shapr

Brief NEWS

This is shapr version 1.0.0 (Released on GitHub Nov 2024), which provides a full restructuring of the code based, and provides a full suit of new functionality, including:

• A long list of approaches for estimating the contribution/value function $v(S)$, including Variational Autoencoders, and regression-based methods
• Iterative Shapley value estimation with convergence detection
• Parallelized computations with progress updates
• Reweighted Kernel SHAP for faster convergence
• New function explain_forecast() for explaining forecasts
• Several other methodological, computational and user-experience improvements
• Python wrapper making the core functionality of shapr available in Python

Below we provide a brief overview of the breaking changes. See the NEWS for the full list of details.

Breaking changes

The new syntax for explaining models essentially amounts to using a single function (explain()) instead of two functions (shapr() and explain()). In addition, custom models are now explained by passing the prediction function directly to explain(), some input arguments got new names, and a few functions for edge cases was removed to simplify the code base.

Note that the CRAN version of shapr (v0.2.2) still uses the old syntax. The examples below uses the new syntax. Here is a version of this README with the syntax of the CRAN version (v0.2.2).

Python wrapper

We now also provide a Python wrapper (shaprpy) which allows explaining python models with the methodology implemented in shapr, directly from Python. The wrapper is available here.

The package

The shapr R package implements an enhanced version of the Kernel SHAP method, for approximating Shapley values, with a strong focus on conditional Shapley values. The core idea is to remain completely model-agnostic while offering a variety of methods for estimating contribution functions, enabling accurate computation of conditional Shapley values across different feature types, dependencies, and distributions. The package also includes evaluation metrics to compare various approaches. With features like parallelized computations, convergence detection, progress updates, and extensive plotting options, shapr is as a highly efficient and user-friendly tool, delivering precise estimates of conditional Shapley values, which are critical for understanding how features truly contribute to predictions.

A basic example is provided below. Otherwise we refer to the pkgdown website and the vignettes there
for details and further examples.

Installation

We highly recommend to install the development version of shapr (with the new explanation syntax and all functionality),

remotes::install_github("NorskRegnesentral/shapr")

To also install all dependencies, use

remotes::install_github("NorskRegnesentral/shapr", dependencies = TRUE)

The CRAN version of shapr (NOT RECOMMENDED) can be installed with

install.packages("shapr")

Example

shapr supports computation of Shapley values with any predictive model which takes a set of numeric features and produces a numeric outcome.

The following example shows how a simple xgboost model is trained using the airquality dataset, and how shapr explains the individual predictions.

We first enable parallel computation and progress updates with the following code chunk. These are optional, but recommended for improved performance and user friendliness, particularly for problems with many features.

# Enable parallel computation
# Requires the future and future_lapply packages
future::plan("multisession", workers = 2) # Increase the number of workers for increased performance with many features

# Enable progress updates of the v(S)-computations
# Requires the progressr package
progressr::handlers(global = TRUE)
progressr::handlers("cli") # Using the cli package as backend (recommended for the estimates of the remaining time)

Here comes the actual example

library(xgboost)
library(shapr)

data("airquality")
data <- data.table::as.data.table(airquality)
data <- data[complete.cases(data), ]

x_var <- c("Solar.R", "Wind", "Temp", "Month")
y_var <- "Ozone"

ind_x_explain <- 1:6
x_train <- data[-ind_x_explain, ..x_var]
y_train <- data[-ind_x_explain, get(y_var)]
x_explain <- data[ind_x_explain, ..x_var]

# Looking at the dependence between the features
cor(x_train)
#>            Solar.R       Wind       Temp      Month
#> Solar.R  1.0000000 -0.1243826  0.3333554 -0.0710397
#> Wind    -0.1243826  1.0000000 -0.5152133 -0.2013740
#> Temp     0.3333554 -0.5152133  1.0000000  0.3400084
#> Month   -0.0710397 -0.2013740  0.3400084  1.0000000

# Fitting a basic xgboost model to the training data
model <- xgboost(
  data = as.matrix(x_train),
  label = y_train,
  nround = 20,
  verbose = FALSE
)

# Specifying the phi_0, i.e. the expected prediction without any features
p0 <- mean(y_train)

# Computing the actual Shapley values with kernelSHAP accounting for feature dependence using
# the empirical (conditional) distribution approach with bandwidth parameter sigma = 0.1 (default)
explanation <- explain(
  model = model,
  x_explain = x_explain,
  x_train = x_train,
  approach = "empirical",
  phi0 = p0
)
#> Note: Feature classes extracted from the model contains NA.
#> Assuming feature classes from the data are correct.
#> Success with message:
#> max_n_coalitions is NULL or larger than or 2^n_features = 16, 
#> and is therefore set to 2^n_features = 16.
#> 
#> ── Starting `shapr::explain()` at 2024-11-20 12:23:18 ──────────────────────────
#> • Model class: <xgb.Booster>
#> • Approach: empirical
#> • Iterative estimation: FALSE
#> • Number of feature-wise Shapley values: 4
#> • Number of observations to explain: 6
#> • Computations (temporary) saved at:
#> '/tmp/Rtmp4yBCHY/shapr_obj_17459f7fdc4b8f.rds'
#> 
#> ── Main computation started ──
#> 
#> ℹ Using 16 of 16 coalitions.

# Printing the Shapley values for the test data.
# For more information about the interpretation of the values in the table, see ?shapr::explain.
print(explanation$shapley_values_est)
#>    explain_id     none    Solar.R      Wind      Temp      Month
#>         <int>    <num>      <num>     <num>     <num>      <num>
#> 1:          1 43.08571 13.2117337  4.785645 -25.57222  -5.599230
#> 2:          2 43.08571 -9.9727747  5.830694 -11.03873  -7.829954
#> 3:          3 43.08571 -2.2916185 -7.053393 -10.15035  -4.452481
#> 4:          4 43.08571  3.3254595 -3.240879 -10.22492  -6.663488
#> 5:          5 43.08571  4.3039571 -2.627764 -14.15166 -12.266855
#> 6:          6 43.08571  0.4786417 -5.248686 -12.55344  -6.645738

# Finally we plot the resulting explanations
plot(explanation)

See the vignette for further basic usage examples.

Contribution

All feedback and suggestions are very welcome. Details on how to contribute can be found here. If you have any questions or comments, feel free to open an issue here.

Please note that the ‘shapr’ project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

References