news

非常著名的曲线拟合程序

Goose v0.0.11
Released October 18, 1999

* Massive type changes
  + Removed all size_t from DataSet object interface.
  + Changed DataSet derivatives to have their elements indexed by ints,
    and introduced offsets so that objects can be indexed over arbitrary
    ranges of integers, rather than 0..(N-1).

* Jon added Permutation::permute(vector<X>&) function.

* Mikkel added some statistical tests: Kruskal-Wallis, Cochran's Q,
  McNemar, Spearman.

* Jon added the Hypergeometric CDF (and the inverse, too).

* Asger made various tweaks for Win32 compatibility.

* Jon finally added t-test, F-test (for equality of variances),
  confidence intervals for variances --- both functions to do the
  actual calculations, as well as a nice (well, I like it) object
  interface.  (in parametric_estimation.{h,cpp}).

* Jon also added various tests of estimating the probability of success
  for a set of binomial trials.  (in parametric_estimation.{h,cpp}).

* Flagged texinfo docs as being very out-of-date.  


Goose v0.0.10
Released Jul 12, 1999.

* Fixed a number of portability problems that were tracked down by
  Philippe Defert <[e-mail]>.

* Bugfixes due to Tom Shannon <[e-mail]>

* Added first hints of XML support.

* Renamed/revised some functions in Random.

* Added "convenience exceptions" to Exception.h, and began integrating
  them into the code.

* Catch-all random hacking.


Goose v0.0.9
Released May 28, 1999

* For configure purposes, the project was renamed to `goose' (from `libgoose').
  That way our tarballs will come out being called `goose-?.?.?.tar.gz'.
  (Suggested by members of the GNU ftp-upload mailing list.)

* Added calculations on empirical distribution functions: Kolmogorov &
  Smirnov's D, D+, D-, Kuiper's V.

* Added standardize() function to RealSet.

* Fixed some potential data/sorted data coherence problems in RealSet.

* Added more guile bindings.


Goose v0.0.8
Released May 27, 1999

* Another bump-the-version-number-for-build-purposes release by Havoc.


Goose v0.0.7
Released May 26, 1999

* Havoc fixed the build to check for guile properly, among other things.

* Jon added to the guile bindings.

* Jon added to the documentation.  Much of what he wrote quickly
  became obsolete.

* Jon wreaked hav... uh... took a chainsaw to the code, decrufting some stuff,
  breaking a lot of other stuff, splitting everything up into a bunch
  of different libraries, breaking and then repairing the build, and
  generally imposed his own ideosyncratic version of cleanliness.

* Jon added a lot of half-finished, undebugged stuff.  If you find new feature
  tucked in there that doesn't seem to work, don't assume that it is supposed
  to.

* Jon generally failed to document everything that happened.  He sucks for
  not keeping a ChangeLog or for updating NEWS more often.

* Jon also sucks for not releasing more often.


Goose v0.0.6
Unreleased

* At some point, for the purposes of getting the build to work
  properly, Havoc had to gratuitously increment the version number.


Goose v0.0.5
Released January 22, 1999

* As of this version, development has been CVSified.

* Everybody added lots of fixes and tweaks.

* Asger added exceptions, replacing a bunch of brain-dead assert()s.

* Havoc made some changes to DataSet to make modifying existing data
  easier.

* Jon improved inv_beta_cdf() some more.  It still doesn't really do the
  right thing for extreme parameters, but works quite nicely in all of 
  the usual, reasonable cases.  ("I'm not a numerical analyst, although
  I play one on TV...")

* Finally, an improvement that actually adds some new statistical capabilities!
  Jon added inference to the LinearRegression class, so that you can now
  generate confidence intervals around the estimates of the slope and
  intercept, as well as around the linear model's predictions.

* Jon added regression testing stuff.  (Regression *testing*, not regression
  in the statistical sense...)

* Asger added a bunch of portability stuff, so we can now build under Win32.
  Jon has mixed feelings about this, but recognizes that portability is
  a Good Thing.

* Asger added tons of improvements to the ASCII import mechanism.

* Jon added MetaCurve, which is a base class for handling a sampling and
  interpolation for 2d curves.  An example of its use is the
  GaussianKD class, which calculates gaussian kernel density estimates from
  a DataSet.  Later, the MetaCurve will also be used for non-linear
  regressions, like Loess curves.

* Asger added more improvements to the ASCII import thing.  It works for
  many files now, but a few more features are planned to make it the killer
  importer.

* Our class heirarchy got jiggled, and stuff got renamed.  The thing that
  used to be called a 'DataSet' is now a 'RealSet'.

* Jon trashed all of the existing numerics code and created a separate
  numerics library, called specfns, derived from on the PD cephes library.

* Havoc started a scm directory, with some preliminary doodles that will
  one day turn into guile bindings.

* Jon re-worked the bootstraping code so that it can use threads to
  distribute calculations between multiple CPUs.  If you have a multi-
  CPU box, this is a big performance winner, since those kinds of
  calculations are completely parallelizable.

* Lots of other stuff happened, but we forgot to report it.  Crufty old
  code has been taken out, and immature new code has been shoved in.


Goose v0.0.4
Released October 7, 1998

* Added gala, the Goose Acceptable Linear Algebra micro-package.  This
  replaces our previous, much grottier, matrix routines.  The new version
  is a bit slower, but (thanks to some hidden, ugly optimizations) it
  isn't really that bad.  (I get 170 40x40 inversions with the gala GMat
  class on my P2-300, versus 227 inversions/sec with the old, hacked-up
  code.)

* Added two (well, three) independence measures to DataSet:
  AR1_independence_z(), autocorr() and autocorr_z().

* Added StatsDB, a way to access descriptive statistics by name, via
  a string.

* Added inv_beta_cdf().  It has not yet been carefully tested for
  numerical stability.


Goose v0.0.3
Released September 19, 1998

* We're back to the LGPL, due to popular demand.  (Well, OK, it was just one
  person... but it was one person who was willing to contribute code!)

* Added Permutation class.

* Added permute() and sorting_permutation() functions to DataSet, making
  use of our fun new Permutation objects.

* Added add_at() and remove_at() to DataSet in order to let you insert
  individual elements at places other than the end, and to remove elements.
  The implementations of both are a bit gross, and both operations are
  somewhat inefficient... but you really shouldn't be doing this sort of
  thing.  DataSet != Array.

* Started putting in some assert()s where throws will eventually go.

* Added global variable common_random to Random.{cc,h}... a global Random
  object that routines can call if they don't need their own.


Goose v0.0.2
Released September 14, 1998

* Switched from LGPL to GPL, to facilitate code-borrowing from other
  free software projects involving math/statistics/numerical analysis.

* Added the a skeletal framework for documentation.  Nothing currently
  useful, though.

* Fixed sortedness() bug which mishandled the case of constant data
  in a DataSet.

* Changed DataSet to use reference-counting arrays w/ copy-on-write semantics
  for storing data.  This allowed lots of DataSet code to be simplified,
  since we no longer need to explicitly check for the case when data_ and
  sdata_ are aliased.

* Changed the interface to the bootstrap, jackknife, and regression routines
  to take advantage of the fact that we can now efficiently return DataSet
  objects from functions.  No more gratuitous dereferencing of returned
  pointers required!

* Added lots of new functions from statfn.h.  I've borrowed a lot of nice code
  from the R project (see http://fangorn.ci.tuwien.ac.at/R) that has
  really helped a lot.

* New functions in DataSet: mean_trimmed(), mean_winsorized(),
  exp_transform(), logit_transform(), pooled_mean(), pooled_var(),
  weighted_mean().
    
* Added objects for runs tests in runs.{cc,h}.

* Added support for the basic chi-square Goodness-of-Fit test in fit.{cc,h}.

* Added DataType abstract base interface; this holds all functions specific to 
  a particular type (currency, real numbers, dates, etc.)

* Changed DataImport parsing routines to be more flexible and more
  GUI-friendly. (stream import temporarily suboptimal)

* Optimized greater_than(), less_than(), equal_to() and between() to use
  sorted data.


Goose v0.0.1
Released August 21, 1998

* Initial release