Conditioner

library("ggplot2")

customCoders = list('c.PiecewiseV.num' = vtreat::solve_piecewise,
                    'n.PiecewiseV.num' = vtreat::solve_piecewise,
                    'c.knearest.num' = vtreat::square_window,
                    'n.knearest.num' = vtreat::square_window)
codeRestriction = c("PiecewiseV", 
                    "knearest",
                    "poolN", "poolC",
                    "NonDecreasingV", "NonIncreasingV",
                    "clean", "isBAD", "catB", "catP")

d <- data.frame(x_numeric = seq(0, 15, by = 0.01))
d$x_cat <- paste0("l_", round(d$x_numeric, digits = 1))
d$y_ideal <- sin(d$x_numeric)
d$x_numeric_noise <- d$x_numeric[sample.int(nrow(d), nrow(d), replace = FALSE)]
d$x_cat_noise <- d$x_cat[sample.int(nrow(d), nrow(d), replace = FALSE)]
d$y <- d$y_ideal + 0.5*rnorm(nrow(d))
d$yc <- d$y>0.5
d$is_train <- runif(nrow(d))>=0.2

dcheck <- d[1:2, ]
dcheck$x_numeric <- NA_real_
dcheck$x_cat[1] <- "new_level"
dcheck$x_cat[2] <- NA_character_
dcheck

head(d)

summary(d)

ggplot(data=d) +
  geom_point(aes(x = x_numeric, y = y, color = yc), alpha=0.5) + 
  geom_line(aes(x = x_numeric, y = y_ideal), color = "lightblue") +
  geom_hline(yintercept = 0.5, color = "red")

cfn <- vtreat::mkCrossFrameNExperiment(
  d[d$is_train, , drop=FALSE], 
  c('x_numeric', 'x_numeric_noise', 'x_cat', 'x_cat_noise'), 'y',
  customCoders = customCoders,
  codeRestriction = codeRestriction,
  verbose = FALSE)

cfn$treatments

vtreat::variable_values(cfn$treatments$scoreFrame)

# or directly
vtreat::value_variables_N(
  d[d$is_train, , drop=FALSE], 
  c('x_numeric', 'x_numeric_noise', 'x_cat', 'x_cat_noise'), 'y')

prepared <- vtreat::prepare(cfn$treatments, d)
d$x_numeric_PiecewiseV <- prepared$x_numeric_PiecewiseV
d$x_numeric_knearest <- prepared$x_numeric_knearest

ggplot(data=d) +
  # geom_point(aes(x = x_numeric, y = y)) + 
  geom_line(aes(x = x_numeric, y = y_ideal), color = "lightblue") + 
  geom_line(aes(x = x_numeric, y = x_numeric_PiecewiseV)) +
  ggtitle("y_ideal as a function of x_numeric_PiecewiseV")

ggplot(data=d) +
  # geom_point(aes(x = x_numeric, y = y)) + 
  geom_line(aes(x = x_numeric, y = y_ideal), color = "lightblue") + 
  geom_line(aes(x = x_numeric, y = x_numeric_knearest)) +
  ggtitle("y_ideal as a function of x_numeric_knearest")

WVPlots::ScatterHist(d[d$is_train, , drop=FALSE], 
                     "x_numeric_PiecewiseV", "y",
                     "x_numeric_PiecewiseV versus observed y on train",
                     smoothmethod = "identity",
                     estimate_sig = TRUE)

WVPlots::ScatterHist(d[d$is_train, , drop=FALSE], 
                     "x_numeric_PiecewiseV", "y_ideal",
                     "x_numeric_PiecewiseV versus ideal y on train",
                     smoothmethod = "identity",
                     estimate_sig = TRUE)

WVPlots::ScatterHist(d[!d$is_train, , drop=FALSE], 
                     "x_numeric_PiecewiseV", "y",
                     "x_numeric_PiecewiseV versus observed y on test",
                     smoothmethod = "identity",
                     estimate_sig = TRUE)

WVPlots::ScatterHist(d[!d$is_train, , drop=FALSE], 
                     "x_numeric_PiecewiseV", "y_ideal",
                     "x_numeric_PiecewiseV versus ideal y on test",
                     smoothmethod = "identity",
                     estimate_sig = TRUE)

vtreat::prepare(cfn$treatments, dcheck)

cfc <- vtreat::mkCrossFrameCExperiment(
  d[d$is_train, , drop=FALSE], 
  c('x_numeric', 'x_numeric_noise', 'x_cat', 'x_cat_noise'), 'yc', TRUE,
  customCoders = customCoders,
  codeRestriction = codeRestriction,
  verbose = FALSE)

cfc$treatments

vtreat::variable_values(cfc$treatments$scoreFrame)

# or directly
vtreat::value_variables_C(
  d[d$is_train, , drop=FALSE], 
  c('x_numeric', 'x_numeric_noise', 'x_cat', 'x_cat_noise'), 'yc', TRUE)


vtreat::prepare(cfc$treatments, dcheck)