From 0c89dc36152a2ec3962993482aecf7cc0f3517f8 Mon Sep 17 00:00:00 2001 From: Michael Mayer Date: Wed, 23 Jul 2025 12:10:54 +0200 Subject: [PATCH 1/5] try progressr --- DESCRIPTION | 1 + NEWS.md | 10 ++++-- R/kernelshap.R | 75 +++++++++++++++++--------------------------- R/permshap.R | 72 +++++++++++++++++------------------------- R/utils_kernelshap.R | 10 +++++- R/utils_permshap.R | 8 ++++- R/zzz.R | 1 + README.md | 7 +++-- packaging.R | 2 ++ 9 files changed, 88 insertions(+), 98 deletions(-) create mode 100644 R/zzz.R diff --git a/DESCRIPTION b/DESCRIPTION index a95507b..6a41597 100644 --- a/DESCRIPTION +++ b/DESCRIPTION @@ -31,6 +31,7 @@ Imports: stats, utils Suggests: + progressr, testthat (>= 3.0.0) Config/testthat/edition: 3 URL: https://github.com/ModelOriented/kernelshap diff --git a/NEWS.md b/NEWS.md index d6c3903..67191cd 100644 --- a/NEWS.md +++ b/NEWS.md @@ -12,8 +12,10 @@ we have fixed a bug in how `kernelshap()` calculates Kernel weights. Fixed in [#168](https://github.com/ModelOriented/kernelshap/pull/168), which also has received unit tests against Python's "shap". -### API +### API changes and improvements +- The traditional progress bar has been replaced by {progressr}. Activate e.g. via `progressr::handlers(global = TRUE). + It works in parallel mode as well, and you can choose your own style ([#173](https://github.com/ModelOriented/kernelshap/pull/173)). - The argument `feature_names` can now also be used with matrix input ([#166](https://github.com/ModelOriented/kernelshap/pull/166)). - `kernelshap()` and `permshap()` have received a `seed = NULL` argument ([#170](https://github.com/ModelOriented/kernelshap/pull/170)). - Parallel mode: If missing packages or globals have to be specified, this now has to be done through `parallel_args = list(packages = ..., globals = ...)` @@ -39,14 +41,16 @@ The list is passed to `[foreach::foreach(.options.future = ...)]`. We have switched from `%dopar%` to `doFuture` ([#170](https://github.com/ModelOriented/kernelshap/pull/170)) with the following impact: - No need for calling `registerDoFuture()` anymore. +- No need to set `parallel = TRUE` anymore ([#173](https://github.com/ModelOriented/kernelshap/pull/173)). - Random seeding is properly handled, and respects `seed`, thanks [#163](https://github.com/ModelOriented/kernelshap/issues/163) for reporting. - If missing packages or globals have to be specified, this now has to be done through `parallel_args = list(packages = ..., globals = ...)` instead of `parallel_args = list(.packages = ..., .globals = ...)`. The list is passed to `[foreach::foreach(.options.future = ...)]`. ### Dependencies -- {MASS}: Dropped from imports -- {doFuture}: suggests -> imports +- {MASS}: Dropped from Imports +- {doFuture}: Suggests -> Imports +- {progressr}: Suggests # kernelshap 0.8.0 diff --git a/R/kernelshap.R b/R/kernelshap.R index 4ddc690..d46077e 100644 --- a/R/kernelshap.R +++ b/R/kernelshap.R @@ -8,6 +8,8 @@ #' Otherwise, a partly exact hybrid algorithm combining exact calculations and #' iterative paired sampling is used, see Details. #' +#' To activate the progress bar, run `progressr::handlers(global = TRUE)` first. +#' #' @details #' The pure iterative Kernel SHAP sampling as in Covert and Lee (2021) works like this: #' @@ -198,6 +200,9 @@ kernelshap.default <- function( verbose = TRUE, seed = NULL, ...) { + if (parallel) { + warning("The 'parallel' argument has been deprecated. Simply set plan(...) to activate parallel computing.") + } p <- length(feature_names) basic_checks(X = X, feature_names = feature_names, pred_fun = pred_fun) stopifnot( @@ -258,54 +263,32 @@ kernelshap.default <- function( warning_burden(max(m, m_exact), bg_n = bg_n) } - # Apply Kernel SHAP to each row of X - if (isTRUE(parallel)) { - future_args <- c(list(seed = TRUE), parallel_args) - parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) - res <- do.call(foreach::foreach, parallel_args) %dofuture% kernelshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], - object = object, - pred_fun = pred_fun, - feature_names = feature_names, - bg_w = bg_w, - exact = exact, - deg = hybrid_degree, - m = m, - tol = tol, - max_iter = max_iter, - v0 = v0, - precalc = precalc, - ... - ) - } else { - if (verbose && n >= 2L) { - pb <- utils::txtProgressBar(max = n, style = 3) - } - res <- vector("list", n) - for (i in seq_len(n)) { - res[[i]] <- kernelshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], - object = object, - pred_fun = pred_fun, - feature_names = feature_names, - bg_w = bg_w, - exact = exact, - deg = hybrid_degree, - m = m, - tol = tol, - max_iter = max_iter, - v0 = v0, - precalc = precalc, - ... - ) - if (verbose && n >= 2L) { - utils::setTxtProgressBar(pb, i) - } - } + pbar <- if (verbose && n >= 2L && requireNamespace("progressr", quietly = TRUE)) { + progressr::progressor(n) } + # Apply Kernel SHAP to each row of X + future_args <- c(list(seed = TRUE), parallel_args) + parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) + + res <- do.call(foreach::foreach, parallel_args) %dofuture% kernelshap_one( + x = X[i, , drop = FALSE], + v1 = v1[i, , drop = FALSE], + object = object, + pred_fun = pred_fun, + feature_names = feature_names, + bg_w = bg_w, + exact = exact, + deg = hybrid_degree, + m = m, + tol = tol, + max_iter = max_iter, + v0 = v0, + precalc = precalc, + pbar = pbar, + ... + ) + # Organize output exact <- exact || trunc(p / 2) == hybrid_degree diff --git a/R/permshap.R b/R/permshap.R index 398bcf8..3a18e65 100644 --- a/R/permshap.R +++ b/R/permshap.R @@ -9,6 +9,8 @@ #' Otherwise, the sampling process iterates until the resulting values #' are sufficiently precise, and standard errors are provided. #' +#' To activate the progress bar, run `progressr::handlers(global = TRUE)` first. +#' #' @details #' During each iteration, the algorithm cycles twice through a random permutation: #' It starts with all feature components "turned on" (i.e., taking them @@ -108,6 +110,9 @@ permshap.default <- function( verbose = TRUE, seed = NULL, ...) { + if (parallel) { + warning("The 'parallel' argument has been deprecated. Simply set plan(...) to activate parallel computing.") + } p <- length(feature_names) if (p <= 1L) { stop("Case p = 1 not implemented. Use kernelshap() instead.") @@ -167,52 +172,31 @@ permshap.default <- function( warning_burden(max(m_eval, m_exact), bg_n = bg_n) } - # Apply permutation SHAP to each row of X - if (isTRUE(parallel)) { - future_args <- c(list(seed = TRUE), parallel_args) - parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) - res <- do.call(foreach::foreach, parallel_args) %dofuture% permshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], - object = object, - pred_fun = pred_fun, - bg_w = bg_w, - v0 = v0, - precalc = precalc, - feature_names = feature_names, - exact = exact, - low_memory = low_memory, - tol = tol, - max_iter = max_iter, - ... - ) - } else { - if (verbose && n >= 2L) { - pb <- utils::txtProgressBar(max = n, style = 3) - } - res <- vector("list", n) - for (i in seq_len(n)) { - res[[i]] <- permshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], - object = object, - pred_fun = pred_fun, - bg_w = bg_w, - v0 = v0, - precalc = precalc, - feature_names = feature_names, - exact = exact, - low_memory = low_memory, - tol = tol, - max_iter = max_iter, - ... - ) - if (verbose && n >= 2L) { - utils::setTxtProgressBar(pb, i) - } - } + pbar <- if (verbose && n >= 2L && requireNamespace("progressr", quietly = TRUE)) { + progressr::progressor(n) } + # Apply permutation SHAP to each row of X + future_args <- c(list(seed = TRUE), parallel_args) + parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) + + res <- do.call(foreach::foreach, parallel_args) %dofuture% permshap_one( + x = X[i, , drop = FALSE], + v1 = v1[i, , drop = FALSE], + object = object, + pred_fun = pred_fun, + bg_w = bg_w, + v0 = v0, + precalc = precalc, + feature_names = feature_names, + exact = exact, + low_memory = low_memory, + tol = tol, + max_iter = max_iter, + pbar = pbar, + ... + ) + # Organize output out <- list( S = reorganize_list(lapply(res, `[[`, "beta")), diff --git a/R/utils_kernelshap.R b/R/utils_kernelshap.R index ed33654..2013908 100644 --- a/R/utils_kernelshap.R +++ b/R/utils_kernelshap.R @@ -15,6 +15,7 @@ kernelshap_one <- function( max_iter, v0, precalc, + pbar, ...) { p <- length(feature_names) K <- ncol(v1) @@ -38,6 +39,9 @@ kernelshap_one <- function( # Some of the hybrid cases are exact as well if (exact || trunc(p / 2) == deg) { beta <- solver(A_exact, b_exact, constraint = v1 - v0) # (p x K) + if (!is.null(pbar)) { + pbar() + } return(list(beta = beta)) } } @@ -90,7 +94,11 @@ kernelshap_one <- function( sigma_n <- NA * beta_n # (p x K) } } - list(beta = beta_n, sigma = sigma_n, n_iter = n_iter, converged = converged) + out <- list(beta = beta_n, sigma = sigma_n, n_iter = n_iter, converged = converged) + if (!is.null(pbar)) { + pbar() + } + return(out) } # Regression coefficients given sum(beta) = constraint diff --git a/R/utils_permshap.R b/R/utils_permshap.R index 1f7ed88..e36b93b 100644 --- a/R/utils_permshap.R +++ b/R/utils_permshap.R @@ -26,9 +26,9 @@ permshap_one <- function( low_memory, tol, max_iter, + pbar, ...) { bg <- precalc$bg_X_rep - p <- length(feature_names) K <- ncol(v1) K_names <- colnames(v1) @@ -54,6 +54,9 @@ permshap_one <- function( w = precalc$shapley_w[pos$on] ) } + if (!is.null(pbar)) { + pbar() + } return(list(beta = beta_n)) } @@ -128,6 +131,9 @@ permshap_one <- function( out <- list( beta = beta_n / n_iter, sigma = sigma_n, n_iter = n_iter, converged = converged ) + if (!is.null(pbar)) { + pbar() + } return(out) } diff --git a/R/zzz.R b/R/zzz.R new file mode 100644 index 0000000..764ce86 --- /dev/null +++ b/R/zzz.R @@ -0,0 +1 @@ +globalVariables("i") diff --git a/README.md b/README.md index b0b9858..339194d 100644 --- a/README.md +++ b/README.md @@ -51,6 +51,8 @@ library(ggplot2) library(ranger) library(shapviz) +progressr::handlers(global = TRUE) # activates progress bar + options(ranger.num.threads = 8) diamonds <- transform( @@ -129,9 +131,7 @@ plan(multisession, workers = 4) # Windows fit <- gam(log_price ~ s(log_carat) + clarity * color + cut, data = diamonds) system.time( # 4 seconds in parallel - ps <- permshap( - fit, X, bg_X = bg_X, parallel = TRUE, parallel_args = list(packages = "mgcv") - ) + ps <- permshap(fit, X, bg_X = bg_X, parallel_args = list(packages = "mgcv")) ) ps @@ -141,6 +141,7 @@ ps # [2,] -0.51546 -0.1174766 0.11122775 0.030243973 # Because there are no interactions of order above 2, Kernel SHAP gives the same: +plan("sequential") system.time( # 12 s non-parallel ks <- kernelshap(fit, X, bg_X = bg_X) ) diff --git a/packaging.R b/packaging.R index cc9ab08..136f234 100644 --- a/packaging.R +++ b/packaging.R @@ -42,6 +42,8 @@ use_package("foreach", "Imports") use_package("stats", "Imports") use_package("utils", "Imports") +use_package("progressr", "Suggests") + use_gpl_license(2) # Your files that do not belong to the package itself (others are added by "use_* function") From c81f3fbc6e5acc4cd158a9bfac673efe857e69bd Mon Sep 17 00:00:00 2001 From: Michael Mayer Date: Wed, 23 Jul 2025 14:09:21 +0200 Subject: [PATCH 2/5] switch to future_lapply() --- DESCRIPTION | 9 +- LICENSE.md | 881 +++++++++++++++++++++------------ NAMESPACE | 1 - NEWS.md | 41 +- R/kernelshap.R | 41 +- R/permshap.R | 25 +- R/utils_kernelshap.R | 3 + R/utils_permshap.R | 4 + README.md | 18 +- backlog/compare_with_python2.R | 19 +- cran-comments.md | 9 +- man/kernelshap.Rd | 21 +- man/permshap.Rd | 21 +- packaging.R | 7 +- tests/testthat/test-basic.R | 18 +- 15 files changed, 695 insertions(+), 423 deletions(-) diff --git a/DESCRIPTION b/DESCRIPTION index 6a41597..765c7d6 100644 --- a/DESCRIPTION +++ b/DESCRIPTION @@ -1,6 +1,6 @@ Package: kernelshap Title: Kernel SHAP -Version: 0.9.0 +Version: 1.0.0 Authors@R: c( person("Michael", "Mayer", , "mayermichael79@gmail.com", role = c("aut", "cre"), comment = c(ORCID = "0009-0007-2540-9629")), @@ -19,15 +19,14 @@ Description: Efficient implementation of Kernel SHAP (Lundberg and Lee, It supports multi-output models, case weights, and parallel computations. Visualizations can be done using the R package 'shapviz'. -License: GPL (>= 2) +License: GPL (>= 3) Depends: R (>= 3.2.0) Encoding: UTF-8 Roxygen: list(markdown = TRUE) RoxygenNote: 7.3.2 -Imports: - doFuture, - foreach, +Imports: + future.apply, stats, utils Suggests: diff --git a/LICENSE.md b/LICENSE.md index 0daa041..175443c 100644 --- a/LICENSE.md +++ b/LICENSE.md @@ -1,336 +1,595 @@ GNU General Public License ========================== -_Version 2, June 1991_ -_Copyright © 1989, 1991 Free Software Foundation, Inc.,_ -_51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA_ - -Everyone is permitted to copy and distribute verbatim copies -of this license document, but changing it is not allowed. - -### Preamble - -The licenses for most software are designed to take away your -freedom to share and change it. By contrast, the GNU General Public -License is intended to guarantee your freedom to share and change free -software--to make sure the software is free for all its users. This -General Public License applies to most of the Free Software -Foundation's software and to any other program whose authors commit to -using it. (Some other Free Software Foundation software is covered by -the GNU Lesser General Public License instead.) You can apply it to +_Version 3, 29 June 2007_ +_Copyright © 2007 Free Software Foundation, Inc. <>_ + +Everyone is permitted to copy and distribute verbatim copies of this license +document, but changing it is not allowed. + +## Preamble + +The GNU General Public License is a free, copyleft license for software and other +kinds of works. + +The licenses for most software and other practical works are designed to take away +your freedom to share and change the works. By contrast, the GNU General Public +License is intended to guarantee your freedom to share and change all versions of a +program--to make sure it remains free software for all its users. We, the Free +Software Foundation, use the GNU General Public License for most of our software; it +applies also to any other work released this way by its authors. You can apply it to your programs, too. -When we speak of free software, we are referring to freedom, not -price. Our General Public Licenses are designed to make sure that you -have the freedom to distribute copies of free software (and charge for -this service if you wish), that you receive source code or can get it -if you want it, that you can change the software or use pieces of it -in new free programs; and that you know you can do these things. - -To protect your rights, we need to make restrictions that forbid -anyone to deny you these rights or to ask you to surrender the rights. -These restrictions translate to certain responsibilities for you if you -distribute copies of the software, or if you modify it. - -For example, if you distribute copies of such a program, whether -gratis or for a fee, you must give the recipients all the rights that -you have. You must make sure that they, too, receive or can get the -source code. And you must show them these terms so they know their -rights. - -We protect your rights with two steps: **(1)** copyright the software, and -**(2)** offer you this license which gives you legal permission to copy, -distribute and/or modify the software. - -Also, for each author's protection and ours, we want to make certain -that everyone understands that there is no warranty for this free -software. If the software is modified by someone else and passed on, we -want its recipients to know that what they have is not the original, so -that any problems introduced by others will not reflect on the original -authors' reputations. - -Finally, any free program is threatened constantly by software -patents. We wish to avoid the danger that redistributors of a free -program will individually obtain patent licenses, in effect making the -program proprietary. To prevent this, we have made it clear that any -patent must be licensed for everyone's free use or not licensed at all. - -The precise terms and conditions for copying, distribution and -modification follow. - -### TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION - -**0.** This License applies to any program or other work which contains -a notice placed by the copyright holder saying it may be distributed -under the terms of this General Public License. The “Program”, below, -refers to any such program or work, and a “work based on the Program” -means either the Program or any derivative work under copyright law: -that is to say, a work containing the Program or a portion of it, -either verbatim or with modifications and/or translated into another -language. (Hereinafter, translation is included without limitation in -the term “modification”.) Each licensee is addressed as “you”. - -Activities other than copying, distribution and modification are not -covered by this License; they are outside its scope. The act of -running the Program is not restricted, and the output from the Program -is covered only if its contents constitute a work based on the -Program (independent of having been made by running the Program). -Whether that is true depends on what the Program does. - -**1.** You may copy and distribute verbatim copies of the Program's -source code as you receive it, in any medium, provided that you -conspicuously and appropriately publish on each copy an appropriate -copyright notice and disclaimer of warranty; keep intact all the -notices that refer to this License and to the absence of any warranty; -and give any other recipients of the Program a copy of this License -along with the Program. - -You may charge a fee for the physical act of transferring a copy, and -you may at your option offer warranty protection in exchange for a fee. - -**2.** You may modify your copy or copies of the Program or any portion -of it, thus forming a work based on the Program, and copy and -distribute such modifications or work under the terms of Section 1 -above, provided that you also meet all of these conditions: - -* **a)** You must cause the modified files to carry prominent notices -stating that you changed the files and the date of any change. -* **b)** You must cause any work that you distribute or publish, that in -whole or in part contains or is derived from the Program or any -part thereof, to be licensed as a whole at no charge to all third -parties under the terms of this License. -* **c)** If the modified program normally reads commands interactively -when run, you must cause it, when started running for such -interactive use in the most ordinary way, to print or display an -announcement including an appropriate copyright notice and a -notice that there is no warranty (or else, saying that you provide -a warranty) and that users may redistribute the program under -these conditions, and telling the user how to view a copy of this -License. (Exception: if the Program itself is interactive but -does not normally print such an announcement, your work based on -the Program is not required to print an announcement.) - -These requirements apply to the modified work as a whole. If -identifiable sections of that work are not derived from the Program, -and can be reasonably considered independent and separate works in -themselves, then this License, and its terms, do not apply to those -sections when you distribute them as separate works. But when you -distribute the same sections as part of a whole which is a work based -on the Program, the distribution of the whole must be on the terms of -this License, whose permissions for other licensees extend to the -entire whole, and thus to each and every part regardless of who wrote it. - -Thus, it is not the intent of this section to claim rights or contest -your rights to work written entirely by you; rather, the intent is to -exercise the right to control the distribution of derivative or -collective works based on the Program. - -In addition, mere aggregation of another work not based on the Program -with the Program (or with a work based on the Program) on a volume of -a storage or distribution medium does not bring the other work under -the scope of this License. - -**3.** You may copy and distribute the Program (or a work based on it, -under Section 2) in object code or executable form under the terms of -Sections 1 and 2 above provided that you also do one of the following: - -* **a)** Accompany it with the complete corresponding machine-readable -source code, which must be distributed under the terms of Sections -1 and 2 above on a medium customarily used for software interchange; or, -* **b)** Accompany it with a written offer, valid for at least three -years, to give any third party, for a charge no more than your -cost of physically performing source distribution, a complete -machine-readable copy of the corresponding source code, to be -distributed under the terms of Sections 1 and 2 above on a medium -customarily used for software interchange; or, -* **c)** Accompany it with the information you received as to the offer -to distribute corresponding source code. (This alternative is -allowed only for noncommercial distribution and only if you -received the program in object code or executable form with such -an offer, in accord with Subsection b above.) - -The source code for a work means the preferred form of the work for -making modifications to it. For an executable work, complete source -code means all the source code for all modules it contains, plus any -associated interface definition files, plus the scripts used to -control compilation and installation of the executable. However, as a -special exception, the source code distributed need not include -anything that is normally distributed (in either source or binary -form) with the major components (compiler, kernel, and so on) of the -operating system on which the executable runs, unless that component -itself accompanies the executable. - -If distribution of executable or object code is made by offering -access to copy from a designated place, then offering equivalent -access to copy the source code from the same place counts as -distribution of the source code, even though third parties are not -compelled to copy the source along with the object code. - -**4.** You may not copy, modify, sublicense, or distribute the Program -except as expressly provided under this License. Any attempt -otherwise to copy, modify, sublicense or distribute the Program is -void, and will automatically terminate your rights under this License. -However, parties who have received copies, or rights, from you under -this License will not have their licenses terminated so long as such -parties remain in full compliance. - -**5.** You are not required to accept this License, since you have not -signed it. However, nothing else grants you permission to modify or -distribute the Program or its derivative works. These actions are -prohibited by law if you do not accept this License. Therefore, by -modifying or distributing the Program (or any work based on the -Program), you indicate your acceptance of this License to do so, and -all its terms and conditions for copying, distributing or modifying -the Program or works based on it. - -**6.** Each time you redistribute the Program (or any work based on the -Program), the recipient automatically receives a license from the -original licensor to copy, distribute or modify the Program subject to -these terms and conditions. You may not impose any further -restrictions on the recipients' exercise of the rights granted herein. -You are not responsible for enforcing compliance by third parties to -this License. - -**7.** If, as a consequence of a court judgment or allegation of patent -infringement or for any other reason (not limited to patent issues), -conditions are imposed on you (whether by court order, agreement or -otherwise) that contradict the conditions of this License, they do not -excuse you from the conditions of this License. If you cannot -distribute so as to satisfy simultaneously your obligations under this -License and any other pertinent obligations, then as a consequence you -may not distribute the Program at all. For example, if a patent -license would not permit royalty-free redistribution of the Program by -all those who receive copies directly or indirectly through you, then -the only way you could satisfy both it and this License would be to -refrain entirely from distribution of the Program. - -If any portion of this section is held invalid or unenforceable under -any particular circumstance, the balance of the section is intended to -apply and the section as a whole is intended to apply in other -circumstances. - -It is not the purpose of this section to induce you to infringe any -patents or other property right claims or to contest validity of any -such claims; this section has the sole purpose of protecting the -integrity of the free software distribution system, which is -implemented by public license practices. Many people have made -generous contributions to the wide range of software distributed -through that system in reliance on consistent application of that -system; it is up to the author/donor to decide if he or she is willing -to distribute software through any other system and a licensee cannot -impose that choice. - -This section is intended to make thoroughly clear what is believed to -be a consequence of the rest of this License. - -**8.** If the distribution and/or use of the Program is restricted in -certain countries either by patents or by copyrighted interfaces, the -original copyright holder who places the Program under this License -may add an explicit geographical distribution limitation excluding -those countries, so that distribution is permitted only in or among -countries not thus excluded. In such case, this License incorporates -the limitation as if written in the body of this License. - -**9.** The Free Software Foundation may publish revised and/or new versions -of the General Public License from time to time. Such new versions will -be similar in spirit to the present version, but may differ in detail to -address new problems or concerns. - -Each version is given a distinguishing version number. If the Program -specifies a version number of this License which applies to it and “any -later version”, you have the option of following the terms and conditions -either of that version or of any later version published by the Free -Software Foundation. If the Program does not specify a version number of -this License, you may choose any version ever published by the Free Software -Foundation. - -**10.** If you wish to incorporate parts of the Program into other free -programs whose distribution conditions are different, write to the author -to ask for permission. For software which is copyrighted by the Free -Software Foundation, write to the Free Software Foundation; we sometimes -make exceptions for this. Our decision will be guided by the two goals -of preserving the free status of all derivatives of our free software and -of promoting the sharing and reuse of software generally. - -### NO WARRANTY - -**11.** BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY -FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN -OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES -PROVIDE THE PROGRAM “AS IS” WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED -OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF -MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS -TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE -PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, -REPAIR OR CORRECTION. - -**12.** IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING -WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR -REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, -INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING -OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED -TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY -YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER -PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE +When we speak of free software, we are referring to freedom, not price. Our General +Public Licenses are designed to make sure that you have the freedom to distribute +copies of free software (and charge for them if you wish), that you receive source +code or can get it if you want it, that you can change the software or use pieces of +it in new free programs, and that you know you can do these things. + +To protect your rights, we need to prevent others from denying you these rights or +asking you to surrender the rights. Therefore, you have certain responsibilities if +you distribute copies of the software, or if you modify it: responsibilities to +respect the freedom of others. + +For example, if you distribute copies of such a program, whether gratis or for a fee, +you must pass on to the recipients the same freedoms that you received. You must make +sure that they, too, receive or can get the source code. And you must show them these +terms so they know their rights. + +Developers that use the GNU GPL protect your rights with two steps: **(1)** assert +copyright on the software, and **(2)** offer you this License giving you legal permission +to copy, distribute and/or modify it. + +For the developers' and authors' protection, the GPL clearly explains that there is +no warranty for this free software. For both users' and authors' sake, the GPL +requires that modified versions be marked as changed, so that their problems will not +be attributed erroneously to authors of previous versions. + +Some devices are designed to deny users access to install or run modified versions of +the software inside them, although the manufacturer can do so. This is fundamentally +incompatible with the aim of protecting users' freedom to change the software. The +systematic pattern of such abuse occurs in the area of products for individuals to +use, which is precisely where it is most unacceptable. Therefore, we have designed +this version of the GPL to prohibit the practice for those products. If such problems +arise substantially in other domains, we stand ready to extend this provision to +those domains in future versions of the GPL, as needed to protect the freedom of +users. + +Finally, every program is threatened constantly by software patents. States should +not allow patents to restrict development and use of software on general-purpose +computers, but in those that do, we wish to avoid the special danger that patents +applied to a free program could make it effectively proprietary. To prevent this, the +GPL assures that patents cannot be used to render the program non-free. + +The precise terms and conditions for copying, distribution and modification follow. + +## TERMS AND CONDITIONS + +### 0. Definitions + +“This License” refers to version 3 of the GNU General Public License. + +“Copyright” also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + +“The Program” refers to any copyrightable work licensed under this +License. Each licensee is addressed as “you”. “Licensees” and +“recipients” may be individuals or organizations. + +To “modify” a work means to copy from or adapt all or part of the work in +a fashion requiring copyright permission, other than the making of an exact copy. The +resulting work is called a “modified version” of the earlier work or a +work “based on” the earlier work. + +A “covered work” means either the unmodified Program or a work based on +the Program. + +To “propagate” a work means to do anything with it that, without +permission, would make you directly or secondarily liable for infringement under +applicable copyright law, except executing it on a computer or modifying a private +copy. Propagation includes copying, distribution (with or without modification), +making available to the public, and in some countries other activities as well. + +To “convey” a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through a computer +network, with no transfer of a copy, is not conveying. + +An interactive user interface displays “Appropriate Legal Notices” to the +extent that it includes a convenient and prominently visible feature that **(1)** +displays an appropriate copyright notice, and **(2)** tells the user that there is no +warranty for the work (except to the extent that warranties are provided), that +licensees may convey the work under this License, and how to view a copy of this +License. If the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + +### 1. Source Code + +The “source code” for a work means the preferred form of the work for +making modifications to it. “Object code” means any non-source form of a +work. + +A “Standard Interface” means an interface that either is an official +standard defined by a recognized standards body, or, in the case of interfaces +specified for a particular programming language, one that is widely used among +developers working in that language. + +The “System Libraries” of an executable work include anything, other than +the work as a whole, that **(a)** is included in the normal form of packaging a Major +Component, but which is not part of that Major Component, and **(b)** serves only to +enable use of the work with that Major Component, or to implement a Standard +Interface for which an implementation is available to the public in source code form. +A “Major Component”, in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system (if any) on which +the executable work runs, or a compiler used to produce the work, or an object code +interpreter used to run it. + +The “Corresponding Source” for a work in object code form means all the +source code needed to generate, install, and (for an executable work) run the object +code and to modify the work, including scripts to control those activities. However, +it does not include the work's System Libraries, or general-purpose tools or +generally available free programs which are used unmodified in performing those +activities but which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for the work, and +the source code for shared libraries and dynamically linked subprograms that the work +is specifically designed to require, such as by intimate data communication or +control flow between those subprograms and other parts of the work. + +The Corresponding Source need not include anything that users can regenerate +automatically from other parts of the Corresponding Source. + +The Corresponding Source for a work in source code form is that same work. + +### 2. Basic Permissions + +All rights granted under this License are granted for the term of copyright on the +Program, and are irrevocable provided the stated conditions are met. This License +explicitly affirms your unlimited permission to run the unmodified Program. The +output from running a covered work is covered by this License only if the output, +given its content, constitutes a covered work. This License acknowledges your rights +of fair use or other equivalent, as provided by copyright law. + +You may make, run and propagate covered works that you do not convey, without +conditions so long as your license otherwise remains in force. You may convey covered +works to others for the sole purpose of having them make modifications exclusively +for you, or provide you with facilities for running those works, provided that you +comply with the terms of this License in conveying all material for which you do not +control copyright. Those thus making or running the covered works for you must do so +exclusively on your behalf, under your direction and control, on terms that prohibit +them from making any copies of your copyrighted material outside their relationship +with you. + +Conveying under any other circumstances is permitted solely under the conditions +stated below. Sublicensing is not allowed; section 10 makes it unnecessary. + +### 3. Protecting Users' Legal Rights From Anti-Circumvention Law + +No covered work shall be deemed part of an effective technological measure under any +applicable law fulfilling obligations under article 11 of the WIPO copyright treaty +adopted on 20 December 1996, or similar laws prohibiting or restricting circumvention +of such measures. + +When you convey a covered work, you waive any legal power to forbid circumvention of +technological measures to the extent such circumvention is effected by exercising +rights under this License with respect to the covered work, and you disclaim any +intention to limit operation or modification of the work as a means of enforcing, +against the work's users, your or third parties' legal rights to forbid circumvention +of technological measures. + +### 4. Conveying Verbatim Copies + +You may convey verbatim copies of the Program's source code as you receive it, in any +medium, provided that you conspicuously and appropriately publish on each copy an +appropriate copyright notice; keep intact all notices stating that this License and +any non-permissive terms added in accord with section 7 apply to the code; keep +intact all notices of the absence of any warranty; and give all recipients a copy of +this License along with the Program. + +You may charge any price or no price for each copy that you convey, and you may offer +support or warranty protection for a fee. + +### 5. Conveying Modified Source Versions + +You may convey a work based on the Program, or the modifications to produce it from +the Program, in the form of source code under the terms of section 4, provided that +you also meet all of these conditions: + +* **a)** The work must carry prominent notices stating that you modified it, and giving a +relevant date. +* **b)** The work must carry prominent notices stating that it is released under this +License and any conditions added under section 7. This requirement modifies the +requirement in section 4 to “keep intact all notices”. +* **c)** You must license the entire work, as a whole, under this License to anyone who +comes into possession of a copy. This License will therefore apply, along with any +applicable section 7 additional terms, to the whole of the work, and all its parts, +regardless of how they are packaged. This License gives no permission to license the +work in any other way, but it does not invalidate such permission if you have +separately received it. +* **d)** If the work has interactive user interfaces, each must display Appropriate Legal +Notices; however, if the Program has interactive interfaces that do not display +Appropriate Legal Notices, your work need not make them do so. + +A compilation of a covered work with other separate and independent works, which are +not by their nature extensions of the covered work, and which are not combined with +it such as to form a larger program, in or on a volume of a storage or distribution +medium, is called an “aggregate” if the compilation and its resulting +copyright are not used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work in an aggregate +does not cause this License to apply to the other parts of the aggregate. + +### 6. Conveying Non-Source Forms + +You may convey a covered work in object code form under the terms of sections 4 and +5, provided that you also convey the machine-readable Corresponding Source under the +terms of this License, in one of these ways: + +* **a)** Convey the object code in, or embodied in, a physical product (including a +physical distribution medium), accompanied by the Corresponding Source fixed on a +durable physical medium customarily used for software interchange. +* **b)** Convey the object code in, or embodied in, a physical product (including a +physical distribution medium), accompanied by a written offer, valid for at least +three years and valid for as long as you offer spare parts or customer support for +that product model, to give anyone who possesses the object code either **(1)** a copy of +the Corresponding Source for all the software in the product that is covered by this +License, on a durable physical medium customarily used for software interchange, for +a price no more than your reasonable cost of physically performing this conveying of +source, or **(2)** access to copy the Corresponding Source from a network server at no +charge. +* **c)** Convey individual copies of the object code with a copy of the written offer to +provide the Corresponding Source. This alternative is allowed only occasionally and +noncommercially, and only if you received the object code with such an offer, in +accord with subsection 6b. +* **d)** Convey the object code by offering access from a designated place (gratis or for +a charge), and offer equivalent access to the Corresponding Source in the same way +through the same place at no further charge. You need not require recipients to copy +the Corresponding Source along with the object code. If the place to copy the object +code is a network server, the Corresponding Source may be on a different server +(operated by you or a third party) that supports equivalent copying facilities, +provided you maintain clear directions next to the object code saying where to find +the Corresponding Source. Regardless of what server hosts the Corresponding Source, +you remain obligated to ensure that it is available for as long as needed to satisfy +these requirements. +* **e)** Convey the object code using peer-to-peer transmission, provided you inform +other peers where the object code and Corresponding Source of the work are being +offered to the general public at no charge under subsection 6d. + +A separable portion of the object code, whose source code is excluded from the +Corresponding Source as a System Library, need not be included in conveying the +object code work. + +A “User Product” is either **(1)** a “consumer product”, which +means any tangible personal property which is normally used for personal, family, or +household purposes, or **(2)** anything designed or sold for incorporation into a +dwelling. In determining whether a product is a consumer product, doubtful cases +shall be resolved in favor of coverage. For a particular product received by a +particular user, “normally used” refers to a typical or common use of +that class of product, regardless of the status of the particular user or of the way +in which the particular user actually uses, or expects or is expected to use, the +product. A product is a consumer product regardless of whether the product has +substantial commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + +“Installation Information” for a User Product means any methods, +procedures, authorization keys, or other information required to install and execute +modified versions of a covered work in that User Product from a modified version of +its Corresponding Source. The information must suffice to ensure that the continued +functioning of the modified object code is in no case prevented or interfered with +solely because modification has been made. + +If you convey an object code work under this section in, or with, or specifically for +use in, a User Product, and the conveying occurs as part of a transaction in which +the right of possession and use of the User Product is transferred to the recipient +in perpetuity or for a fixed term (regardless of how the transaction is +characterized), the Corresponding Source conveyed under this section must be +accompanied by the Installation Information. But this requirement does not apply if +neither you nor any third party retains the ability to install modified object code +on the User Product (for example, the work has been installed in ROM). + +The requirement to provide Installation Information does not include a requirement to +continue to provide support service, warranty, or updates for a work that has been +modified or installed by the recipient, or for the User Product in which it has been +modified or installed. Access to a network may be denied when the modification itself +materially and adversely affects the operation of the network or violates the rules +and protocols for communication across the network. + +Corresponding Source conveyed, and Installation Information provided, in accord with +this section must be in a format that is publicly documented (and with an +implementation available to the public in source code form), and must require no +special password or key for unpacking, reading or copying. + +### 7. Additional Terms + +“Additional permissions” are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. Additional +permissions that are applicable to the entire Program shall be treated as though they +were included in this License, to the extent that they are valid under applicable +law. If additional permissions apply only to part of the Program, that part may be +used separately under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + +When you convey a copy of a covered work, you may at your option remove any +additional permissions from that copy, or from any part of it. (Additional +permissions may be written to require their own removal in certain cases when you +modify the work.) You may place additional permissions on material, added by you to a +covered work, for which you have or can give appropriate copyright permission. + +Notwithstanding any other provision of this License, for material you add to a +covered work, you may (if authorized by the copyright holders of that material) +supplement the terms of this License with terms: + +* **a)** Disclaiming warranty or limiting liability differently from the terms of +sections 15 and 16 of this License; or +* **b)** Requiring preservation of specified reasonable legal notices or author +attributions in that material or in the Appropriate Legal Notices displayed by works +containing it; or +* **c)** Prohibiting misrepresentation of the origin of that material, or requiring that +modified versions of such material be marked in reasonable ways as different from the +original version; or +* **d)** Limiting the use for publicity purposes of names of licensors or authors of the +material; or +* **e)** Declining to grant rights under trademark law for use of some trade names, +trademarks, or service marks; or +* **f)** Requiring indemnification of licensors and authors of that material by anyone +who conveys the material (or modified versions of it) with contractual assumptions of +liability to the recipient, for any liability that these contractual assumptions +directly impose on those licensors and authors. + +All other non-permissive additional terms are considered “further +restrictions” within the meaning of section 10. If the Program as you received +it, or any part of it, contains a notice stating that it is governed by this License +along with a term that is a further restriction, you may remove that term. If a +license document contains a further restriction but permits relicensing or conveying +under this License, you may add to a covered work material governed by the terms of +that license document, provided that the further restriction does not survive such +relicensing or conveying. + +If you add terms to a covered work in accord with this section, you must place, in +the relevant source files, a statement of the additional terms that apply to those +files, or a notice indicating where to find the applicable terms. + +Additional terms, permissive or non-permissive, may be stated in the form of a +separately written license, or stated as exceptions; the above requirements apply +either way. + +### 8. Termination + +You may not propagate or modify a covered work except as expressly provided under +this License. Any attempt otherwise to propagate or modify it is void, and will +automatically terminate your rights under this License (including any patent licenses +granted under the third paragraph of section 11). + +However, if you cease all violation of this License, then your license from a +particular copyright holder is reinstated **(a)** provisionally, unless and until the +copyright holder explicitly and finally terminates your license, and **(b)** permanently, +if the copyright holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + +Moreover, your license from a particular copyright holder is reinstated permanently +if the copyright holder notifies you of the violation by some reasonable means, this +is the first time you have received notice of violation of this License (for any +work) from that copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + +Termination of your rights under this section does not terminate the licenses of +parties who have received copies or rights from you under this License. If your +rights have been terminated and not permanently reinstated, you do not qualify to +receive new licenses for the same material under section 10. + +### 9. Acceptance Not Required for Having Copies + +You are not required to accept this License in order to receive or run a copy of the +Program. Ancillary propagation of a covered work occurring solely as a consequence of +using peer-to-peer transmission to receive a copy likewise does not require +acceptance. However, nothing other than this License grants you permission to +propagate or modify any covered work. These actions infringe copyright if you do not +accept this License. Therefore, by modifying or propagating a covered work, you +indicate your acceptance of this License to do so. + +### 10. Automatic Licensing of Downstream Recipients + +Each time you convey a covered work, the recipient automatically receives a license +from the original licensors, to run, modify and propagate that work, subject to this +License. You are not responsible for enforcing compliance by third parties with this +License. + +An “entity transaction” is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an organization, or +merging organizations. If propagation of a covered work results from an entity +transaction, each party to that transaction who receives a copy of the work also +receives whatever licenses to the work the party's predecessor in interest had or +could give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if the predecessor +has it or can get it with reasonable efforts. + +You may not impose any further restrictions on the exercise of the rights granted or +affirmed under this License. For example, you may not impose a license fee, royalty, +or other charge for exercise of rights granted under this License, and you may not +initiate litigation (including a cross-claim or counterclaim in a lawsuit) alleging +that any patent claim is infringed by making, using, selling, offering for sale, or +importing the Program or any portion of it. + +### 11. Patents + +A “contributor” is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The work thus +licensed is called the contributor's “contributor version”. + +A contributor's “essential patent claims” are all patent claims owned or +controlled by the contributor, whether already acquired or hereafter acquired, that +would be infringed by some manner, permitted by this License, of making, using, or +selling its contributor version, but do not include claims that would be infringed +only as a consequence of further modification of the contributor version. For +purposes of this definition, “control” includes the right to grant patent +sublicenses in a manner consistent with the requirements of this License. + +Each contributor grants you a non-exclusive, worldwide, royalty-free patent license +under the contributor's essential patent claims, to make, use, sell, offer for sale, +import and otherwise run, modify and propagate the contents of its contributor +version. + +In the following three paragraphs, a “patent license” is any express +agreement or commitment, however denominated, not to enforce a patent (such as an +express permission to practice a patent or covenant not to sue for patent +infringement). To “grant” such a patent license to a party means to make +such an agreement or commitment not to enforce a patent against the party. + +If you convey a covered work, knowingly relying on a patent license, and the +Corresponding Source of the work is not available for anyone to copy, free of charge +and under the terms of this License, through a publicly available network server or +other readily accessible means, then you must either **(1)** cause the Corresponding +Source to be so available, or **(2)** arrange to deprive yourself of the benefit of the +patent license for this particular work, or **(3)** arrange, in a manner consistent with +the requirements of this License, to extend the patent license to downstream +recipients. “Knowingly relying” means you have actual knowledge that, but +for the patent license, your conveying the covered work in a country, or your +recipient's use of the covered work in a country, would infringe one or more +identifiable patents in that country that you have reason to believe are valid. + +If, pursuant to or in connection with a single transaction or arrangement, you +convey, or propagate by procuring conveyance of, a covered work, and grant a patent +license to some of the parties receiving the covered work authorizing them to use, +propagate, modify or convey a specific copy of the covered work, then the patent +license you grant is automatically extended to all recipients of the covered work and +works based on it. + +A patent license is “discriminatory” if it does not include within the +scope of its coverage, prohibits the exercise of, or is conditioned on the +non-exercise of one or more of the rights that are specifically granted under this +License. You may not convey a covered work if you are a party to an arrangement with +a third party that is in the business of distributing software, under which you make +payment to the third party based on the extent of your activity of conveying the +work, and under which the third party grants, to any of the parties who would receive +the covered work from you, a discriminatory patent license **(a)** in connection with +copies of the covered work conveyed by you (or copies made from those copies), or **(b)** +primarily for and in connection with specific products or compilations that contain +the covered work, unless you entered into that arrangement, or that patent license +was granted, prior to 28 March 2007. + +Nothing in this License shall be construed as excluding or limiting any implied +license or other defenses to infringement that may otherwise be available to you +under applicable patent law. + +### 12. No Surrender of Others' Freedom + +If conditions are imposed on you (whether by court order, agreement or otherwise) +that contradict the conditions of this License, they do not excuse you from the +conditions of this License. If you cannot convey a covered work so as to satisfy +simultaneously your obligations under this License and any other pertinent +obligations, then as a consequence you may not convey it at all. For example, if you +agree to terms that obligate you to collect a royalty for further conveying from +those to whom you convey the Program, the only way you could satisfy both those terms +and this License would be to refrain entirely from conveying the Program. + +### 13. Use with the GNU Affero General Public License + +Notwithstanding any other provision of this License, you have permission to link or +combine any covered work with a work licensed under version 3 of the GNU Affero +General Public License into a single combined work, and to convey the resulting work. +The terms of this License will continue to apply to the part which is the covered +work, but the special requirements of the GNU Affero General Public License, section +13, concerning interaction through a network will apply to the combination as such. + +### 14. Revised Versions of this License + +The Free Software Foundation may publish revised and/or new versions of the GNU +General Public License from time to time. Such new versions will be similar in spirit +to the present version, but may differ in detail to address new problems or concerns. + +Each version is given a distinguishing version number. If the Program specifies that +a certain numbered version of the GNU General Public License “or any later +version” applies to it, you have the option of following the terms and +conditions either of that numbered version or of any later version published by the +Free Software Foundation. If the Program does not specify a version number of the GNU +General Public License, you may choose any version ever published by the Free +Software Foundation. + +If the Program specifies that a proxy can decide which future versions of the GNU +General Public License can be used, that proxy's public statement of acceptance of a +version permanently authorizes you to choose that version for the Program. + +Later license versions may give you additional or different permissions. However, no +additional obligations are imposed on any author or copyright holder as a result of +your choosing to follow a later version. + +### 15. Disclaimer of Warranty + +THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. +EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES +PROVIDE THE PROGRAM “AS IS” WITHOUT WARRANTY OF ANY KIND, EITHER +EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF +MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE +QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE +DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + +### 16. Limitation of Liability + +IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL ANY +COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS THE PROGRAM AS +PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, +INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE +PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE +OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE +WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. -END OF TERMS AND CONDITIONS +### 17. Interpretation of Sections 15 and 16 + +If the disclaimer of warranty and limitation of liability provided above cannot be +given local legal effect according to their terms, reviewing courts shall apply local +law that most closely approximates an absolute waiver of all civil liability in +connection with the Program, unless a warranty or assumption of liability accompanies +a copy of the Program in return for a fee. -### How to Apply These Terms to Your New Programs +_END OF TERMS AND CONDITIONS_ -If you develop a new program, and you want it to be of the greatest -possible use to the public, the best way to achieve this is to make it -free software which everyone can redistribute and change under these terms. +## How to Apply These Terms to Your New Programs -To do so, attach the following notices to the program. It is safest -to attach them to the start of each source file to most effectively -convey the exclusion of warranty; and each file should have at least -the “copyright” line and a pointer to where the full notice is found. +If you develop a new program, and you want it to be of the greatest possible use to +the public, the best way to achieve this is to make it free software which everyone +can redistribute and change under these terms. + +To do so, attach the following notices to the program. It is safest to attach them +to the start of each source file to most effectively state the exclusion of warranty; +and each file should have at least the “copyright” line and a pointer to +where the full notice is found. Copyright (C) - - This program is free software; you can redistribute it and/or modify + + This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 2 of the License, or + the Free Software Foundation, either version 3 of the License, or (at your option) any later version. - + This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. - - You should have received a copy of the GNU General Public License along - with this program; if not, write to the Free Software Foundation, Inc., - 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . Also add information on how to contact you by electronic and paper mail. -If the program is interactive, make it output a short notice like this +If the program does terminal interaction, make it output a short notice like this when it starts in an interactive mode: - Gnomovision version 69, Copyright (C) year name of author - Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + Copyright (C) + This program comes with ABSOLUTELY NO WARRANTY; for details type 'show w'. This is free software, and you are welcome to redistribute it - under certain conditions; type `show c' for details. - -The hypothetical commands `show w` and `show c` should show the appropriate -parts of the General Public License. Of course, the commands you use may -be called something other than `show w` and `show c`; they could even be -mouse-clicks or menu items--whatever suits your program. - -You should also get your employer (if you work as a programmer) or your -school, if any, to sign a “copyright disclaimer” for the program, if -necessary. Here is a sample; alter the names: - - Yoyodyne, Inc., hereby disclaims all copyright interest in the program - `Gnomovision' (which makes passes at compilers) written by James Hacker. - - , 1 April 1989 - Ty Coon, President of Vice - -This General Public License does not permit incorporating your program into -proprietary programs. If your program is a subroutine library, you may -consider it more useful to permit linking proprietary applications with the -library. If this is what you want to do, use the GNU Lesser General -Public License instead of this License. + under certain conditions; type 'show c' for details. + +The hypothetical commands `show w` and `show c` should show the appropriate parts of +the General Public License. Of course, your program's commands might be different; +for a GUI interface, you would use an “about box”. + +You should also get your employer (if you work as a programmer) or school, if any, to +sign a “copyright disclaimer” for the program, if necessary. For more +information on this, and how to apply and follow the GNU GPL, see +<>. + +The GNU General Public License does not permit incorporating your program into +proprietary programs. If your program is a subroutine library, you may consider it +more useful to permit linking proprietary applications with the library. If this is +what you want to do, use the GNU Lesser General Public License instead of this +License. But first, please read +<>. diff --git a/NAMESPACE b/NAMESPACE index 85b8c40..c41b10a 100644 --- a/NAMESPACE +++ b/NAMESPACE @@ -10,4 +10,3 @@ export(additive_shap) export(is.kernelshap) export(kernelshap) export(permshap) -importFrom(doFuture,"%dofuture%") diff --git a/NEWS.md b/NEWS.md index 67191cd..2ab5062 100644 --- a/NEWS.md +++ b/NEWS.md @@ -1,4 +1,4 @@ -# kernelshap 0.9.0 +# kernelshap 1.0.0 ### Bug fix @@ -12,23 +12,29 @@ we have fixed a bug in how `kernelshap()` calculates Kernel weights. Fixed in [#168](https://github.com/ModelOriented/kernelshap/pull/168), which also has received unit tests against Python's "shap". +### Major improvements for parallel computation + +We have switched from `%dopar%` to `future_lapply()`: + +- We have a progress bar! Activate via `progressr::handlers(global = TRUE). +- No need for calling `registerDoFuture()` anymore. +- No need to set `parallel = TRUE` anymore. +- No need to declare missing packages anymore via `parallel_args`. Should you still encounter missing packages anyway, + use the new argument `future.packages`. +- Random seeding is now properly handled, thanks [#163](https://github.com/ModelOriented/kernelshap/issues/163) for reporting. + +Implemented in PRs [#170](https://github.com/ModelOriented/kernelshap/pull/170) and [#173](https://github.com/ModelOriented/kernelshap/pull/173). + ### API changes and improvements - The traditional progress bar has been replaced by {progressr}. Activate e.g. via `progressr::handlers(global = TRUE). It works in parallel mode as well, and you can choose your own style ([#173](https://github.com/ModelOriented/kernelshap/pull/173)). - The argument `feature_names` can now also be used with matrix input ([#166](https://github.com/ModelOriented/kernelshap/pull/166)). - `kernelshap()` and `permshap()` have received a `seed = NULL` argument ([#170](https://github.com/ModelOriented/kernelshap/pull/170)). -- Parallel mode: If missing packages or globals have to be specified, this now has to be done through `parallel_args = list(packages = ..., globals = ...)` -instead of `parallel_args = list(.packages = ..., .globals = ...)`, see section on parallelism below. -The list is passed to `[foreach::foreach(.options.future = ...)]`. -### Speed and memory improvements +### Memory improvements - `permshap()` and `kernelshap()` require about 10% less memory ([#166](https://github.com/ModelOriented/kernelshap/pull/166)). -- `permshap()` and `kernelshap()` are faster for data.frame input, - and slightly slower for matrix input ([#166](https://github.com/ModelOriented/kernelshap/pull/166)). -- Additionally, `permshap(, exact = TRUE)` is faster by pre-calculating more - elements used across rows ([#165](https://github.com/ModelOriented/kernelshap/pull/165)). ### Internal changes @@ -36,21 +42,12 @@ The list is passed to `[foreach::foreach(.options.future = ...)]`. - `kernelshap()` solver: Replacing the Moore-Penrose pseudo-inverse by two direct solves, a trick of [Ian Covert](https://github.com/iancovert/shapley-regression/blob/master/shapreg/shapley.py), and ported to R in ([#171](https://github.com/ModelOriented/kernelshap/pull/171)). -### Changes in parallelism - -We have switched from `%dopar%` to `doFuture` ([#170](https://github.com/ModelOriented/kernelshap/pull/170)) with the following impact: - -- No need for calling `registerDoFuture()` anymore. -- No need to set `parallel = TRUE` anymore ([#173](https://github.com/ModelOriented/kernelshap/pull/173)). -- Random seeding is properly handled, and respects `seed`, thanks [#163](https://github.com/ModelOriented/kernelshap/issues/163) for reporting. -- If missing packages or globals have to be specified, this now has to be done through `parallel_args = list(packages = ..., globals = ...)` -instead of `parallel_args = list(.packages = ..., .globals = ...)`. The list is passed to `[foreach::foreach(.options.future = ...)]`. - -### Dependencies +### Change in dependencies - {MASS}: Dropped from Imports -- {doFuture}: Suggests -> Imports -- {progressr}: Suggests +- {foreach}: Dropped from Imports +- {doFuture}: Dropped from suggests and replaced by {future.apply} in Imports +- {progressr}: New in Suggests # kernelshap 0.8.0 diff --git a/R/kernelshap.R b/R/kernelshap.R index d46077e..84f58a5 100644 --- a/R/kernelshap.R +++ b/R/kernelshap.R @@ -8,7 +8,10 @@ #' Otherwise, a partly exact hybrid algorithm combining exact calculations and #' iterative paired sampling is used, see Details. #' -#' To activate the progress bar, run `progressr::handlers(global = TRUE)` first. +#' To activate the progress bar, e.g., run `progressr::handlers(global = TRUE)` first. +#' +#' To activate parallel processing, run `future::plan(multisession)` or similar. +#' To deactivate later, run `plan("sequential")`. #' #' @details #' The pure iterative Kernel SHAP sampling as in Covert and Lee (2021) works like this: @@ -55,8 +58,6 @@ #' should not be higher than 10 for exact calculations. #' For similar reasons, degree 2 hybrids should not use \eqn{p} larger than 40. #' -#' @importFrom doFuture %dofuture% -#' #' @param object Fitted model object. #' @param X \eqn{(n \times p)} matrix or `data.frame` with rows to be explained. #' The columns should only represent model features, not the response @@ -106,16 +107,10 @@ #' For `permshap()`, the default is 0.01, while for `kernelshap()` it is set to 0.005. #' @param max_iter If the stopping criterion (see `tol`) is not reached after #' `max_iter` iterations, the algorithm stops. Ignored if `exact = TRUE`. -#' @param parallel If `TRUE`, use [foreach::foreach()] and `%dofuture%` to loop over rows -#' to be explained. Must register backend beforehand, e.g., `plan(multisession)`, -#' see README for an example. Currently disables the progress bar. -#' @param parallel_args Named list of arguments passed to -#' `foreach::foreach(.options.future = ...)`, ideally `NULL` (default). -#' Only relevant if `parallel = TRUE`. -#' Example on Windows: if `object` is a GAM fitted with package 'mgcv', -#' then one might need to set `parallel_args = list(packages = "mgcv")`. -#' Similarly, if the model has been fitted with `ranger()`, then it might be necessary -#' to pass `parallel_args = list(packages = "ranger")`. +#' @param parallel Deprecated. +#' @param parallel_args Deprecated (see `future.packages`). +#' @param future.packages Character vector with packages to be attached in the +#' R environment evaluating the future. Only if parallel computing. #' @param verbose Set to `FALSE` to suppress messages and the progress bar. #' @param seed Optional integer random seed. Note that it changes the global seed. #' @param survival Should cumulative hazards ("chf", default) or survival @@ -197,12 +192,19 @@ kernelshap.default <- function( max_iter = 100L, parallel = FALSE, parallel_args = NULL, + future.packages = NULL, verbose = TRUE, seed = NULL, ...) { if (parallel) { warning("The 'parallel' argument has been deprecated. Simply set plan(...) to activate parallel computing.") } + if (!is.null(parallel_args)) { + warning( + "The 'parallel_args' argument has been deprecated. + If your parallel sessions lack a package, use argument `future.packages`." + ) + } p <- length(feature_names) basic_checks(X = X, feature_names = feature_names, pred_fun = pred_fun) stopifnot( @@ -268,12 +270,13 @@ kernelshap.default <- function( } # Apply Kernel SHAP to each row of X - future_args <- c(list(seed = TRUE), parallel_args) - parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) - - res <- do.call(foreach::foreach, parallel_args) %dofuture% kernelshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], + res <- future.apply::future_lapply( + seq_len(n), + FUN = kernelshap_one, + future.packages = future.packages, + future.seed = TRUE, + x = X, + v1 = v1, object = object, pred_fun = pred_fun, feature_names = feature_names, diff --git a/R/permshap.R b/R/permshap.R index 3a18e65..9340a54 100644 --- a/R/permshap.R +++ b/R/permshap.R @@ -9,7 +9,10 @@ #' Otherwise, the sampling process iterates until the resulting values #' are sufficiently precise, and standard errors are provided. #' -#' To activate the progress bar, run `progressr::handlers(global = TRUE)` first. +#' To activate the progress bar, e.g., run `progressr::handlers(global = TRUE)` first. +#' +#' To activate parallel processing, run `future::plan(multisession)` or similar. +#' To deactivate later, run `plan("sequential")`. #' #' @details #' During each iteration, the algorithm cycles twice through a random permutation: @@ -107,12 +110,19 @@ permshap.default <- function( max_iter = 10L * length(feature_names), parallel = FALSE, parallel_args = NULL, + future.packages = NULL, verbose = TRUE, seed = NULL, ...) { if (parallel) { warning("The 'parallel' argument has been deprecated. Simply set plan(...) to activate parallel computing.") } + if (!is.null(parallel_args)) { + warning( + "The 'parallel_args' argument has been deprecated. + If your parallel sessions lack a package, use argument `future.packages`." + ) + } p <- length(feature_names) if (p <= 1L) { stop("Case p = 1 not implemented. Use kernelshap() instead.") @@ -177,12 +187,13 @@ permshap.default <- function( } # Apply permutation SHAP to each row of X - future_args <- c(list(seed = TRUE), parallel_args) - parallel_args <- c(list(i = seq_len(n)), list(.options.future = future_args)) - - res <- do.call(foreach::foreach, parallel_args) %dofuture% permshap_one( - x = X[i, , drop = FALSE], - v1 = v1[i, , drop = FALSE], + res <- future.apply::future_lapply( + seq_len(n), + FUN = permshap_one, + future.packages = future.packages, + future.seed = TRUE, + x = X, + v1 = v1, object = object, pred_fun = pred_fun, bg_w = bg_w, diff --git a/R/utils_kernelshap.R b/R/utils_kernelshap.R index 2013908..db99c3d 100644 --- a/R/utils_kernelshap.R +++ b/R/utils_kernelshap.R @@ -2,6 +2,7 @@ # If exact, a single call to predict() is necessary. # If sampling is involved, we need at least two additional calls to predict(). kernelshap_one <- function( + i, x, v1, object, @@ -17,6 +18,8 @@ kernelshap_one <- function( precalc, pbar, ...) { + x <- x[i, , drop = FALSE] + v1 <- v1[i, , drop = FALSE] p <- length(feature_names) K <- ncol(v1) K_names <- colnames(v1) diff --git a/R/utils_permshap.R b/R/utils_permshap.R index e36b93b..60b13cf 100644 --- a/R/utils_permshap.R +++ b/R/utils_permshap.R @@ -6,6 +6,7 @@ #' @keywords internal #' #' @inheritParams permshap +#' @param i Row ID to be explained. #' @param v1 Prediction of `x`. #' @param v0 Average prediction on background data. #' @param x A single row to be explained. @@ -14,6 +15,7 @@ #' a (p x K) matrix of standard errors, number of iterations, #' and convergence status. permshap_one <- function( + i, x, v1, object, @@ -28,6 +30,8 @@ permshap_one <- function( max_iter, pbar, ...) { + x <- x[i, , drop = FALSE] + v1 <- v1[i, , drop = FALSE] bg <- precalc$bg_X_rep p <- length(feature_names) K <- ncol(v1) diff --git a/README.md b/README.md index 339194d..fbb44ac 100644 --- a/README.md +++ b/README.md @@ -24,7 +24,7 @@ To explain your model, select an explanation dataset `X` (up to 1000 rows from t **Remarks to `permshap()` and `kernelshap()`** - Both algorithms need a representative background data `bg_X` to calculate marginal means (up to 500 rows from the training data). In cases with a natural "off" value (like MNIST digits), this can also be a single row with all values set to the off value. If unspecified, 200 rows are randomly sampled from `X`. -- Exact Kernel SHAP gives identical results as exact permutation SHAP. Both algorithms are fast up to 8 features. +- Exact Kernel SHAP gives identical results as exact permutation SHAP. The algorithms are fast for up to 8 features. With more features, `kernelshap()` switches to a partly exact algorithm with faster convergence than the sampling version of permutation SHAP. - For models with interactions of order up to two, the sampling versions provide the same results as the exact versions. - Sampling versions iterate until standard errors of SHAP values are sufficiently small. @@ -113,25 +113,19 @@ The {kernelshap} package can deal with almost any situation. We will show some o ### Parallel computing -Parallel computing for `permshap()` and `kernelshap()` is supported via {doFuture} and {foreach}. -Note that this does not work for all models. - -On Windows, sometimes not all packages or global objects are passed to the parallel sessions. -Often, this can be fixed via `parallel_args` using the arguments "packages" and "globals" passed -to `foreach(.options.future = ...)`, see this example: +Parallel computing for `permshap()` and `kernelshap()` is supported via {future.apply}. ```r -library(doFuture) library(mgcv) -plan(multisession, workers = 4) # Windows -# plan(multicore, workers = 4) # Linux, macOS, Solaris +future::plan(multisession, workers = 4) # Windows +# future::plan(multicore, workers = 4) # Linux, macOS, Solaris # GAM with interactions - we cannot use additive_shap() fit <- gam(log_price ~ s(log_carat) + clarity * color + cut, data = diamonds) system.time( # 4 seconds in parallel - ps <- permshap(fit, X, bg_X = bg_X, parallel_args = list(packages = "mgcv")) + ps <- permshap(fit, X, bg_X = bg_X) ) ps @@ -141,7 +135,7 @@ ps # [2,] -0.51546 -0.1174766 0.11122775 0.030243973 # Because there are no interactions of order above 2, Kernel SHAP gives the same: -plan("sequential") +future::plan("sequential") system.time( # 12 s non-parallel ks <- kernelshap(fit, X, bg_X = bg_X) ) diff --git a/backlog/compare_with_python2.R b/backlog/compare_with_python2.R index 8e23801..565bcf7 100644 --- a/backlog/compare_with_python2.R +++ b/backlog/compare_with_python2.R @@ -33,7 +33,7 @@ kss <- kernelshap( max_iter = 100, tol = 0.0005 ) -kss # -1.198078 -1.246508 -0.9580638 3.877532 -0.3241824 0.541247 +kss # -1.203484 -1.241578 -0.9653531 3.889111 -0.3360786 0.5493293 set.seed(2) ksh <- kernelshap( @@ -46,7 +46,7 @@ ksh <- kernelshap( max_iter = 10000, tol = 0.0005 ) -ksh # -1.191981 -1.240656 -0.9516264 3.86776 -0.3342143 0.5426642 +ksh # -1.195273 -1.240459 -0.9546994 3.884469 -0.3382316 0.5361404 set.seed(1) ksh2 <- kernelshap( @@ -60,4 +60,17 @@ ksh2 <- kernelshap( max_iter = 10000, tol = 0.0001 ) -ksh2 # 1.195976 -1.241107 -0.9565121 3.878891 -0.3384621 0.5451118 +ksh2 # -1.195725 -1.241576 -0.9567066 3.87909 -0.3385894 0.5454539 + +set.seed(1) +ps0 <- permshap( + pf, + head(X, 1), + bg_X = X, + pred_fun = pf, + exact = FALSE, + low_memory = FALSE, + max_iter = 20000, + tol = 0.0001 +) +ps0 # -1.20929 -1.225766 -0.9602608 3.879889 -0.33825 0.5456252 diff --git a/cran-comments.md b/cran-comments.md index 3ba42fb..91699d1 100644 --- a/cran-comments.md +++ b/cran-comments.md @@ -1,11 +1,10 @@ -# kernelshap 0.9.0 +# kernelshap 1.0.0 -We have figured out a bug in the weighting logic of Kernel SHAP. - -This update comes with a fix which has been tested against two other implementations. +- We have fixed a problematic bug in the weighting logic of Kernel SHAP. +- Much better parallel processing via {future.apply}. I am aware that the last release of {kernelshap} is not too long ago, but I still would love to see -this fixed before the (well-deserved) summer break. +this fixed before your (well-deserved) summer break. Thanks a lot! diff --git a/man/kernelshap.Rd b/man/kernelshap.Rd index f274067..ee17f92 100644 --- a/man/kernelshap.Rd +++ b/man/kernelshap.Rd @@ -23,6 +23,7 @@ kernelshap(object, ...) max_iter = 100L, parallel = FALSE, parallel_args = NULL, + future.packages = NULL, verbose = TRUE, seed = NULL, ... @@ -115,17 +116,12 @@ For \code{permshap()}, the default is 0.01, while for \code{kernelshap()} it is \item{max_iter}{If the stopping criterion (see \code{tol}) is not reached after \code{max_iter} iterations, the algorithm stops. Ignored if \code{exact = TRUE}.} -\item{parallel}{If \code{TRUE}, use \code{\link[foreach:foreach]{foreach::foreach()}} and \verb{\%dofuture\%} to loop over rows -to be explained. Must register backend beforehand, e.g., \code{plan(multisession)}, -see README for an example. Currently disables the progress bar.} +\item{parallel}{Deprecated.} -\item{parallel_args}{Named list of arguments passed to -\code{foreach::foreach(.options.future = ...)}, ideally \code{NULL} (default). -Only relevant if \code{parallel = TRUE}. -Example on Windows: if \code{object} is a GAM fitted with package 'mgcv', -then one might need to set \code{parallel_args = list(packages = "mgcv")}. -Similarly, if the model has been fitted with \code{ranger()}, then it might be necessary -to pass \code{parallel_args = list(packages = "ranger")}.} +\item{parallel_args}{Deprecated (see \code{future.packages}).} + +\item{future.packages}{Character vector with packages to be attached in the +R environment evaluating the future. Only if parallel computing.} \item{verbose}{Set to \code{FALSE} to suppress messages and the progress bar.} @@ -166,6 +162,11 @@ By default, for up to p=8 features, exact SHAP values are returned (with respect to the selected background data). Otherwise, a partly exact hybrid algorithm combining exact calculations and iterative paired sampling is used, see Details. + +To activate the progress bar, e.g., run \code{progressr::handlers(global = TRUE)} first. + +To activate parallel processing, run \code{future::plan(multisession)} or similar. +To deactivate later, run \code{plan("sequential")}. } \details{ The pure iterative Kernel SHAP sampling as in Covert and Lee (2021) works like this: diff --git a/man/permshap.Rd b/man/permshap.Rd index 6abcb34..9ecfe8e 100644 --- a/man/permshap.Rd +++ b/man/permshap.Rd @@ -22,6 +22,7 @@ permshap(object, ...) max_iter = 10L * length(feature_names), parallel = FALSE, parallel_args = NULL, + future.packages = NULL, verbose = TRUE, seed = NULL, ... @@ -98,17 +99,12 @@ For \code{permshap()}, the default is 0.01, while for \code{kernelshap()} it is \item{max_iter}{If the stopping criterion (see \code{tol}) is not reached after \code{max_iter} iterations, the algorithm stops. Ignored if \code{exact = TRUE}.} -\item{parallel}{If \code{TRUE}, use \code{\link[foreach:foreach]{foreach::foreach()}} and \verb{\%dofuture\%} to loop over rows -to be explained. Must register backend beforehand, e.g., \code{plan(multisession)}, -see README for an example. Currently disables the progress bar.} +\item{parallel}{Deprecated.} -\item{parallel_args}{Named list of arguments passed to -\code{foreach::foreach(.options.future = ...)}, ideally \code{NULL} (default). -Only relevant if \code{parallel = TRUE}. -Example on Windows: if \code{object} is a GAM fitted with package 'mgcv', -then one might need to set \code{parallel_args = list(packages = "mgcv")}. -Similarly, if the model has been fitted with \code{ranger()}, then it might be necessary -to pass \code{parallel_args = list(packages = "ranger")}.} +\item{parallel_args}{Deprecated (see \code{future.packages}).} + +\item{future.packages}{Character vector with packages to be attached in the +R environment evaluating the future. Only if parallel computing.} \item{verbose}{Set to \code{FALSE} to suppress messages and the progress bar.} @@ -148,6 +144,11 @@ By default, for up to p=8 features, exact SHAP values are returned (exact with respect to the selected background data). Otherwise, the sampling process iterates until the resulting values are sufficiently precise, and standard errors are provided. + +To activate the progress bar, e.g., run \code{progressr::handlers(global = TRUE)} first. + +To activate parallel processing, run \code{future::plan(multisession)} or similar. +To deactivate later, run \code{plan("sequential")}. } \details{ During each iteration, the algorithm cycles twice through a random permutation: diff --git a/packaging.R b/packaging.R index 136f234..e2dcead 100644 --- a/packaging.R +++ b/packaging.R @@ -15,7 +15,7 @@ library(usethis) use_description( fields = list( Title = "Kernel SHAP", - Version = "0.9.0", + Version = "1.0.0", Description = "Efficient implementation of Kernel SHAP (Lundberg and Lee, 2017, ) permutation SHAP, and additive SHAP for model interpretability. @@ -37,14 +37,13 @@ use_description( roxygen = TRUE ) -use_package("doFuture", "Imports") -use_package("foreach", "Imports") +use_package("future.apply", "Imports") use_package("stats", "Imports") use_package("utils", "Imports") use_package("progressr", "Suggests") -use_gpl_license(2) +use_gpl_license(3) # Your files that do not belong to the package itself (others are added by "use_* function") use_build_ignore(c( diff --git a/tests/testthat/test-basic.R b/tests/testthat/test-basic.R index eea4634..5e44853 100644 --- a/tests/testthat/test-basic.R +++ b/tests/testthat/test-basic.R @@ -280,7 +280,7 @@ test_that("kernelshap and permshap work for models with high-order interactions" ) expect_equal( c(ksh2$S), - c(-1.194878, -1.24747, -0.9596389, 3.883523, -0.3349787, 0.5453894), + c(-1.202621, -1.245655, -0.9510531, 3.880391, -0.3384063, 0.5492909), tolerance = 1e-4 ) @@ -299,7 +299,7 @@ test_that("kernelshap and permshap work for models with high-order interactions" ) expect_equal( c(ksh1$S), - c(-1.196958, -1.256924, -0.9603291, 3.886163, -0.3277153, 0.5477104), + c(-1.203028, -1.240056, -0.9574796, 3.89151, -0.3360694, 0.5370693), tolerance = 1e-3 ) @@ -314,13 +314,13 @@ test_that("kernelshap and permshap work for models with high-order interactions" exact = FALSE, m = 10000, max_iter = 10000, - tol = 0.003, + tol = 0.005, verbose = FALSE ) ) expect_equal( c(ksh0$S), - c(-1.18917, -1.2298, -0.9247995, 3.80673, -0.3144175, 0.5434034), + c(-1.212142, -1.238348, -0.9795299, 3.937577, -0.3726428, 0.5570322), tolerance = 1e-3 ) }) @@ -351,13 +351,3 @@ test_that("Random seed works", { expect_false(identical(s1a$S, s2$S)) } }) - -test_that("using foreach (non-parallel) gives the same as normal mode", { - for (algo in c(kernelshap, permshap)) { - s <- algo(fit, iris[J, x], bg_X = iris, verbose = FALSE) - s2 <- suppressWarnings( - algo(fit, iris[J, x], bg_X = iris, verbose = FALSE, parallel = TRUE) - ) - expect_equal(s, s2) - } -}) From 7205b29783440ce0f6cb25151a440e22f855ac21 Mon Sep 17 00:00:00 2001 From: Michael Mayer Date: Wed, 23 Jul 2025 17:49:25 +0200 Subject: [PATCH 3/5] Reverse dependency check --- cran-comments.md | 8 ++++---- revdep/README.md | 34 +++++++++++++++++----------------- 2 files changed, 21 insertions(+), 21 deletions(-) diff --git a/cran-comments.md b/cran-comments.md index 91699d1..460cde2 100644 --- a/cran-comments.md +++ b/cran-comments.md @@ -1,7 +1,7 @@ # kernelshap 1.0.0 - We have fixed a problematic bug in the weighting logic of Kernel SHAP. -- Much better parallel processing via {future.apply}. +- Better parallel processing via {future.apply}. I am aware that the last release of {kernelshap} is not too long ago, but I still would love to see this fixed before your (well-deserved) summer break. @@ -20,8 +20,8 @@ Status: OK ### Revdep OK -✔ survex 1.2.0 ── E: 0 | W: 0 | N: 0 -✔ XAItest 1.0.1 ── E: 1 | W: 0 | N: 0 -✔ SEMdeep 1.0.0 ── E: 1 | W: 1 | N: 0 +✔ SEMdeep 1.0.0 ── E: 0 | W: 0 | N: 0 +✔ XAItest 1.0.1 ── E: 0 | W: 1 | N: 1 +✔ survex 1.2.0 ── E: 0 | W: 0 | N: 0 OK: 3 diff --git a/revdep/README.md b/revdep/README.md index 5e3d79c..56ba958 100644 --- a/revdep/README.md +++ b/revdep/README.md @@ -1,31 +1,31 @@ # Platform -|field |value | -|:--------|:---------------------------------------| -|version |R version 4.4.1 (2024-06-14 ucrt) | -|os |Windows 11 x64 (build 22631) | -|system |x86_64, mingw32 | -|ui |RStudio | -|language |(EN) | -|collate |English_Switzerland.utf8 | -|ctype |English_Switzerland.utf8 | -|tz |Europe/Zurich | -|date |2025-07-21 | -|rstudio |2025.05.0+496 Mariposa Orchid (desktop) | -|pandoc |NA | +|field |value | +|:--------|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| +|version |R version 4.5.1 (2025-06-13 ucrt) | +|os |Windows 11 x64 (build 22631) | +|system |x86_64, mingw32 | +|ui |RStudio | +|language |(EN) | +|collate |English_Switzerland.utf8 | +|ctype |English_Switzerland.utf8 | +|tz |Europe/Zurich | +|date |2025-07-23 | +|rstudio |2025.05.0+496 Mariposa Orchid (desktop) | +|pandoc |NA | +|quarto |ERROR: Unknown command "TMPDIR=C:/Users/mayer/AppData/Local/Temp/RtmpQX07M4/file57e03fb40a5". Did you mean command "create"? @ C:\Users\mayer\AppData\Local\Programs\Quarto\bin\quarto.exe | # Dependencies |package |old |new |Δ | |:------------|:------|:------|:--| -|kernelshap |0.8.0 |0.9.0 |* | +|kernelshap |0.8.0 |1.0.0 |* | |digest |NA |0.6.37 |* | -|doFuture |NA |1.1.2 |* | -|foreach |1.5.2 |1.5.2 | | +|foreach |1.5.2 |NA |* | |future |NA |1.58.0 |* | |future.apply |NA |1.20.0 |* | |globals |NA |0.18.0 |* | -|iterators |1.0.14 |1.0.14 | | +|iterators |1.0.14 |NA |* | |listenv |NA |0.9.1 |* | |parallelly |NA |1.45.0 |* | From c761addf1fb0dbf68d8fececf6ff76eddc72f13a Mon Sep 17 00:00:00 2001 From: Michael Mayer Date: Wed, 23 Jul 2025 19:57:15 +0200 Subject: [PATCH 4/5] Reduce number of progress updates as they cost time --- R/kernelshap.R | 4 +++- R/permshap.R | 4 +++- R/utils_kernelshap.R | 9 +++++++-- R/utils_permshap.R | 9 +++++++-- backlog/compare_with_python.R | 2 ++ tests/testthat/test-deprecation.R | 18 ++++++++++++++++++ 6 files changed, 40 insertions(+), 6 deletions(-) create mode 100644 tests/testthat/test-deprecation.R diff --git a/R/kernelshap.R b/R/kernelshap.R index 84f58a5..f294c14 100644 --- a/R/kernelshap.R +++ b/R/kernelshap.R @@ -265,8 +265,9 @@ kernelshap.default <- function( warning_burden(max(m, m_exact), bg_n = bg_n) } + pbar_step <- max(1L, n %/% 20L) pbar <- if (verbose && n >= 2L && requireNamespace("progressr", quietly = TRUE)) { - progressr::progressor(n) + progressr::progressor(ceiling(n / pbar_step)) } # Apply Kernel SHAP to each row of X @@ -289,6 +290,7 @@ kernelshap.default <- function( v0 = v0, precalc = precalc, pbar = pbar, + pbar_step = pbar_step, ... ) diff --git a/R/permshap.R b/R/permshap.R index 9340a54..1226806 100644 --- a/R/permshap.R +++ b/R/permshap.R @@ -182,8 +182,9 @@ permshap.default <- function( warning_burden(max(m_eval, m_exact), bg_n = bg_n) } + pbar_step <- max(1L, n %/% 20L) pbar <- if (verbose && n >= 2L && requireNamespace("progressr", quietly = TRUE)) { - progressr::progressor(n) + progressr::progressor(ceiling(n / pbar_step)) } # Apply permutation SHAP to each row of X @@ -205,6 +206,7 @@ permshap.default <- function( tol = tol, max_iter = max_iter, pbar = pbar, + pbar_step = pbar_step, ... ) diff --git a/R/utils_kernelshap.R b/R/utils_kernelshap.R index db99c3d..d41c678 100644 --- a/R/utils_kernelshap.R +++ b/R/utils_kernelshap.R @@ -17,6 +17,7 @@ kernelshap_one <- function( v0, precalc, pbar, + pbar_step, ...) { x <- x[i, , drop = FALSE] v1 <- v1[i, , drop = FALSE] @@ -42,9 +43,11 @@ kernelshap_one <- function( # Some of the hybrid cases are exact as well if (exact || trunc(p / 2) == deg) { beta <- solver(A_exact, b_exact, constraint = v1 - v0) # (p x K) - if (!is.null(pbar)) { + + if (!is.null(pbar) && (i %% pbar_step == 0L)) { pbar() } + return(list(beta = beta)) } } @@ -98,9 +101,11 @@ kernelshap_one <- function( } } out <- list(beta = beta_n, sigma = sigma_n, n_iter = n_iter, converged = converged) - if (!is.null(pbar)) { + + if (!is.null(pbar) && (i %% pbar_step == 0L)) { pbar() } + return(out) } diff --git a/R/utils_permshap.R b/R/utils_permshap.R index 60b13cf..4769f1f 100644 --- a/R/utils_permshap.R +++ b/R/utils_permshap.R @@ -29,6 +29,7 @@ permshap_one <- function( tol, max_iter, pbar, + pbar_step, ...) { x <- x[i, , drop = FALSE] v1 <- v1[i, , drop = FALSE] @@ -58,9 +59,11 @@ permshap_one <- function( w = precalc$shapley_w[pos$on] ) } - if (!is.null(pbar)) { + + if (!is.null(pbar) && (i %% pbar_step == 0L)) { pbar() } + return(list(beta = beta_n)) } @@ -135,9 +138,11 @@ permshap_one <- function( out <- list( beta = beta_n / n_iter, sigma = sigma_n, n_iter = n_iter, converged = converged ) - if (!is.null(pbar)) { + + if (!is.null(pbar) && (i %% pbar_step == 0L)) { pbar() } + return(out) } diff --git a/backlog/compare_with_python.R b/backlog/compare_with_python.R index 936464f..ad3c8c8 100644 --- a/backlog/compare_with_python.R +++ b/backlog/compare_with_python.R @@ -1,6 +1,8 @@ library(ggplot2) library(kernelshap) +progressr::handlers(global = TRUE) + # Turn ordinal factors into unordered ord <- c("clarity", "color", "cut") diamonds[, ord] <- lapply(diamonds[ord], factor, ordered = FALSE) diff --git a/tests/testthat/test-deprecation.R b/tests/testthat/test-deprecation.R new file mode 100644 index 0000000..380a09e --- /dev/null +++ b/tests/testthat/test-deprecation.R @@ -0,0 +1,18 @@ +test_that("Deprecation warnings work", { + fit <- lm(Sepal.Length ~ ., data = iris) + + for (algo in c(kernelshap, permshap)) { + expect_warning( + algo(fit, X = iris[1:2, -1], bg_X = iris, parallel = TRUE, verbose = FALSE) + ) + expect_warning( + algo( + fit, + X = iris[1:2, -1], + bg_X = iris, + parallel_args = list(a = 1), + verbose = FALSE + ) + ) + } +}) From faef166a87800eb5f6a8821ccfd44e8eec561794 Mon Sep 17 00:00:00 2001 From: Michael Mayer Date: Wed, 23 Jul 2025 20:03:43 +0200 Subject: [PATCH 5/5] Hint that not *all* models are possible with parallel computing, e.g. keras --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index fbb44ac..0ccb86d 100644 --- a/README.md +++ b/README.md @@ -113,7 +113,7 @@ The {kernelshap} package can deal with almost any situation. We will show some o ### Parallel computing -Parallel computing for `permshap()` and `kernelshap()` is supported via {future.apply}. +Parallel computing for `permshap()` and `kernelshap()` is supported via {future.apply}. It works for most, but not all models (e.g., keras). ```r library(mgcv)