R中XGBoost模型的SHAP汇总图，图上不显示Mean Absolute SHAP值

Question

我不想在r中显示SHAP摘要图上的平均绝对值，我想要类似于用python生成的输出。哪一行代码将有助于从R中的汇总图中删除平均值绝对值?

我现在正在使用这行代码:

shap.plot.summary.wrap1(xgb_model, X = x, top_n = 10)

Answer 1

您可以稍微修改shape .plot.summary()的源代码，如下所示:

shap.plot.summary.edited <- function(data_long,
x_bound = NULL,
dilute = FALSE,
scientific = FALSE,
my_format = NULL){
if (scientific){label_format = "%.1e"} else {label_format = "%.3f"}

if (!is.null(my_format)) label_format <- my_format
# check number of observations
N_features <- setDT(data_long)[,uniqueN(variable)]
if (is.null(dilute)) dilute = FALSE
nrow_X <- nrow(data_long)/N_features # n per feature
if (dilute!=0){
# if nrow_X <= 10, no dilute happens
dilute <- ceiling(min(nrow_X/10, abs(as.numeric(dilute)))) # not allowed to dilute to fewer than 10 obs/feature
set.seed(1234)
data_long <- data_long[sample(nrow(data_long),
min(nrow(data_long)/dilute, nrow(data_long)/2))] # dilute
}
x_bound <- if (is.null(x_bound)) max(abs(data_long$value))*1.1 else as.numeric(abs(x_bound))
plot1 <- ggplot(data = data_long) +
coord_flip(ylim = c(-x_bound, x_bound)) +
geom_hline(yintercept = 0) + # the y-axis beneath
# sina plot:
ggforce::geom_sina(aes(x = variable, y = value, color = stdfvalue),
method = "counts", maxwidth = 0.7, alpha = 0.7) +
# print the mean absolute value:
#geom_text(data = unique(data_long[, c("variable", "mean_value")]),
#          aes(x = variable, y=-Inf, label = sprintf(label_format, mean_value)),
#          size = 3, alpha = 0.7,
#          hjust = -0.2,
#          fontface = "bold") + # bold
# # add a "SHAP" bar notation
# annotate("text", x = -Inf, y = -Inf, vjust = -0.2, hjust = 0, size = 3,
#          label = expression(group("|", bar(SHAP), "|"))) +
scale_color_gradient(low="#FFCC33", high="#6600CC",
breaks=c(0,1), labels=c(" Low","High "),
guide = guide_colorbar(barwidth = 12, barheight = 0.3)) +
theme_bw() +
theme(axis.line.y = element_blank(),
axis.ticks.y = element_blank(), # remove axis line
legend.position="bottom",
legend.title=element_text(size=10),
legend.text=element_text(size=8),
axis.title.x= element_text(size = 10)) +
# reverse the order of features, from high to low
# also relabel the feature using `label.feature`
scale_x_discrete(limits = rev(levels(data_long$variable))#,
#labels = label.feature(rev(levels(data_long$variable)))
)+
labs(y = "SHAP value (impact on model output)", x = "", color = "Feature value  ")
return(plot1)
}