Shap values don't add up to prediction values

[prescient]

I'm not sure if I'm using this correctly, but I would have thought that the sum of the shap values should equal the prediction. In this case they don't seem to:

require(fastshap)
require(ranger)
require(tidyverse)

data(iris)

mdl <- ranger(Sepal.Length ~ . -Species,
              data = iris)
ranger_predict <- function(object, newdata){
  return(predict(object, newdata)$predictions)
}

rng_predictions <- predict(mdl, iris)$predictions

shaps <- explain(object = mdl, 
                 X = iris %>% select(-Species, - Sepal.Length), 
                 pred_wrapper = ranger_predict,
                 nsim = 1)
shap_predictions <- rowSums(shaps)
diffs <- rng_predictions - shap_predictions
plot(diffs)

[bgreenwell]

Hi @prescient sorry for the delay. The exact Shapley values for an instance should add up to the difference between the prediction for that instance and the average predictions across the entire training set. This is not the case, however, for approximate Shapley values, which is the default in fastshap. The sum should converge though as nsim goes to infinity which is briefly demonstrated here: https://bgreenwell.github.io/fastshap/articles/fastshap.html.

I've thought about adding an option to scale the explanations, but not completely sure of what the downside would be.

Does that help?

[bgreenwell]

Linking to this issue: slundberg/shap#998

[bgreenwell]

Just pushed an update (currently experimental). Now you can try setting adjust = TRUE as follows (note, however, that this requires nsim > 1):

require(ranger)
require(tidyverse)

data(iris)

mdl <- ranger(Sepal.Length ~ . -Species,
              data = iris)
ranger_predict <- function(object, newdata){
  return(predict(object, newdata)$predictions)
}

rng_predictions <- predict(mdl, iris)$predictions

shaps <- fastshap::explain(
  object = mdl, 
  X = iris %>% select(-Species, - Sepal.Length), 
  pred_wrapper = ranger_predict,
  nsim = 10,  # needs to be > 1
  adjust = TRUE
)
diffs <- rng_predictions - mean(rng_predictions)
plot(rowSums(shaps), diffs); abline(0, 1)

[prescient]

Hey Ben. I appreciate the response. I didn't know if I was just using the package wrong or not. Turns out this was the expected behavior. I'll test out the adjust option. Im often working with stakeholders in the business and things have to add up =]