-
Notifications
You must be signed in to change notification settings - Fork 14
/
NEWS
377 lines (237 loc) · 12.5 KB
/
NEWS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
Changes in version 1.30.1:
o fct johnsonshap() has been corrected and does work now,
with bootstrap, argument model=NULL and tell() usage
Changes in version 1.30.0:
o Nex fct EPtest
o New fct johnsonshap() (Warning: does not work yet)
Changes in version 1.29.0:
o New fct johnson()
o Bugs corrected in testHSIC.R concerning some functions' names
that have been renamed
Changes in version 1.28.1:
o Bug corrected in fct parameterSets()
o Bugs corrected about all the ggplot.xxx() fcts issues (required by CRAN)
o Correction in the tell() fct of sensiHSIC() for direct call
Changes in version 1.28.0:
o New fct correlratio()
o Many bugs corrected in fct sensiHSIC()
o Many additional options added in fct sensiHSIC(), especially HSIC-ANOVA indices
o New fct testHSIC() for independence testing based on HSIC indices
o New benchmark matyas.fun() added in "testmodels.R"
Changes in version 1.27.1:
o Adding the topic of 'interpretability of machine learning models'
in the description of the package
o emvd() fct and names have been changed to pmvd()
o Several bugs corrected in pme_knn()
o R requirements: Changes from 'class()' to 'inherits()' in several fcts
o Corrections in parameterSets(),
see https://github.com/cran/sensitivity/pull/2
Changes in version 1.27.0:
o New fct pme_knn() for proportional marginal effects computation
o Addition of an example about visualiz. of perturbed law in PLI() fct
Changes in version 1.26.1:
o Simplification in shapleyBlockestimation.R and
parameter values changes in shapleyBlockestimation.Rd
o extract and ggplot functionalities added in shapleysobol_knn() fct
o new examples for nice visualizations (via plotCI() of plotrix package)
in fcts PLIquantile(), PLIsuperquantile(), PLIquantile_multivar()
and PLIsuperquantile_multivar()
Changes in version 1.26.0:
o New function shapleysobol_knn() which allows optimized and parallelized
computations of Shapley effects/Sobol' indices using the
nearest-neighbors method. This fct also replicates the same functionalities
than sobolshap_knn() that will be removed in a future package's version
o \donttest examples have been changed in the .Rd of several functions
in order to shorten their execution (because the tests are judged too long
by CRAN)
o Bug corrected in all PLIxxx() fcts:
- conditions (delta != 0) replaced by conditions (delta != mu),
mu being the input mean)
- pti=phi(lambda[K]) is not recomputed at each bootstrap iteration
(causing numerical instabilities)
o Bug corrected in PLIquantile_multivar() and PLIsuperquantile_multivar()
fcts (conditions "|" instead of "&")
o Correction in docs of PLIquantile_multivar() and
PLIsuperquantile_multivar() to precise that the delta are perturbed means
o In the fct shapleyPermrand(), the test about 'Multiserver queue model'
has been suppressed (unsolved bug in Xset)
Changes in version 1.25.0:
o New function PLIsuperquantile_multivar()
o Bootstrap added in PLIquantile_multivar()
o Bug corrected in weightTSA() (data structures)
o Change in lmg(): new \code{max.iter} argument to increase the maximum
number of iterative optimization step for logistic regression
o Change in emvd():
- new \code{max.iter} argument to increase the maximum number of
iterative optimization step for logistic regression
- new \code{tol} argument to detect spurious variables
- faster implementation
- recursive emvd computation can now be partially done in parallel
o Add an example with Gaussian inverse transforms in PLI() fct
Changes in version 1.24.1:
o Bug for y format in lmg() and emvd(): allow dataframe
Changes in version 1.24.0:
o New function lmg() for LMG sensitivity indices
o New function emvd() for E-MVD sensitivity indices
o Package foreach is necessary (Imports)
Changes in version 1.23.1:
o Bug of tests of sensiHSIC()
o Bug in sobolshap_knn() (matrix option in argument U)
Changes in version 1.23.0:
o New function qosa() computing quantile-oriented sensitivity analysis
via the contrast-function formulation
o Corrections in sensiHSIC() for
- taking into account categorical inputs (and the choice of their
hyperparameters)
- taking into account multiple outputs when some are categorical
o Corrections in sobolshap_knn() for taking into account categorical
inputs
o Correction in bootstats() relative to the use of sobolrep() fct
Changes in version 1.22.2:
o Correction in PLIsuperquantile() for "bias=TRUE" & "type=VAR" case
o New toy function in testmodels(): Friedman function
o New argument 'varoutput' in delsa() in order to be able to
take the output variance as the empirical variance from the sample
o New argument 'randperm' in sobolshap_knn() in order to use or not
random permutations to estimate Shapley effects
Changes in version 1.22.1:
o Correction in sensiHSIC.R with test.method = "No" in order to
not perform the statistical tests
o Correction in nested.cpp
o Change of colors in shapleyBlockEstimation.Rd
Changes in version 1.22.0:
o New function shapleyBlockEstimation()
o New functionalities in sensiHSIC(): PCA, kernel for categorical
inputs and/or outputs, optimization of permutations numbers,
convergence plot for p-values, U-stat estimator for conditional HSIC
o Corrected plot with new fct plotMultOut() (for ubiquitous Sobol'
indices) in sobolMultOut() and associated sobolxxx() fcts
o New bootstrap process in PLIquantile() and PLIsuperquantile() via
the new argument bootsample
Changes in version 1.21.0:
o New function sobolshap_knn() (sensitivity analysis via
ranking / nearest neighbours)
o New functionalities in sensiHISIC(): aggregated HSIC
Changes in version 1.20.0:
o New functions sobolrep() (Sobol' indices estimation based on
replicated orthogonal arrays) and sobolrec() (Recursive estimation
of Sobol' indices), which use the new functions addelman_const(),
discrepancyCriteria_cplus() and maximin_cplus() (discrepancy and
maximin criteria using a C++ implementation)
o bootstrap implemented in sobolrank()
o New function shapleyLinearGaussian() which computes Shapley effects
for linear models with Gaussian inputs
o bug corrected in shapleySubsetMc() with the modified line:
cost=sum(apply(U,1,function(u) round(Ntot/choose(p,sum(u))/(p-1))))
Changes in version 1.19.0:
o New author Sebastien Da veiga
o New function sobolrank()
o Functions src() and pcc() adapted to work with logistic GLM model
(via the new argument logistic = TRUE)
o New argument 'title' in ggplot.shapleyPermRand and
ggplot.shapleyPermEx
o bug solved in shapleyPermEx() in order to compute the CI for the
full first order and independent total Sobol' indices
o New argument 'noise' to consider noisy observations in
shapleySubsetMc()
Changes in version 1.18.1:
o New argument semi (= TRUE or FALSE) in function pcc() in order
to compute semi-partial correlation coefficient (SPCC)
o bug (caused by new R version 4.0.0) solved in:
- sobolmartinez.R, base.R, morris.R, morris_oat.R and simplex.R
(matrix objects now also inherit from class "array")
- sobolMultOut.Rd, testmodels.Rd and delsa.Rd (screen devices
x11() should not be used in examples)
Changes in version 1.18.0:
o New function: weightTSA.R in order to perform Target Sensitivity
Analysis by transforming the output sample using aspecific weight
function
o Improvements in sensiHSIC(): 1) for the parameters estimation via
the new argument crit.option, 2) for a sequential estimation of the
pvalue (independence test based on permutations), 3) new argument
target in order to perform Target Sensitivity Analysis
o New argument 'bias' in PLIsuperquantile(): "TRUE" gives the
previous computation; "FALSE" corrects this computation (the sample
points were not weighted by the likelihood ratio when computing the
perturbed superquantile)
Changes in version 1.17.1:
o bug (caused by new R version 4.0.0) solved in morris_oat.R
(matrix objects now also inherit from class "array")
o exception "multiIndex[inputIndex] == 0 & der" solved in
PoincareChaosSqCoef.R
o example with given data in sensiHSIc()
Changes in version 1.17.0:
o New functions: squaredIntEstim.R and PoincareChaosSqCoef.R
(which allows to compute the squared coefficients in
generalized chaos via Poincare differential operators)
o Modification in PoincareOptimal.Rd concerning the
returned eigenvalues
Changes in version 1.16.3:
o Unresolved matrix/array classes related bug on R4.0.0 version
in morris.R, morris_oat.R, soboljansen.R, sobolmartinez.R
and sobolroalhs.R
Examples that do not work have been put in \donttest{} (these
examples work in practice)
o Resolved bug in plot.soboljansen() and plot.sobolmartinez
when dealing with matrice-type model outputs
o ylabel changed (from S to PLI) when plotting PLI examples
o legends precised (between full and independent Sobol' indices)
when plotting Shapley effects
Changes in version 1.16.2:
o New functionalities in sensiHSIC() function
(statistical tests of independence)
o Graph legends corrected in sensiHSIC(), sensiFdiv(), sb() and
sobolMultOut() functions
o Corrected bug in nodeggplot()
Changes in version 1.16.1:
o New function nodeggplot() (in order to ebautify the plots)
o ggplot functions for the following functions:
pcc, src, sensiFdiv, sensiHSIC, shapleyPermEx, shapleyPermrand,
sobol, sobol2002, sobol2007, sobolEff, soboljansen, sobolmara,
sobolmartinez, sobolMultOut, sobolowen, sobolroalhs, sobolroauc,
sobolsalt, sobolTIIlo, sobolTIIpf, soboltouati
o Functions sobolroalhs() and sobolSalt(): Modifications of
plot functions in order to have "XiXj" instead of "Xij""
o New output of function delsa(): Brute values of derivatives
o Test case corrected in pcc.Rd
Changes in version 1.16.0:
o Added functions for calculating Shapley indices from data:
shapleySubsetMC()
o Added functions for calculating PLI: PLIsuperquantile() and
PLIquantile_multivar()
o In the 3 PLI functions related to quantiles:
- new arguments in order to have the PLI in percentage
- the result also contains the perturbed quantiles values
- bootstrap confidence intervals can be obtained
o Added the toy function heterdisc.fun()
Changes in version 1.15.2:
o Added functions for Morris method for multidimensional outputs:
morrisMultOut()
o Added the toy function linkletter.fun()
o Change the names of the fcts (dtnorm, ..., dtgumbel, ...) by
(dnorm.trunc, ...) (reverse dependency pb with ATmet package).
Export now the fcts, documentation entries in truncateddistrib.Rd
o Corrected bug: Replace \dontrun by \donttest in all Rd-files
o Corrected bug: package triangle is suggested and called in
PoincareOptimal.Rd
o Function in sobolGPmethods.R and sobolpickfreeze.R have been
integrated in sobolGP.R
o Detected bug (not corrected) in tell.sobolGP()
Unused argument in km(), passed by eval(), update(), tell.sobolGP() ??)
=> comment in the example of sobolGP.Rd
Changes in version 1.15.1:
o Functions shapleyPermEx() corrected (confidence intervals)
Changes in version 1.15.0:
o Function "sobolCert.R" has been (temporarily) suppressed
(compilation bug unresolved)
o Function "support.R" has been changed by O. Roustant (new plotting
functionalities are available, as the function "scatterplot")
o Resolved bug about "sobolmartinez" method (now suppressed) in
the sobolMultOut() function
o sobolMultOut() now returns the functional Sobol' indices via
the attributes "Sfct" and "Tfct" of the output object
Changes in version 1.14.0:
o Added functions for calculating Shapley indices: shapleyPermEx()
and shapleyPermRand()
o Added function for calculating first order Sobol' indices via
smoothing technique: sobolSmthSpl()