proto_basis1d_distribution {cheem}

Adds the distribution of the row local attributions to a ggtour

Description

A spinifex proto_*-like function, that adds the distribution of orthonormalized row values of the specified local explanation attr_df. Does not draw the basis bars; use in conjunction with proto_basis1d().

Usage

proto_basis1d_distribution(
  attr_df,
  primary_obs = NULL,
  comparison_obs = NULL,
  position = c("top1d", "floor1d", "bottom1d", "off"),
  group_by = as.factor(FALSE),
  pcp_shape = c(3, 142, 124),
  do_add_pcp_segments = TRUE,
  inc_var_nms = NULL,
  row_index = NULL
)

Arguments

attr_df

An data frame, the attributions of a local explanation.

primary_obs

The rownumber of the primary observation. Its local attribution becomes the 1d projection basis, and the point it highlighted as a dashed line. Defaults to NULL, no highlighting.

comparison_obs

The rownumber of the comparison observation. Point is highlighted as a dotted line. Defaults to NULL, no highlighting.

position

The position for the basis, one of: c("top1d", "floor1d", "bottom1d", "off"). Defaults to "top1d"; basis above the density curves.

group_by

Vector to group densities by. Originally predicted class.

pcp_shape

The number of the shape character to add. Expects 3, 142, or 124, '+', '|' in plotly, and '|' in gganimate, respectively. Defaults to 3, '+' in either output.

do_add_pcp_segments

Logical, whether or not to add to add faint parallel coordinate lines on the 1D basis. Defaults to TRUE.

inc_var_nms

A character vector, the names of the variables to keep. Defaults to NULL, all variables kept.

row_index

A numeric or logical vector, the index of the rows to keep. Defaults to NULL, all rows kept.

Value

A ggplot object of the the distribution of the local explanation's attributions.

See Also

radial_cheem_tour()

Examples

library(cheem)
library(spinifex)

## Regression setup
dat  <- amesHousing2018_NorthAmes
X    <- dat[, 1:9]
Y    <- dat$SalePrice
clas <- dat$SubclassMS

## Basis, manipulation var, manual tour path, & predictions to fix to y-axis
bas     <- sug_basis(ames_rf_shap, 1)
mv      <- sug_manip_var(ames_rf_shap, 1, 2)
mt_path <- manual_tour(bas, mv)
## Also consumed by: ?radial_cheem_tour()

## Compose and animate the tour
ggt <- ggtour(mt_path, scale_sd(X), angle = .3) +
  append_fixed_y(fixed_y = scale_sd(Y)) +
  proto_point(list(color = clas, shape = clas)) +
  proto_basis1d_distribution(
    attr_df = ames_rf_shap,
    primary_obs = 1, comparison_obs = 2,
    position = "top1d", group_by = clas) +
  proto_basis1d(position = "bottom1d") +
  proto_origin()

if(interactive()){
  animate_plotly(ggt)
}