anyway.

Made sure the destructor for mp_base_base gets called before anyone clones 
data into the mp_base_base's memory space.  

Updated the spec to reflect the fact that member function pointer 
objects no longer make memory allocations and thus will never
throw bad_alloc.

Added something to the base64 object to allow the user to 
control how it writes out end of lines.

Fixed a compile time bug in the pinv() function.  It didn't compile 
when used on statically sized matrices when they weren't square.

Removed dangerous implicit casts between unrelated member function pointer types.
Also made the spec more clear.

Renamed is_const to is_const_type.

- added is_const template
- changed the member_function_pointer_kernel_1 slightly so that it
  compiles in visual studio.

Changed the member function pointer object so that:
   - It never calls new or delete
   - It can point to const member functions
   - It has an operator bool and operator! so that it can now
     be used in an if statement like a normal pointer

Changed the base_window so that it doesn't have any requirement that it not
be closed before calling its member functions.  Now doing so is just a no op.

Fixed cmake files so they still work with the old version of cmake.

Changed some of the events that are about the mouse leaving
a widget so that they still trigger even if the widget has been disabled
or hidden.  

Made the graph specs more clear

Added a user settable style to the text_field widget.

Renamed stuff so that this still compiles

Changed code to avoid compiler warnings.

made svd spec more clear

Added some anti-aliasing to the draw_line function.

Removed all the arrow button styles and replaced them with a single 
button style.

Cleaned up the list_box's objects code by making it use the scrollable_region widget.
Also gave the list_box a user settable style.

changed svn:ignore settings

Made the button_style_toolbar1 look nicer and also gave the button_style 
interface the ability to draw outside a button.

Fixed a bug in the fill_gradient_rounded() function. It didn't always
draw the entire rectangle.

Removed the confusing and unnecessary hidden bool argument to the style
drawing functions.

Made the scrollable_region and zoomable_region widgets have user settable styles.

updated spec

Renamed dragable to draggable.

- removed scroll_bar::set_orientation()
- added the dragable::on_drag_stop() event
- added the dragable::is_being_dragged() function
- Made the scroll_bar have a user settable style and made a default 
  style
- Added some missing mutex locks to the scroll_bar widget

Removed the arrow_button widget and moved its functionality into the 
button widget.  I also added 4 new button styles that correspond to 
the 4 types of arrow button.  Lastly, I had to move some things around 
so that is why there seem to be a lot of code changes in this commit.

Changed cmakelists.txt so that cmake 2.6 doesn't give a warning.

Moved the dlib_assert_breakpoint() call in the assert macro to the front.  This avoids
the case where your program has had its heap corrupted and is very nearly about to crash,
and so when an assert triggers it crashes on the code that constructs the error 
message because the heap is hosed.  Thus preventing you from hitting a break point
on dlib_assert_breakpoint().  So this change mitigates that.

changed svm test a little

Changed the code to avoid a compiler warning.

Changed code to suppress compiler warnings

Moved a bunch of code from the base_widgets.h file to the base_widgets.cpp file.

Moved a bunch of stuff from the widgets header file to the widgets cpp file.

Cleaned up some more gui code.  I fixed a memory leak in the native_font object, 
made the get_native_font() function return a shared_ptr_thread_safe object instead
of a raw pointer, and also made the letter object noncopyable (as it should have been).
I also moved some code from the widgets header file to the cpp file.

Removed an unnecessary COMPILE_TIME_ASSERT statement from the
svd function.

Added a tensor_product() function for the matrix object.

Fixed an #include statement so that it works even if you don't
specify an include path argument to your compiler.

Removed the forced setting of DLIB_NO_GUI_SUPPORT in the cmake file
since it is sort of irritating.

Fixed a compile time error and also added back in the call to start the
event_handler_thread in the gui core that I accidentally deleted a few minutes ago.

changed the shared_ptr_thread_safe object so that it unlocks its
internal mutex before calling delete on the pointer that it contains.

Cleaned up more of the gui core code

Cleaned up the gui_core code.  Also made it so that the gui event handler thread
isn't created at all unless some part of an application calls some of the
gui_core code.

made the spec more clear

Added some missing asserts

Added the normalized_function object.

Added the vector_normalizer object

Added the sort_columns() and rsort_columns() functions

added another svn:ignore item

Removed the default argument to the is_dlib_thread() function and 
instead added an overload that doesn't take any arguments.  I did this
because -O3 sometimes makes gcc ignore more complex default arguments. 
I don't actually have any cases where this happened to this function
but I'm changing it just to be safe.

Fixed more Intel compiler incompatibilities.

Fixed a bunch of compiler warnings and errors in the
Intel compiler.

Added some things to the initializer lists of some exception classes because
without them you get errors when compiling with the Intel compiler.

cleaned up the code

  - Made the spec for the rank_features() function a little more clear.  Also made
    the implementation do recursive feature elimination when the user tries to
    rank all the features.
  - The report format that comes out of the rank_features() function is now
    also slightly different.

Fixed a bug in the covariance() function that prevented it from compiling
sometimes.

Fixed typo in spec

Added the popup_menu_region widget.

Added more control over how the scrollable_region scrolls its region.
You can now adjust how much it scrolls when the mouse wheel is scrolled 
as well as enabling scrolling via a mouse drag.

Fixed another destruction order problem, this time in the gui_core.

Fixed a destruction order problem in the timer object and also added
some extra mutex locks to the global singleton in the gui core.

Changed all the font* pointers to shared_ptr_thread_safe objects.

Added a thread safe shared pointer object

Removed the comments about doing stuff before main() has been entered
since this is now mostly untrue.

Made the X11 version of the gui_core work with global window objects.

Changed the gui core code around so that it should be safe to make window 
objects at the global scope

Removed the register_program_ending_handler() function from the threading API
and also made the dlib thread pool not block the termination of the program
if there are still threads executing.

Made the on_wheel_down() and on_wheel_up() gui events take a new
argument that lets you know the status of any buttons currently 
depressed

Made one of the arguments to font::draw_string() not be a reference because
some versions of gcc don't end up doing the right thing when -O3 is supplied.

  - Fixed a bug in the scrollable_region widget that caused it to scroll in an
    unpleasant way when the horizontal and vertical scroll increments weren't set
    to the same value.
  - Generally made the specs more clear and added some missing requires clauses.

Changed the name of the mutex in the threaded_object and multithreaded_object 
so that it won't cause any ambiguities when you derive a class from a
gui widget and a thread base class.

Fixed a bug in the linearly_independent_subset_finder object.  Also
added a way to set a minimum tolerance.  This also breaks backwards
compatibility with the previous serialization format for the object.

Cleaned up the kkmeans class and made it actually use
the min_change parameter.

Fixed some of the errors in the X11 gui stuff that
come out of valgrind.

Changed code a little to avoid compiler warnings about 
uninitialized variables.

Fixed a bug in the directory navigation gui.  If you tried to go
into a drive on windows that wasn't mounted you got an error.
This is now fixed.

 - made sockets test compile in vc7
 - made the dir_nav stuff compile in vc, cygwin, and mingw

Made the dir_nav stuff work with std::vector and dlib::std_vector_c 
as well as dlib::queue objects.

Added an is_std_vector to the is_kind.h file.

cleaned up the sockets test

Added some overloads of the create_connection() and create_listener()
functions that use scoped_ptr objects instead of just plain pointers.

cleaned up the code

changed slightly to avoid a warning in visual studio

Fixed another bug in the thread_specific_data object.  It should now
work right regardless of the destruction order of the relevant global
bits of state hanging around.

changed the tests a little

updated makefile

Added some regression tests for the svm/kernel methods stuff

Cleaned up the rvm code and made the regression version more 
numerically robust.

Made the rvm classifier more numerically robust

Organized the Unicode overloads into a more readable format.  Also added some missing
overloads to the spec file.

Added some overloads for move_rect() and translate_rect() that take a
point object.

Added the unregister_thread_end_handler() function and also used it to
fix a bug that can trigger when the thread_specific_data object is destructed.

Added an additional stopping condition to the kkmeans object.  It is now
possible to tell it to top when a certain fraction of centers don't change.

Fixed the serialization code for the kkmeans object so that it actually works.

Fixed some potential bugs in the rvm

Fixed the cmake option to toggle the ENABLE_ASSERTS macro

Added a missing mutex lock to the tooltip widget.  

Optimized matrix multiplication a little

Added some tests to catch problems with optimized versions of
matrix multiplication 

Changed code slightly to avoid another warning from valgrind.

Changed the code slightly to avoid a warning from valgrind.

Changed the test slightly to avoid warning from valgrind.

Added the spec for the new overload of close_gracefully()

Added some stuff to encourage the use of scoped_ptr with the connection
objects.

Refactored the matrix code so that all matrix assignments from a matrix_exp
ultimately happen inside the matrix_assign() function.

Made an example less confusing.

Fixed a potential bug in the hysteresis_threshold() function.

Added some extra checks in the stopping conditions to avoid
getting stuck while training an rvm.

added a spec for the rvm_regression_trainer 

fixed a spelling error

worked on the rvm regression

Added an initial cut of the rvm stuff

optimized the squared and cubed functions a little

Added more error handling to the nativefont.  Also changed the
code so that the native font object isn't instantiated unless 
you try to use it for something.

Added the length and length_squared functions.  Also cleaned up
the sigmoid function's code and spec

Fixed some typos in the matrix utilities spec file. 

turned the test_trainer function into the test_binary_decision_function

Made the reduced trainers copyable so that they can be used
with the threaded cross validation functions.

changed gui code a little so that windows don't popup in funny 
places when used with the cygwin X windows system.

added the test_trainer and cross_validate_trainer_threaded functions

Fixed grammar and make spec more clear.

Changed a few things so that the code works regardless of how you set the 
character set option in visual studio.

Changed test to avoid false error

Changed the default stopping tolerance in the reduced2() function
from 1e-5 to 1e-3.

Added a sigmoid_kernel object and also added a kernel_derivative
for the polynomial_kernel.

Made the spec more clear and also added some tests

Changed a few things so that the library works in cygwin.

Fixed the native font object so that it can find the X11 display 
on Mac OS X even when the application is run outside an xterm.

Made the dlib::vector object convertible to a dlib::matrix and
also gave it a constructor to convert a matrix to a vector as 
well.

fixed cmake options

Added a cmake option for defining ENABLE_ASSERTS

Added an option to the cmake lists file to enable or disable
the stack trace stuff.

Added get_kernel() member functions to the reduced set trainers.  
Also added missing comments to the spec.

Finished the reduced_decision_function_trainer2 object.  Also added
the kernel_derivative template.

Moved the reduced set stuff to its own file.  Also added a prototype 
version of another reduced set algorithm.

updated cmake file to avoid a warning from cmake

Improved the reduced_decision_function_trainer object by making it use
the linearly_independent_subset_finder.

Updated this header to include the linearly_independent_subset_finder

Changed the rbf_network_trainer to use the linearly_independent_subset_finder
to find centers. 

Cleaned up the code a bit

Added the linearly_independent_subset_finder object

Made the spec more clear

Fixed compiler error I just introduced in this file.

cleaned up this spec a little

Added some missing member variables to the .swap and serialization 
functions.

Added a check for division by zero to the conjugate gradient
optimization functions.

Removed a bug in the sum() and variance() functions that
caused them to seg fault when they were used on certain 
matrix of matrix objects.

Made the members of the function object non-const since
that was sort of obnoxious.

Added the inv_upper_triangular() and inv_upper_triangular()
functions.

Made the rbf_network's template argument be a kernel type
instead of a sample type.

made the specs more clear

Removed unnecessary restrictions on the rbf_network_trainer
object.

Added an RBF network trainer

Optimized the pinv function a little

Removed an unneeded and potentially numerical instability 
inducing std::abs call.

Added some missing requires clauses and asserts

Added the reduced_decision_function_trainer object and 
reduced() function.

made the spec more clear

Added the distance_function object

Optimized the kcentroid code a little by making it wait
to compute the bias term until it actually needs it.

turned some tabs into spaces

Added some missing asserts and cleaned up some things

Refactored a bunch of the svm training code into a much cleaner form.

Made the specs more clear

made the rank_features() function work with any kind of 
vector like object

Made it so the kkmeans train function can take any kind of
vector container.  Not just dlib::matrix.

Made it so you can pass a dlib::matrix to the array_to_matrix
and vector_to_matrix functions.

Added the rank_features() function

made spec more clear

Added some overloads of the randomize_samples() functions to take
std vector objects and to take just one vector of things without
labels as well.

Added a missing requires clause to the kkmeans object.

Fixed the running_stats object so that the variance can never be negative.

Added the ability to compare kcentroid objects to each other

Made the kernel objects comparable via operator==

Oops.  Put an important #include back

Applied another Unicode patch from Keita Mochizuki.

fixed out of date spec

Fixed a bug in the min and max calculation. 

set the subversion eol-style to native for these files

fixed some compiler warnings

Added a min() and max() to the running_stats object.

Fixed some grammar and typos

minor code cleanup

changed to avoid compiler warning in visual studio

Added the running_stats object

Just code clean up stuff like making sure everything uses the right 
templated scalar types and checking for degenerate numerical conditions 
and so forth.

Added some missing requires clauses to the max and min functions.

Changed code slightly to avoid compiler error in some versions of gcc.

Fixed typo in spec

Merging in the unicode branch.  The change log for that branch is as follows:

	------------------------------------------------------------------------
	r2331 | davisking | 2008-06-15 15:03:05 -0400 (Sun, 15 Jun 2008) | 1 line
	Changed paths:
	   M /branches/unicode/dlib/CMakeLists.txt

	Added another link library to the windows part of the cmake file
	------------------------------------------------------------------------
	r2330 | davisking | 2008-06-15 14:11:47 -0400 (Sun, 15 Jun 2008) | 3 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_abstract.h
	   M /branches/unicode/dlib/gui_widgets/base_widgets.cpp
	   M /branches/unicode/dlib/gui_widgets/base_widgets.h
	   M /branches/unicode/dlib/gui_widgets/base_widgets_abstract.h
	   M /branches/unicode/dlib/gui_widgets/drawable.cpp
	   M /branches/unicode/dlib/gui_widgets/drawable.h
	   M /branches/unicode/dlib/gui_widgets/fonts.cpp
	   M /branches/unicode/dlib/gui_widgets/fonts.h
	   M /branches/unicode/dlib/gui_widgets/widgets.cpp
	   M /branches/unicode/dlib/gui_widgets/widgets.h
	   M /branches/unicode/dlib/gui_widgets/widgets_abstract.h

	Added Keita Mochizuki's name to a bunch of files.  Also added a missing
	prototype for the get_native_font() function in fonts.h

	------------------------------------------------------------------------
	r2329 | davisking | 2008-06-15 14:05:17 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_widgets/fonts.cpp
	   M /branches/unicode/dlib/gui_widgets/fonts_abstract.h
	   A /branches/unicode/dlib/gui_widgets/nativefont.h

	Added the nativefont object from Keita Mochizuki

	------------------------------------------------------------------------
	r2328 | davisking | 2008-06-15 14:04:44 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp

	Removed an unused variable

	------------------------------------------------------------------------
	r2327 | davisking | 2008-06-15 13:33:31 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.h
	   M /branches/unicode/dlib/gui_widgets/base_widgets.h
	   M /branches/unicode/dlib/gui_widgets/base_widgets_abstract.h
	   M /branches/unicode/dlib/gui_widgets/drawable.cpp
	   M /branches/unicode/dlib/gui_widgets/drawable.h
	   M /branches/unicode/dlib/gui_widgets/drawable_abstract.h
	   M /branches/unicode/dlib/gui_widgets/widgets.cpp
	   M /branches/unicode/dlib/gui_widgets/widgets.h

	Converted tabs to spaces

	------------------------------------------------------------------------
	r2326 | davisking | 2008-06-15 13:31:32 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_abstract.h
	   M /branches/unicode/dlib/gui_widgets/drawable.cpp
	   M /branches/unicode/dlib/gui_widgets/drawable.h
	   M /branches/unicode/dlib/gui_widgets/drawable_abstract.h
	   M /branches/unicode/dlib/gui_widgets/widgets.cpp
	   M /branches/unicode/dlib/gui_widgets/widgets.h

	Checking in the on_string_put() event patch from Keita Mochizuki.

	------------------------------------------------------------------------
	r2325 | davisking | 2008-06-15 13:13:49 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_widgets/base_widgets.h
	   M /branches/unicode/dlib/gui_widgets/base_widgets_abstract.h
	   M /branches/unicode/dlib/gui_widgets/style.cpp
	   M /branches/unicode/dlib/gui_widgets/style.h
	   M /branches/unicode/dlib/gui_widgets/style_abstract.h
	   M /branches/unicode/dlib/gui_widgets/widgets.cpp
	   M /branches/unicode/dlib/gui_widgets/widgets.h
	   M /branches/unicode/dlib/gui_widgets/widgets_abstract.h
	   M /branches/unicode/dlib/serialize.h

	Added the first part of the UNICODE patch from Keita Mochizuki 

	------------------------------------------------------------------------
	r2324 | davisking | 2008-06-15 13:05:17 -0400 (Sun, 15 Jun 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_widgets/fonts_abstract.h

	Fixed a typo in a comment

	------------------------------------------------------------------------
	r2244 | davisking | 2008-05-15 17:40:47 -0400 (Thu, 15 May 2008) | 3 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/unicode/unicode.cpp
	   M /branches/unicode/dlib/unicode/unicode.h
	   M /branches/unicode/dlib/unicode/unicode_abstract.h

	Put the convert_wstring_to_mbstring() function back in but in a way that 
	works with mingw.

	------------------------------------------------------------------------
	r2219 | davisking | 2008-05-04 11:13:23 -0400 (Sun, 04 May 2008) | 1 line
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp

	changed to be the same as the windows version
	------------------------------------------------------------------------
	r2218 | davisking | 2008-05-04 10:42:28 -0400 (Sun, 04 May 2008) | 4 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/unicode/unicode.cpp
	   M /branches/unicode/dlib/unicode/unicode.h
	   M /branches/unicode/dlib/unicode/unicode_abstract.h

	Moved some more stuff around and cleaned up the code.  The wcstombs got 
	moved into the gui_core_kernel_2.cpp file since I removed it from the 
	unicode.cpp file.  This was because mingw doesn't seem to support it.

	------------------------------------------------------------------------
	r2216 | davisking | 2008-05-03 16:03:40 -0400 (Sat, 03 May 2008) | 1 line
	Changed paths:
	   M /branches/unicode/dlib/algs.h
	   M /branches/unicode/dlib/string/string.h
	   M /branches/unicode/dlib/unicode/unicode.h

	Changed code to avoid compiler trouble in mingw and inane warnings in 
	visual studio.
	------------------------------------------------------------------------
	r2215 | davisking | 2008-05-03 10:22:35 -0400 (Sat, 03 May 2008) | 3 lines
	Changed paths:
	   M /branches/unicode/dlib/unicode/unicode.h

	Added a char_traits for dlib::unichar since gcc 3.3 (and probably others) 
	won't compile without it.

	------------------------------------------------------------------------
	r2214 | davisking | 2008-05-03 10:21:32 -0400 (Sat, 03 May 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_widgets/style.h

	Added a missing #include

	------------------------------------------------------------------------
	r2213 | davisking | 2008-05-02 15:11:23 -0400 (Fri, 02 May 2008) | 3 lines
	Changed paths:
	   M /branches/unicode/dlib/string/string.h
	   M /branches/unicode/dlib/unicode/unicode.cpp

	mingw doesn't have a std::wcstombs so I replaced calls to it with calls to 
	dlib::narrow.  Also changed dlib::narrow so it works in mingw as well since 
	the previous version ended up calling something that mingw doesn't have as 
	well.   
	------------------------------------------------------------------------
	r2212 | davisking | 2008-05-02 11:23:41 -0400 (Fri, 02 May 2008) | 3 lines
	Changed paths:
	   M /branches/unicode/dlib/all/source.cpp

	Applying more of Keita Mochizuki's unicode patch.


	------------------------------------------------------------------------
	r2211 | davisking | 2008-05-02 11:22:58 -0400 (Fri, 02 May 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/unicode/unicode.h
	   M /branches/unicode/dlib/unicode/unicode_abstract.h

	updated specs a little

	------------------------------------------------------------------------
	r2210 | davisking | 2008-05-02 11:16:33 -0400 (Fri, 02 May 2008) | 2 lines
	Changed paths:
	   A /branches/unicode/dlib/unicode/unicode.cpp
	   M /branches/unicode/dlib/unicode/unicode.h

	Applying more of the unicode patch from Keita Mochizuki 

	------------------------------------------------------------------------
	r2209 | davisking | 2008-05-02 11:10:19 -0400 (Fri, 02 May 2008) | 2 lines
	Changed paths:
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_1.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.cpp
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_2.h
	   M /branches/unicode/dlib/gui_core/gui_core_kernel_abstract.h

	Applying the first part of the unicode patch from Keita Mochizuki 

	------------------------------------------------------------------------
	r2208 | davisking | 2008-05-02 10:49:15 -0400 (Fri, 02 May 2008) | 3 lines
	Changed paths:
	   A /branches/unicode (from /trunk:2207)

	Adding a branch to deal with the unicode patch from Keita Mochizuki.

Made the stack trace code a little cleaner

Added a comment to cite the kmeans++ paper

Added the pick_initial_centers() function

Added a get_kernel() function to these objects.  Also fixed some incorrect
parts of the specs.

Added a max iteration argument to the kkmeans train function.

made the code a little more clear

Renamed the get_tolerance() and get_max_dictionary_size() functions
so that they don't start with get_ to be a little more consistent.

   - Added a max dictionary setting to the kcentroid object
   - Removed the "discount" thing and replaced it with a 
     much less confusing pair of scale arguments to the 
     training functions.
   - Changed the kcentroid's serialization format.

   - Fixed a minor numerical error in the krls code so now it gets slightly better
     results.
   - Added the ability to cap the number of dictionary vectors used by the krls
     object at a user specified number.  This changes the serialization format
     of the object.  I also removed the function to set the threshold after
     the object has been constructed.

Made the specs a little more clear and added some more tests

Fixed a typo in the spec for round_zeros

  - Fixed some issues with the requires clause of the removerc function.
  - Added the remove_row and remove_col functions.  Also made all
    three of the above functions capable of taking arguments at 
    run time as well as compile time.

Added some clarifying comments to the svd function.

Fixed an aliasing bug in the set_subm(), set_rowm(), and set_colm() 
functions.  It was possible that you could get incorrect results
if you used these functions to copy one part of a matrix to another
part of the same matrix if the two areas overlapped.

Fixed compiler error I added to the matrix a moment ago.

Changed the matrix operations so that they result in shorter type names
when compiled.  This avoids problems in compilers like visual studio.

Added an initial cut of the http client code from Steven Van Ingelgem.

Added some missing assert statements.

Modified the radio_button style so that it looks nicer.

Fixed typo in spec

Cleaned up the kcentroid code a bit.

Cleaned up the code for the krls object.

Added some more tests to the matrix regression test.

Fixed some bugs in the require clause of the subm and set_subm functions.
Also suppressed some compiler warnings.

Added the set_subm, set_rowm, and set_colm functions.

Added some asserts

Added the kkmeans object.

Changed the checks in the matrix object so that you are allowed to 
assign a matrix of matrices to another matrix of matrices so long
as long as the assignment makes sense.  Previously you were only
allowed to perform this assignment if the inner most matrix types
were exactly the same.

Added two new events to the text_field object.  One for detecting when the 
user hits enter and another for detecting when input focus is lost.

Changed code to avoid a compiler warning.

Changed tests so that they compile in visual studio.

Fixed typo in spec

Made the code more portable

Added the wrap_function and is_function templates.

Added versions of the find_min_* functions that don't take
a derivative function.

Made the optimization functions not return the value of the objective
function.

saving optimization test

Added an empty test file for the optimization stuff

Removed unneeded stuff

Fixed a typo in a comment

Renamed some functions.

Added a bunch of unconstrained optimization stuff to the library.

Fixed some spelling errors

Made the is_matrix template a little more robust.

Added an identity_matrix() that can take a runtime defined size.

Made a comment more clear

renamed the one_class object to kcentroid

Added a function that lets you test and train at the same time

Added the max discount parameter to the one class algorithm as well
as cleaned up the code a bit.

Added a requires clause to the set_tolerance() member function and
renamed clear() to clear_dictionary().

Added a simple kernel based one class classifier.

Fixed incorrect comments.

Changed the stack trace stuff so that it doesn't perform memory 
allocations or copy strings.

Added a missing dlib:: qualifier

Added a missing scope resolution statement.

Made the stack trace stuff more robust

fixed typo in comment

Added stack trace stuff

Added missing comments and fixed some existing ones.