make force_plot compatible with shap >= 0.36.0

.Rproj.user/shared/notebooks/paths

@@ -1,3 +0,0 @@
-/Users/bgreenwell/Dropbox/devel/fastshap/R/gen_pkg_bib.R="E1B8FA98"
-/Users/bgreenwell/Dropbox/devel/fastshap/rjournal/greenwell.Rmd="44C6DFCC"
-/Users/bgreenwell/Dropbox/devel/fastshap/rjournal/greenwell.bib="6C1920D6"

DESCRIPTION

@@ -36,4 +36,4 @@ Suggests:
 LinkingTo: 
     Rcpp,
     RcppArmadillo
-RoxygenNote: 7.0.2
+RoxygenNote: 7.1.1

NEWS.md

@@ -1,3 +1,9 @@
+# fastshap 0.0.5.9000
+
+## Bug fixes
+
+* The `force_plot()` function should now be compatible with **shap** (>=0.36.0); thanks to @hfshr and @jbwoillard for reporting [(#12)](https://github.com/bgreenwell/fastshap/issues/12).
+
 # fastshap 0.0.5
 
 ## Bug fixes

R/force_plot.R

@@ -110,6 +110,7 @@ force_plot.explain <- function(object, baseline = NULL, feature_values = NULL,
 
   # Import shap module and construct HTML for plot
   shap <- reticulate::import("shap")
+
   # shap$initjs()
   fp <- shap$force_plot(
     base_value = if (is.null(baseline)) 0 else baseline, 
@@ -121,8 +122,15 @@ force_plot.explain <- function(object, baseline = NULL, feature_values = NULL,
   )
 
   # Display results
+  # FIXME: Is this the best way to determine/compare Python package versions?
+  shap_version <- reticulate::py_get_attr(shap, name = "__version__")
+  shap_version <- package_version(as.character(shap_version))
   tfile <- tempfile(fileext = ".html")
-  shap$save_html(tfile, plot_html = fp)
+  if (shap_version < "0.36.0") {
+    shap$save_html(tfile, plot_html = fp)
+  } else {
+    shap$save_html(tfile, plot = fp)
+  }
   display <- match.arg(display)
   # Check for dependencies
   if (display == "viewer") {

inst/tinytest/test_force_plot.R

@@ -22,7 +22,8 @@ trn <- gen_friedman(500, seed = 101)
 X <- subset(trn, select = -y)
 
 # Fit a MARS model to the simulated Friedman benchmark data
-mars <- eartth::earth(y ~ ., data = trn, degree = 2)
+mars <- earth::earth(y ~ ., data = trn, degree = 2)
+preds <- predict(mars, newdata = trn)[, 1L, drop = TRUE]
 
 # Prediction wrapper
 pfun <- function(object, newdata) {

rjournal/greenwell.Rmd

@@ -59,10 +59,11 @@ In particular, the Shapley contribution of the $i$-th feature to an instance $x$
 \nonumber
 \phi_i\left(x\right) = \frac{1}{p!} \sum_{\mathcal{O} \in \pi\left(p\right)} \left[\Delta Pre^i\left(\mathcal{O}\right) \cup \left\{i\right\} - Pre^i\left(\mathcal{O}\right)\right], \quad i = 1, 2, \dots, p,
 \end{equation}
-where ...
+where $\pi\left(p\right)$ is the set of all permutations of feature indices $\left\{1, 2, \dots, p\right\}$, and $Pre^i\left(\mathcal{O}\right)$ is the set of all feature indices that appear before $i$ in $\mathcal{O} \in \pi\left(p\right)$.
 
 A simple example may help clarify the main ideas.
 
+
 ### Fairly splitting a bar tab
 
 Alex, Brad, and Brandon decide to go out for drinks after work. They shared a few pitchers of beer, but nobody payed attention to how much each person drank. What's a fair way to split the tab? Suppose we knew the follow information, perhaps based on historical data:
@@ -111,7 +112,7 @@ So the next time the bartender asks how you want to split the tab, whip out a pe
 
 ## Estimating Shapley values via Monte Carlo simulation: SampleSHAP
 
-A single estimate of the contribution of $x_i$ to $f\left(x\right)$ is nothing the more than the difference between two predictions, where each prediction is based on a sort of Frankenstein instance that' are's constructed by swapping out values between the instance being explained ($x$) and an instance selected at random from the training data. To help stabilize the results, the procedure is repeated a large number, say, $R$, times, and the result averaged together. 
+If we assume the features are independent, then $\phi_i\left(x\right)$ can be approximated via a simple Monte-Carlo simulation. In particular, a single estimate of the contribution of $x_i$ to $f\left(x\right)$ is nothing the more than the difference between two predictions, where each prediction is based on a sort of Frankenstein instance that' are's constructed by swapping out values between the instance being explained ($x$) and an instance selected at random from the training data. To help stabilize the results, the procedure is repeated a large number, say, $R$, times, and the result averaged together. 
 
 \begin{algorithm}
 \begin{enumerate}
@@ -157,7 +158,7 @@ Recall that the contribution of the $i$-th feature to the prediction $f\left(X\r
   &= \beta_i \left(X_i - \E\left(X_i\right)\right)
 \end{split},
 \end{equation}
-where we estimate $\E\left(X_i\right)$ with the corresponding sample mean $\bar{X}_i$. The quantity $\phi_i\left(X\right)$ is also referred to as the \emph{situational importance of $X_i$} \citep{achen-1982-interpreting}.
+where we estimate $\E\left(X_i\right)$ with the corresponding sample mean $\bar{X}_i$. The quantity $\phi_i\left(X\right)$ is also referred to as the \emph{situational importance of $X_i$} \citep{achen-1982-interpreting}. So, for additive linear models, the contribution of feature $X_i$ is the difference between the feature's effect minus the average effect.
 
 
 ### Tree-based models: TreeSHAP

rjournal/greenwell.bib

@@ -1,135 +1,112 @@
-% Books ------------------------------------------------------------------------
-
-@book{achen-1982-interpreting,
-  title = {Interpreting and Using Regression},
-  author = {Achen, Christopher H.},
-  isbn = {9780803900004},
-  lccn = {82042675},
-  series = {Interpreting and Using Regression},
-  year = {1982},
-  publisher = {Sage Publications}
+@Manual{R-Rcpp,
+  title = {Rcpp: Seamless R and C++ Integration},
+  author = {Dirk Eddelbuettel and Romain Francois and JJ Allaire and Kevin Ushey and Qiang Kou and Nathan Russell and Douglas Bates and John Chambers},
+  year = {2020},
+  note = {R package version 1.0.5},
+  url = {https://CRAN.R-project.org/package=Rcpp},
 }
 
-
-% Articles ---------------------------------------------------------------------
-
-@article{strumbelj-2014-explaining,
-  author = {{Š}trumbelj, Erik and Kononenko, Igor},
-  title = {Explaining prediction models and individual predictions with feature contributions},
-  journal = {Knowledge and Information Systems},
-  year = {2014},
-  volume = {31},
-  number = {3},
-  pages = {647--665},
-  url = {https://doi.org/10.1007/s10115-013-0679-x}
+@Manual{R-SHAPforxgboost,
+  title = {SHAPforxgboost: SHAP Plots for XGBoost},
+  author = {Yang Liu and Allan Just},
+  year = {2021},
+  note = {R package version 0.1.0},
+  url = {https://github.com/liuyanguu/SHAPforxgboost},
 }
 
-@article{gosiewska-2019-iBreakDown,
-  author    = {Alicja Gosiewska and Przemyslaw Biecek},
-  title     = {iBreakDown: Uncertainty of Model Explanations for Non-additive Predictive
-               Models},
-  journal   = {CoRR},
-  volume    = {abs/1903.11420},
-  year      = {2019},
-  url       = {http://arxiv.org/abs/1903.11420},
-  archivePrefix = {arXiv},
-  eprint    = {1903.11420},
-  timestamp = {Tue, 02 Apr 2019 11:16:55 +0200},
-  biburl    = {https://dblp.org/rec/journals/corr/abs-1903-11420.bib},
-  bibsource = {dblp computer science bibliography, https://dblp.org}
+@Manual{R-fastshap,
+  title = {fastshap: Fast Approximate Shapley Values},
+  author = {Brandon Greenwell},
+  note = {R package version 0.0.5},
+  url = {https://github.com/bgreenwell/fastshap},
+  year = {2020},
 }
 
-@article{veronika-2017-catboost,
-  author = {Anna Veronika Dorogush and Andrey Gulin and Gleb Gusev and Nikita Kazeev and Liudmila Ostroumova Prokhorenkova and
-               Aleksandr Vorobev},
-  title = {Fighting biases with dynamic boosting},
-  journal = {CoRR},
-  volume = {abs/1706.09516},
-  year = {2017},
-  url = {http://arxiv.org/abs/1706.09516},
-  archivePrefix = {arXiv},
-  eprint = {1706.09516},
-  timestamp = {Mon, 13 Aug 2018 16:46:23 +0200},
-  biburl = {https://dblp.org/rec/journals/corr/DorogushGGKPV17.bib},
-  bibsource = {dblp computer science bibliography, https://dblp.org}
+@Manual{R-iBreakDown,
+  title = {iBreakDown: Model Agnostic Instance Level Variable Attributions},
+  author = {Przemyslaw Biecek and Alicja Gosiewska and Hubert Baniecki and Adam Izdebski},
+  year = {2020},
+  note = {R package version 1.3.1},
+  url = {https://CRAN.R-project.org/package=iBreakDown},
 }
 
-@article{chen-2016-xgboost,
-  author = {Tianqi Chen and Carlos Guestrin},
-  title = {XGBoost: {A} Scalable Tree Boosting System},
-  journal = {CoRR},
-  volume = {abs/1603.02754},
-  year = {2016},
-  url = {http://arxiv.org/abs/1603.02754},
-  archivePrefix = {arXiv},
-  eprint = {1603.02754},
-  timestamp = {Mon, 13 Aug 2018 16:47:00 +0200},
-  biburl = {https://dblp.org/rec/journals/corr/ChenG16.bib},
-  bibsource = {dblp computer science bibliography, https://dblp.org}
+@Manual{R-iml,
+  title = {iml: Interpretable Machine Learning},
+  author = {Christoph Molnar and Patrick Schratz},
+  year = {2020},
+  note = {R package version 0.10.1},
+  url = {https://CRAN.R-project.org/package=iml},
 }
 
-@article{R-sellereite,
-  doi = {10.21105/joss.02027},
-  url = {https://doi.org/10.21105/joss.02027},
-  year = {2019},
-  publisher = {The Open Journal},
-  volume = {5},
-  number = {46},
-  pages = {2027},
-  author = {Nikolai Sellereite and Martin Jullum},
-  title = {shapr: An R-package for explaining machine learning models with dependence-aware Shapley values},
-  journal = {Journal of Open Source Software}
+@Manual{R-reticulate,
+  title = {reticulate: Interface to Python},
+  author = {Kevin Ushey and JJ Allaire and Yuan Tang},
+  year = {2020},
+  note = {R package version 1.18},
+  url = {https://github.com/rstudio/reticulate},
 }
 
+@Manual{R-shapper,
+  title = {shapper: Wrapper of Python Library shap},
+  author = {Szymon Maksymiuk and Alicja Gosiewska and Przemyslaw Biecek},
+  year = {2020},
+  note = {R package version 0.1.3},
+  url = {https://github.com/ModelOriented/shapper},
+}
 
-% Other ------------------------------------------------------------------------
+@Article{Rcpp2011,
+  title = {{Rcpp}: Seamless {R} and {C++} Integration},
+  author = {Dirk Eddelbuettel and Romain Fran\c{c}ois},
+  journal = {Journal of Statistical Software},
+  year = {2011},
+  volume = {40},
+  number = {8},
+  pages = {1--18},
+  url = {http://www.jstatsoft.org/v40/i08/},
+  doi = {10.18637/jss.v040.i08},
+}
 
-@misc{aas-2019-explaining,
-  title = {Explaining individual predictions when features are dependent: More accurate approximations to Shapley values},
-  author = {Kjersti Aas and Martin Jullum and Anders Løland},
-  year = {2019},
-  eprint = {1903.10464},
-  archivePrefix = {arXiv},
-  primaryClass = {stat.ML}
+@Book{Rcpp2013,
+  title = {Seamless {R} and {C++} Integration with {Rcpp}},
+  author = {Dirk Eddelbuettel},
+  publisher = {Springer},
+  address = {New York},
+  year = {2013},
+  note = {ISBN 978-1-4614-6867-7},
+  doi = {10.1007/978-1-4614-6868-4},
 }
 
-@misc{R-shapFlex,
-  title = {Shapley Decomposition of R-Squared in Machine Learning Models},
-  author = {Nickalus Redell},
-  year = {2019},
-  eprint = {1908.09718},
-  archivePrefix = {arXiv},
-  primaryClass = {stat.ME}
+@Article{Rcpp2017,
+  title = {{Extending 	extit{R} with 	extit{C++}: A Brief Introduction to 	extit{Rcpp}}},
+  author = {Dirk Eddelbuettel and James Joseph Balamuta},
+  journal = {PeerJ Preprints},
+  year = {2017},
+  month = {aug},
+  volume = {5},
+  pages = {e3188v1},
+  issn = {2167-9843},
+  url = {https://doi.org/10.7287/peerj.preprints.3188v1},
+  doi = {10.7287/peerj.preprints.3188v1},
 }
 
-@misc{frye-2019-asymmetric,
-  title = {Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability},
-  author = {Christopher Frye and Ilya Feige and Colin Rowat},
+@Misc{iBreakDown2019,
+  title = {Do Not Trust Additive Explanations},
+  author = {Alicja Gosiewska and Przemyslaw Biecek},
   year = {2019},
-  eprint = {1910.06358},
-  archivePrefix = {arXiv},
-  primaryClass = {stat.ML}
+  eprint = {arXiv:1903.11420},
+  url = {https://arxiv.org/abs/1903.11420},
 }
 
-@incollection{lundberg-2017-KernelSHAP,
-	Author = {Lundberg, Scott M and Lee, Su-In},
-	Booktitle = {Advances in Neural Information Processing Systems 30},
-	Editor = {I. Guyon and U. V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett},
-	Pages = {4765--4774},
-	Publisher = {Curran Associates, Inc.},
-	Title = {A Unified Approach to Interpreting Model Predictions},
-	Url = {http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf},
-	Year = {2017},
-	Bdsk-Url-1 = {http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf}}
-	
+@Article{iml2018,
+  author = {Christoph Molnar and Bernd Bischl and Giuseppe Casalicchio},
+  title = {iml: An R package for Interpretable Machine Learning},
+  doi = {10.21105/joss.00786},
+  url = {https://joss.theoj.org/papers/10.21105/joss.00786},
+  year = {2018},
+  publisher = {Journal of Open Source Software},
+  volume = {3},
+  number = {26},
+  pages = {786},
+  journal = {JOSS},
+}
 
-@incollection{ke-2017-lightgbm,
-	Author = {Ke, Guolin and Meng, Qi and Finley, Thomas and Wang, Taifeng and Chen, Wei and Ma, Weidong and Ye, Qiwei and Liu, Tie-Yan},
-	Booktitle = {Advances in Neural Information Processing Systems 30},
-	Editor = {I. Guyon and U. V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett},
-	Pages = {3146--3154},
-	Publisher = {Curran Associates, Inc.},
-	Title = {LightGBM: A Highly Efficient Gradient Boosting Decision Tree},
-	Url = {http://papers.nips.cc/paper/6907-lightgbm-a-highly-efficient-gradient-boosting-decision-tree.pdf},
-	Year = {2017},
-	Bdsk-Url-1 = {http://papers.nips.cc/paper/6907-lightgbm-a-highly-efficient-gradient-boosting-decision-tree.pdf}}