Nonlinear Time Series Routines
(FORTRAN branch)
List of changes
TISEAN_2.0 -> TISEAN_2.1
- The input routines for time series had to be changed. A safety feature made
them too slow on many platforms. In the new version, the user has to ensure
that the file has enough columns for the task planned since otherwise,
missing data is read from the following line(s).
- An internal change worth mentioning is the calling sequence for
neigh. The array containing the time series
has to be given twice. The reason is that this allows two different arrays
to be specified. Distances across series are necessary for xc2.
- The program c1 now accepts mutivariate time series
and can perform mixed embeddings. Also, embedding dimension 1 is now
allowed. As a further important change, the center points are now selected
randomly ratehr than taken to be the first points in the set. The comparison
is however still performed with points starting from the beginning of the
series.
-
In project,
predictlazy,
and xc2,
the restriction that the embedding dimension must be at least 2 has been
removed.
-
The Wiener filter wiener1/2 has
been improved by offering a smoothed spectral estimator rather than raw
periodograms.
TISEAN_1.2 -> TISEAN_2.0
-
Utilities for the conversion of inter-event intervals to event times and
vice versa have been added. They are called events
and intervals.
-
noise is no longer supported. If you are
desperate, you can find it still in the unsupported directory.
-
autocor3 is no longer supported. If you are
desperate, you can find it still in the unsupported directory. I never used it myself and
it was in C. It had to go.
-
lyapunov is no longer supported due to its
redundancy with lyap_k. If you are
desperate, you can find it still in the unsupported directory.
-
surrogates has been rewritten in FORTRAN
with a slight change in its functionality.
-
timerev has been rewritten in FORTRAN,
same functionality.
-
addnoise
can now run without input files, just as a random number generator.
-
There are three new members in the randomize
family, one for unevenly sampled data and two suitable for inter-spike
interval series.
- An interference between option names has been removed. In the randomize family, -w has been changed to -u.
-
The output of the correlation sum programs has been partially
restructured. Seperate files for different embedding dimensions are no
longer written, everything come in a single file, seperated by double blank
lines. Thus, c1, c2, and
c2naive are now compatible with
d2. These changes also affect
c2d,
c2g, and c2t,
which thereby also become usable for d2.
Writing to stdout is now the default and stdin is acceptable as input file.
- The output of stp can now be better
controlled, specifying the spacing of level lines and the number of time
steps.
-
The output of the correlation sum programs has been partially
restructured. This applies to xc2 as
well. Also, rather than reading from two files, now two columns from the
same file are read. The latter also applies to
compare.
- The program randomize_generic_exp_random
was quite useless and has therefore been removed.
- As a personal service to Remo Badii, a "feature" of c1 was removed that caused small values
of k to be computed more than once, a result of having k discrete but in a
logarithmic loop.
- The module structure is changed slightly to enable the
future incorporation of multivariate data. Please check the new extension guide.
- A new program is available to cluster a
dissimilarity matrix.
TISEAN_1.0 -> TISEAN_1.2
-
An oddity with the command line parsing of
c2t and
c2g has been fixed.
-
A back end to ar-model
has been added that runs an
autoregressive model. It is called ar-run.
TISEAN_0.9 -> TISEAN_1.0
- A few of the FORTRAN programs confused one compiler due to the order
of declarations, which has now been changed to be conservative.
- Depending on the number of data points, the necessary zero padding in
wiener1
was awkward, if not outright wrong. This has been fixed now.
TISEAN_0.5 -> TISEAN_0.9
- Changed the output file behaviour in accordance with what the Dresden
programs do: By default, most programs write to stdout now, independent of
the input source. The option -o tells the
program to build file names from the input file name, or "stdin" if
appropriate. Finally, -o outfile
specifies an output file name explicitly. Somewhat awkwardly, this will be
the same for all input files processed and results are overwritten if you don't move the files fast enough.
This behaviour may be changed in later releases once I have a good idea
what else to do.
- Quite a number of options have changed in the course of the re-unification
process. Most propminently,
-r/-R
replaces
-e/-E
for absolute lengths scales.
- Gradually, more and more programs respond meaningfully to the -V (verbosity) option.
TISEAN_0.4 -> TISEAN_0.5
- Recovered correct version of
c2 which was
(slightly) broken in release 0.4.
- The realisation of randomize that
implements the
autocorrelation function suffered from a last-minute change that didn't
quite work out and was unusable. Now fixed.
- All programs interpret anything which is not a valid option or its value as
a file name. This allows to give file names at arbitrary places on the
command line, a tiny step towards re-unification of the Dresden and
Wuppertal branches.
- All programs, when called without a file name, act as if a "-" was given,
and read from stdin if that is appropriate. As a side effect, a few
programs have to be called with -h in
order to see the command line options, since a call with no arguments would
be a valid execution.
TISEAN_0.3 -> TISEAN_0.4
- Almost all (FORTRAN) programs now accept an option
-V# that specifies the level of
diagnostic output. The default is as a rule what was there in
TISEAN_0.3. Specifying -V0 now switches
off all but fatal error messages. No more reporting of file opening etc.
- Introduced utilities for usage message, removing repetitions of the same
code sequences through all sources.
- Unstable periodic orbits and their
stabilities (upo). Stabilities are still somewhat
experimental.
- Embedding of periodic orbits (upoembed).
- General constrained
randomization (randomize). Contains modules
for constraints involving linear correlations, others are planned. Own
programming efforts to specify exotic constraints are encouraged.
TISEAN_0.2 -> TISEAN_0.3
- timerev and
autocor3:
bug that prevented from reading input removed.
- Made parsing of commandline arguments more stable (hopefully): larger
numbers (up to 20 digits) are handled.
TISEAN_0.1 -> TISEAN_0.2
- pc prints out the
fractions of the variance covered by the principal directions.
- Recovered correct versions of
lazy and
project:
the difference in the second columns are now formed corerctly,
with respect to the raw data, not to the last iterate.