Customizing labels in SHAPforxgboost plots

user_11598009 0
Viewed 3861

I'm creating some plots of SHAP-scores for visualizing a model I created with xgboost. I've used the SHAPforxgboost package which has worked very well, and I now want to use the figures (especially the one from shap.plot.summary()) in a text document I'm writing. However, the font sizes of the labels/titles on the x and y-axes are very small and I was wondering if there was a way I could make these larger and more readable.

I've used a very similar setup as shown here; https://www.rdocumentation.org/packages/SHAPforxgboost/versions/0.0.2 :

library("SHAPforxgboost")
y_var <-  "diffcwv"
dataX <- dataXY_df[,-..y_var]
# hyperparameter tuning results
param_dart <- list(objective = "reg:linear",  # For regression
                   nrounds = 366,
                   eta = 0.018,
                   max_depth = 10,
                   gamma = 0.009,
                   subsample = 0.98,
                   colsample_bytree = 0.86)

mod <- xgboost::xgboost(data = as.matrix(dataX), label = as.matrix(dataXY_df[[y_var]]), 
                       xgb_param = param_dart, nrounds = param_dart$nrounds,
                       verbose = FALSE, nthread = parallel::detectCores() - 2,
                       early_stopping_rounds = 8)

# To return the SHAP values and ranked features by mean|SHAP|
shap_values <- shap.values(xgb_model = mod, X_train = dataX)
# The ranked features by mean |SHAP|
shap_values$mean_shap_score

# To prepare the long-format data:
shap_long <- shap.prep(xgb_model = mod, X_train = dataX)
# is the same as: using given shap_contrib
shap_long <- shap.prep(shap_contrib = shap_values$shap_score, X_train = dataX)
# (Notice that there will be a data.table warning from `melt.data.table` due to `dayint` coerced from integer to double)

# **SHAP summary plot**
shap.plot.summary(shap_long)

The output of shap.plot.summary() is: something like this

More specifically, I would be interested in increasing the font size of each descriptor on the y-axis

shap xgboost machine-learning r
2 Answers

So as cbo posted a sufficient answer for most cases, I was not able to edit the size of the labels on the y-axis (ie. 0.629, 0.219, 0.029). I found the best solution for this was to use the function

shap.plot.summary <- edit(shap.plot.summary)

to edit the ggplot settings. For anyone curious, the ggplot settings I found to be relevant in the plot were:

theme(axis.line.y = element_blank(),
      axis.ticks.y = element_blank(), legend.position = "bottom",
      legend.title = element_text(size = 25),
      legend.text = element_text(size = 25), 
      axis.title.x = element_text(size = 25),
      axis.text.y = element_text(size = 40),
      axis.text.x.bottom = element_text(size  = 20))
user_115980090

Looking at the code here since it is made with ggplot you should be able to overwrite default label size argument.

Using the example of the shap.plot.summary.wrap2 function :

library("SHAPforxgboost")
library("ggplot2")

data("iris")
X1 = as.matrix(iris[,-5])
mod1 = xgboost::xgboost(
        data = X1, label = iris$Species, gamma = 0, eta = 1,
        lambda = 0,nrounds = 1, verbose = FALSE)


# shap.values(model, X_dataset) returns the SHAP
# data matrix and ranked features by mean|SHAP|
shap_values <- shap.values(xgb_model = mod1, X_train = X1)
shap_values$mean_shap_score
#> Petal.Length  Petal.Width Sepal.Length  Sepal.Width 
#>   0.62935975   0.21664035   0.02910357   0.00000000
shap_values_iris <- shap_values$shap_score

# shap.prep() returns the long-format SHAP data from either model or
shap_long_iris <- shap.prep(xgb_model = mod1, X_train = X1)
# is the same as: using given shap_contrib
shap_long_iris <- shap.prep(shap_contrib = shap_values_iris, X_train = X1)

# **SHAP summary plot**
# shap.plot.summary(shap_long_iris, scientific = TRUE)
# shap.plot.summary(shap_long_iris, x_bound  = 1.5, dilute = 10)

# Alternatives options to make the same plot:
# option 1: from the xgboost model
# shap.plot.summary.wrap1(mod1, X = as.matrix(iris[,-5]), top_n = 3)

# option 2: supply a self-made SHAP values dataset
# (e.g. sometimes as output from cross-validation)
shap.plot.summary.wrap2(shap_values_iris, X1, top_n = 3) +
        ggplot2::theme(axis.text.y = element_text(size = 20))

user_109834700
Related