Note: For changes to the Python wrapper
shaprpy, see the shaprpy
CHANGELOG.
shapr 1.0.8
Improvements
Enhanced plot.shapr() with smarter defaults for
index_x_explain: now defaults to 1:10 for bar
and waterfall plots, and all observations for scatter and beeswarm
plots. Informative messages are shown if more than 10 observations are
available for bar/waterfall plots (#487).
Renamed ARIMA model methods from forecast_ARIMA to
fc_model in S3 method registrations and implementations
after class name change in the forecast package (to align
with fable package conventions). This affects both
get_model_specs and predict_model methods.
CRAN tests added to catch such up front in the future (#487).
Updated plot_MSEv_eval_crit() and related documentation
to use “coalition” instead of “comb” for plot types, improving clarity
and consistency (#487).
Documentation
Various minor documentation clarifications and improvements (#487)
shapr 1.0.7
Improvements
Updated summary.shapr() to return a
summary.shapr class object (rather than just printing) for
improved reusability and consistency with R conventions (#482).
Enhanced plot.shapr() with default
index_x_explain = 1:10 for more convenient plotting of
first observations (and printing info about this) (#482).
Bug fixes
Fixed compatibility with xgboost after their breaking
change class update (version >= 3.1.2.1). We now support both the
previous and the new xgboost versions (#481).
Added non-snapshot tests for all natively supported model classes
(#481).
Minor updates to examples and vignettes the updates in
summary.shapr() and the xgboost class (#483).
shapr 1.0.6
Improvements
Added “proglang” to the what argument in
get_results() to extract the programming language used (R
or Python) (#477), to
ease reuse of the R methods in the shaprpy
Python wrapper.
Added print_ggplot argument to
plot.shapr() (default TRUE) to ensure ggplots
are displayed when sourced, in functions, loops etc. Applied the same
fix to plot_SV_several_approaches(). (#478)
Minor fixes
Converted all exported function titles to Title Case in the
documentation (#477)
shapr 1.0.5
New features
Added get_results() for extracting key components from
a shapr object, including Shapley values, model parameters,
and iterative computation details (#460).
Added summary.shapr() method, which builds on
get_results() and internal format_xyz()
functions to provide a concise overview of explanation objects (#460).
Improvements
Enhanced print.shapr() to support printing of specific
components, with customizable control over digits and decimal precision
(#460).
Refactored cli-based output functions for improved
reuse across multiple methods (#460).
Updated vignettes, examples, and tests to reflect the new
functionality (#460).
Fixed spelling, grammar, and other textual inconsistencies in
documentation, comments, and vignettes (#465).
Applied various minor visual improvements (#460, #464).
Converted vignette figures to WebP format to significantly reduce
tarball size (#468)
Bug fixes
Fixed error when handling many features by removing
as.integer() coercion in the computation of coalitions per
coalition size (#462).
Implement semi-deterministic sampling of coalitions similar to
the default in the shap python library and described and
discussed as the PySHAP* strategy in Olsen & Jullum (2024).
It is disabled by default, but can be set via
extra_computation_args = list(semi_deterministic_sampling = TRUE)
in explain(). The functionality is available when paired
coalition sampling (the default) is enabled. See #449 for
details.
Deletes the regression-surrogate parsnip object when testing to
avoid future conflicts with model object changes. (Second last commit in
#447).
Improve and update the logic and print for setting the number of
coalitions in the next iteration for iterative = TRUE (#452)
Allow passing vS_batching_method to
explain()/explain_forecast() to specify the batch
computation method (default is "future" for both,
"forloop" available mainly for dev purposes) (#452)
Transform to use the cli and rlang
packages to provide all messages/warnings/stops with nicer formatting
and layout. The messages (via cli_inform()) now also obey
the verbose argument and are displayed only if
'basic' %in% verbose is TRUE. The header printout also
differs between explain()/explain_forecast()
and whether called from Python. This also adds cli and
rlang to imports. (#453)
Now using testthat::skip_if_not_installed for all
tests requiring suggested packages to ensure they are skipped gracefully
when dependencies are unavailable (#451)
Other minor fixes
Two minor bugs related to KernelSHAP_reweighing() (#448)
Two minor bugs related to weighting for asymmetric Shapley values
(#449)
Check seed argument, and only pass to
torch if not NULL (#452)
Make explain_forecast() use future for
batch computation as well (by default) (#452)
Fix bug for the AICc + independence variants of
approach = 'empirical' occurring when
n_features < n_explain (#453)
shapr 1.0.3
Fix documentation issues detected during shapr 1.0.2 release (#442)
Remove examples from unexported functions
Replace by for long running examples
Replace usage of print() by warning() on
two occasions
Fix issue with
Expected <nn_module> but got object of type <NULL>
for approach='vaeac' after recent torch update
broke it (#444)
Changes default seed in explain() and
explain_forecast() from 1 to NULL to avoid set.seed() to
conflict with later called code (#445)
Other minor fixes
Add digits as arguments to the internal snapshot testing
expect_snapshot_rds() to reduce false positive
roundoff-errors between platforms (#444)
Adjust dependencies in Suggests and their usage in tests (#446)
Used skip_on_cran() to reduce check time on CRAN for snapshot tests
which where not tested anyway
shapr 1.0.2
Fix CRAN NOTE which turned out to be bug related to returning NULL
rather than integer(0) to identify (unconditional) asymmetric causal
sampling (#435)
[Py] Get ‘shaprpy’ up to speed with the latest ‘shapr’ version and
return the internal list properly (#436)
Bugfix keep_samp_for_vS with iterative approach (#417)
Python Brought the python code base up to
speed with essentially all functionality in explain() in R
(#416)
Please CRAN with dontrun on long running examples + skip_on_cran for
the parallelized tests.
shapr 1.0.0 (GitHub only)
Breaking changes
Moved from explaining predictions using two functions
(shapr() for initial setup + explain() for
explanation for specific observations), to a single function
call (also named explain()). The data used for training and
to be explained have gotten explicit names (x_train and
x_explain). The order of the input arguments has also been
slightly changed (model is now the first argument).
Prediction and checking functions for custom models are now passed
directly as arguments to explain() instead of being defined
as functions of a specific class in the global env.
The previously exported function make_dummies used to
explain xgboost models with categorical data, is removed to
simplify the code base. This is rather handled with a custom prediction
model.
The function explain.ctree_comb_mincrit, which allowed
combining models with approch=ctree with different
mincrit parameters, has been removed to simplify the code
base. It may return in a completely general manner in later version of
shapr.
Iterative Shapley value estimation with convergence detection
New approaches: vaeac, regression_separate, regression_surrogate,
timeseries, categorical
verbose argument for explain() to control the amount of output
Parallelized computation of v(S) with future, including progress
updates
Paired_sampling of coalitions
prev_shapr_object argument to explain() to continue explanation from
a previous object
asymmetric and causal Shapley values
Improved KernelSHAP estimation with adjusted weights for reduced
variance
Release a Python wrapper (shaprpyr, #325) for
explaining predictions from Python models (from Python) utilizing almost
all functionality of shapr. The wrapper moves back and
forth back and forth between Python and R, doing the prediction in
Python, and almost everything else in R. This simplifies maintenance of
shaprpy significantly. The wrapper is available here.
Introduce batch computation of conditional expectations (#244).
This essentially compute \(v(S)\) for a
portion of the \(S\)-subsets at a time,
to reduce the amount of data needed to be held in memory. The user can
control the number of batches herself, but we set a reasonable value by
default (#327).
This allows models with large number of features to be explained with a
significantly lower RAM consumption (at the cost of a slight increase in
the computation time)
Parallelization over batches (#38)
using the future
framework.
Progress bar (#293)
using the progressr
package. Must be activated by the user with
progressr::handlers(global = TRUE) or wrapping the call to
explain() around
progressr::with_progress({})
Added approach = 'categorical' (#256,
#307)
used to explain models with solely categorical features by directly
using/estimating the joint distribution of all feature
combinations.
Added approch='timeseries' (#314) for
explaining classifications based on time series data/models with the
method described in Sec 4.3 of the groupShapley
paper.
Implemented unique sampling of Shapley value subsets (#227)
Added new function explain_forecast to explain
forecasts from time series models, at various prediction horizons (#328).
Uses a different set of input argument which is more appropriate for
these models.
Re-implementation of approach = 'independence' method
providing significantly faster computation (no longer as a special case
of the empirical method). Also allow the method to be used
on models with categorical data (#315).
Added ‘beeswarm’ and ‘waterfall’ plots + new coloring scheme for all
plots. See the vignette
for examples.
Added timing of the various parts of the explanation process.
Under the hood
The test base have been completely rewritten (#249).
Now heavily utilizing snapshots
on a large set of benchmark calls to explain, also using vdiffr for plot tests. Test
functions are only written for exported core functions. Internal
functions are only tested through the exported ones.
Avoid unnecessary computation of inverse for weight matrix (#280)
Minor improvements and bug
fixes
The vignette/readme/tests now uses the
datasets::airquality dataset. This avoids including a new
package just for the dataset (#248).
Allows lm/glm/gam models with interactions (#303).
Previously, this was not possible with the prediction functions defined
internally due to a bug.
Sampling of group subsets implemented also for grouping, not only
features.
Documentation improvements
The vignette
has been updated to reflect the new framework for explaining
predictions, and all the new package features/functionality.
New vignettes also for the regression paradigm, vaeac and the
asymmetric/causal Shapley values
shapr 0.2.3 (GitHub only)
Development version
Added support for groupSHAP, including check of appropriate groups,
examples and tests
Various modifications to input of internal functions to reflect that
Shapley values may be computed both feature-wise and group-wise
Fixed bug when passing non-named data to shapr() or explain()
(e.g. shapr(data[,1:5],model...)
shapr 0.2.2
Patch to fix failing CRAN-tests on R-devel due to changed behavior
of attach(): Fixed by changing how we simulate adding a
function to .GlobalEnv in the failing test. Actual package not
affected.
shapr 0.2.1
Patch to fix warning from development version of data.table due to
the use of nomatch argument in merge(), as requested from data.table
developers.
shapr 0.2.0
Minor CRAN release
Added the new dependence modeling approach “ctree” which handles
categorical features in addition to numerical ones. For more information
see our paper https://doi.org/10.1007/978-3-030-57321-8_7
Added support to explain models which take as input categorical
features for model classes like xgboost which originally takes only
numeric input. On the user side, an additional call to the new
make_dummies function is required. See the vignette for
details.
Slight change in the user procedure for explaining predictions from
custom models. This now requires only a single function
predict_model.
Introduced a thorough system for extracting and checking the feature
information in the model and the data passed to shapr and
explain. The features in the data are checked for consistency
with what can be extracted from the model object. If the model object is
missing some of the necessary information, the info from the data is
used instead. The system checks feature labels, classes, and any factor
levels.
Due to the previous point, the feature_names option
previously used for custom models is removed.
Added a manual testing script for custom model (currently cannot be
handled by testthat due to environment issues).
A few under-the-hood changes for checking in the shapr
function.
shapr 0.1.4
Patch to fulfill CRAN policy of using packages under Suggests
conditionally (in tests and examples)
shapr 0.1.3
Fix installation error on Solaris
Updated README with CRAN installation instructions and badges
shapr 0.1.2
CRAN release
Removed unused clustering code
Removed several package dependencies
Moved automatic check and pkgdown site build from Circle CI to
GitHub actions
Some minor efficiency fixes
Changed stopping threshold from 12 to 13 features for none-sampling
version of KernelSHAP for consistency with our recommendation
Changed package title (shortened)
Minor fixes to fulfill CRAN policy
Improved documentation
Revised internal/external and exported/non-exported functions,
leading to far fewer external functions and a cleaner manual.
shapr 0.1.1
Journal of Open Source Software release
Improved installation instructions and community guidelines in
README