-------------------------------------------------------------------
Fri Jan 27 05:53:39 UTC 2023 - Stefan Brüns <stefan.bruens@rwth-aachen.de>
- Update to version 2.8.0:
* Required Octave version number has been raised to 4.0.1.
* Test functions:
+ stk_testfun_hartman3.m, stk_testfun_hartman6.m: New test
functions (“Hartman3” and “Hartman6”) from the Dixon & Szego
(1978) benchmark.
+ stk_testfun_hartman4.m: New test function (“Hartman4”) based
on Picheny et al (2013), with a different scaling.
+ Most test functions in examples/test_functions are now also
available under the CC0 license (see each file).
-------------------------------------------------------------------
Sun Aug 21 22:21:17 UTC 2022 - Atri Bhattacharya <badshah400@gmail.com>
- Update to version 2.7.0:
* Required Octave version number has been raised to 4.0.0.
* Prediction:
- stk_predict.m now officially supports a shorter calling
syntax: `[zp, ...] = skt_predict (model, x_prd)` that can be
used either to compute predictions from any type of model
- @stk_model_gpposterior/stk_predict.m: no longer exists. The
base stk_predict function now provides the recommended
interface for all possible types of models, and then
dispatches to the (new) internal function stk_predict_ to do
the actual computations.
* Models: Removed get_input_data, get_output_data,
get_prior_model.
* Arrays:
- stk_dataframe arrays now possess a sample_size property,
which can be read using the usual dot-indexing notation or
through the (slightly faster) getter stk_get_sample_size.
- stk_model_gpposterior objects also have a sample_size
property, which corresponds to the sample size of the
underlying data set and can be read similarly.
- stk_length: deprecated (use stk_get_sample_size instead).
- @stk_model_gpposterior/stk_{set,get}_optimizable_parameters.m:
Fix a bug (erroneous property name).
* Sequential design of experiments: @stk_sampcrit_akg/set.m:
* Remove replicated input points when the AKG reference grid is
* construted using Scott's original method (i.e., taking past
* observations points plus the candidate point).
* Examples:
- stk_example_kb07.m: Add a second figure where nu is fixed to
5/2 and rho varies.
- stk_example_doe05.m: Make it possible to use the exact KG
criterion instead of Scott's approximation.
- stk_example_doe06.m: New example script, which illustrates a
simple sequential design strategy to estimate an excursion
set (failure region) and its volume.
* Miscellaneous:
- stk_init.m: Make STK startup faster.
- startup.m: Run stk_init automatically for in-place use.
- stk_select_optimizer.m, stk_optim_hasfmincon.m: Removed.
- README.md: Explain how to initialize STK when using the
Parallel Computing Toolbox in Matlab, by means of the
pctRunOnAll function.
- Internal: source formatting, comments, etc.
-------------------------------------------------------------------
Wed Nov 27 12:06:45 UTC 2019 - Stefan Brüns <stefan.bruens@rwth-aachen.de>
- Update to version 2.6.1:
* Documentation
+ stk_factorialdesign: Provide missing help text. (ticket #91)
+ Fix HTML doc (INDEX structure was broken).
- Update to version 2.6.0:
* Required Octave version
+ Required Octave version number has been raised to 3.8.
* Parameter estimation
+ stk_param_init: The syntax with an additional do_estim_lnv parameter
is now deprecated. Set model.lognoisevariance to NaN if you want an
initial estimate for the variance of the noise.
+ stk_param_init_lnv: Refuse noiseless models (error).
+ stk_model_gpposterior: Trigger parameter estimation when lognoisevariance
is NaN. In the (experimental) case where lognoisevariance is a noise
model object, parameter estimation is triggered as soon as at least one
of the optimizable parameters of the model is NaN. (ticket #93)
+ stk_param_relik: Return [] instead of NaN when the derivative with
respect to lnv is requested for a noiseless model.
+ stk_param_relik: Detect duplicated observation points automatically
and raise an error for noiseless models.
+ stk_param_getdefaultbounds: Estimation bounds for the regularity parameter
of the Matérn covariance function are now customizable through the options
nu_min_dimfun and nu_max_dimfun.
* Covariance functions
+ Deprecated: stk_noisecov.
* Models
+ stk_get_input_data, stk_get_output_data, stk_get_prior_model: New getters
for model properties, that replace get_input_data, get_output_data and
get_prior_model respectively (now deprecated).
+ stk_get_observation_variances: New getter, which returns the vector of
variances associated to the observations that have been used to build a model.
+ stk_gaussiannoise_, stk_gaussiannoise_het0: New classes representing Gaussian
noise models. Currently in experimental state, these two classes are merely a
proof-of-concept. Other noise model classes can be defined by subclassing
stk_gaussiannoise_. (ticket #38)
+ stk_example_misc05: Demonstrate the use of noise model objects to estimate
the dispersion parameter in an heteroscedatic case.
+ stk_simulate_noise: New function to simulate noise sample replicates.
* stk_dataframe and related classes
+ stk_dataframe: Accept char vectors (or strings) as colnames or rownames
argument when there is only one column or one row (ticket #92)
+ @stk_dataframe/sort: Argument dim can now be skipped, as in the base sort function.
+ @stk_dataframe/unique: Overload base function.
* Graphics
+ stk_plot1d: Do not create a legend systematically. (ticket #82)
+ stk_legend: New function to create a legend, using the graphical objects for
which a non-empty DisplayName has been provided.
+ stk_plot_shadedci: Change gray levels and improve legend.
* Miscellaneous
+ stk_options_set: Make it possible to set all options at once using an option
structure. Add help text.
+ stk_runtests: Return test results in a structure.
+ stk_sampling_nesteddesign: Fix help text.
-------------------------------------------------------------------
Wed Jul 18 08:48:23 UTC 2018 - badshah400@gmail.com
- Update to version 2.5.1:
* tk_dominatedhv: Workaround for Octave 4.4 issue (ticket #89)
* Documentation fixes.
* stk_option_set: Return option struct only if nargout > 0.
-------------------------------------------------------------------
Wed Sep 27 13:17:47 UTC 2017 - badshah400@gmail.com
- Update to version 2.5.0
* Required Octave version number has been raised to 3.6.0.
* Sequential design of experiments:
- `stk_sampcrit_akg_eval`: New function. Computes the
Approximate Knowledge Gradient (AKG) sampling criterion.
- `stk_example_doe05`: Example script that demonstrates the
use of the AKG and EQI sampling criteria for noisy
optimization.
- New (experimental) classes to represent sampling criterion
objects: `stk_sampcrit_ei`, `stk_sampcrit_akg`,
`stk_sampcrit_eqi`.
- `stk_sampcrit_ei_eval`: Remove deprecated calling forms.
* Design of experiments:
- `stk_factorialdesign`: Accept one-variable `stk_dataframe`
objects as factors and preserve column names in this case.
- `stk_sampling_nesteddesign`: New function to create nested
designs.
- `stk_sampling_nestedlhs`: New function to create nested LHS
(NLHS).
- `stk_sampling_sobol`: Fix help text.
* Validation of models:
- `stk_distrib_normal_crps`: New function to compare
observations and predictive (Gaussian) density.
- Calling `stk_predict_leaveoneout` with no output arguments
now automatically creates two cross-validation diagnostics
in two subplots: prediction VS observations (left panel) and
normalized residuals (right panel).
- `stk_predict_leaveoneout` uses now the virtual LOO formula.
* `stk_dataframe` and related classes
- `stk_hrect`: Preserve column names for `stk_dataframe`
inputs.
- `@stk_dataframe/find`: Overload base function to support
calling find with an stk_dataframe as first input argument.
- `@stk_dataframe/plotmatrix`: Overload base function to
enrich `plotmatrix` graphics with variable names when
possible.
* Logical functions:
- Operations that normally return logical (`lt`, `eq`,
`and`...) now return logical values for `stk_dataframe`
arguments.
- New overloaded functions for `stk_dataframe` objects:
`isinf`, `isnan`, `isfinite`.
* Testing array membership:
- `@stk_dataframe/ismember`: No longer assumes `rows` flag for
consistency with the base `ismember` function.
- `stk_factorialdesign/ismember`: New function. Tests
membership for factorial designs much more efficiently than
for plain arrays or data frames.
* `stk_generate_samplepaths` now returns a plain numerical array instead of
a data frame.
* Miscellaneous:
- `stk_plot_probdom2d`: New function to represent the
uncertainty about a 2D Pareto front. This function is
currently considered experimental and should not be expected
to produce a reliable representation of the uncertainty for
difficult or high-dimensional optimization problems.
- New test case: "truss3" (Koski, 1985; Das, 1997).
- Many functions have been optimized for speed.
- `stk_plot1d`: Handle the case where `xt` is not sorted.
- Support old-style STK structures (with a `.a` field) has
been removed.
-------------------------------------------------------------------
Wed Nov 11 15:20:54 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.3.4
* Bug fixes and minor changes
-------------------------------------------------------------------
Tue Oct 13 06:08:14 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.3.3
* Bug fixes and minor changes
-------------------------------------------------------------------
Tue Jul 7 17:09:17 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.3.2
* Bug fixes and minor changes
-------------------------------------------------------------------
Mon Jun 22 19:41:57 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.3.1
* Bugfix release
-------------------------------------------------------------------
Mon May 18 19:44:24 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.3.0
-------------------------------------------------------------------
Thu Apr 16 21:50:44 UTC 2015 - dmitry_r@opensuse.org
- Update to version 2.2.1
* Octave 4.0 compatibility
-------------------------------------------------------------------
Thu Feb 26 16:06:43 UTC 2015 - dmitry_r@opensuse.org
- Split from octave-forge package, version 2.2.0