################################################################################
# 
#
#             Teaching with loon
#             
#
################################################################################
#  
#  
#  
#  There are several ways that the looniverse can help with teaching.
#  
#  E.g. loon.data has a few artificial data sets which illustrate
#       statistical concepts.
#       see  http://great-northern-diver.github.io/loon.data/reference/
#       Or
# https://great-northern-diver.github.io/loon.data/reference/index.html#section-teaching
#       They include
#         - pandemic ... a cyclical preference of treatments
#         - medicalRecords ... double reversing Simpson's paradox
#         - lizards ... ecological fallacy like Simpson's
#         - blocks ... sampling methods
#         - judgment ... sampling methods
#         
#  
#    - produces high-quality presentation graphics
#    - uses a pipeline style to build visualizations
#    - based on a variation on the original grammar of graphics (hence the gg)
#    - part of the tidyverse (with magrittr, dplyr, purrr, ...)
#
#
# There are also a few demos() whose code could be adapted.
# 
library(loon)
library(loon.data)
# to see all of the demos available
demo(package = "loon")

# Box-Cox power transformations;  
# A must use for the data analysis classroom
demo("l_power")

# Add polynomial regressions to selected points
demo("l_add_regressions")

# Explore the effect of influential points seen by
# actually moving them around in the scatterplot.
# Great for illustrating breakdown 
demo("l_regression_influential")

# Hans Rosling's famous gapminder demonstration.
# Now the modern statistical equivalent of "Hello world" 
# statistical visualization
demo("l_us_and_them_slider")

# Some new kinds of linking
# Many to one (polygons in map)
demo("l_us_and_them_choropleth")

# Comparing/using many dimension reduction methods
# at once in exploration
demo("l_ng_dimred")


########
#
# The code from the above demos show you how to build them.
# (the code prints out when you run the demo)
# 
# But, it really isn't too hard to build something straightforward.
# 
# Because `loon` is built on the base `R` `tcltk` package (which ships with `R`), 
# you can always build your own interactive displays 
# using a mix of `loon` and `tcltk` functionality.
#
# For example, a simple slider is built as

slider_window <- tktoplevel()

tktitle(slider_window) <- "slider"

slider_val <- tclVar('1') # default value

slider <- tkscale(parent = slider_window, 
                  orient = "horizontal",
                  variable = slider_val,  
                  from = -5, 
                  to = 5, 
                  resolution = 0.1,
                  command = function(...) print(paste0("Slider value = ", ...)))

tkgrid(slider, row = 0, column = 0, sticky="we", padx = 50)


# Now move the slider.
# 
#
#
########
#
#  An important concept is that of binding events
#  
#  In a loon plot, it is EASY to bind a function to fire 
#  whenever named states are changed.
#
#  The main function here is `l_bind_state(target, event, callback)`
#
#  For example

myPlot <- l_plot(iris)

makeItHappen <- function() {print ("It happened!")}

l_bind_state(myPlot, c("color", "xTemp", "yTemp"), makeItHappen)

#  Now change the colour of some points.
#  Move some points.
#  It is that simple to extend behaviour
#  
#  We could attach different functions to fire for different state changes
#  
WhoIsSpecial <- function() {
    cat("Selected points were: \n\t", 
        myPlot["itemLabel"][myPlot["selected"]], "\n"
    )
}

# Loon now allow you to bind myPlot so that
# when the value of any of its "selected" state
# values change, the function WhoIsSpecial
# is called (with no argumenta)

l_bind_state(myPlot, c("selected"), WhoIsSpecial)

# 
# Which opens up a whole range of possibilities!!
# 
# 
# Note that having a global variable like myPlot
# appear within a function (as above) is generally
# a very bad idea.  
# This was to simplify the illustration.
# Generally, more care needs to be taken to ensure
# that myPlot can only be the plot you are interested in.
# 
# 
# A more realistic (and careful) example follows.
# 
# Suppose we fit a least-squares line to some data:
x <- longley$GNP
y <- longley$Unemployed
data <- data.frame(x = x, y = y)
p <- l_plot(data$x, data$y)


lm_fit <- lm( y ~ x, data = data)
x_line <- extendrange(x)
y_line <- predict(lm_fit, newdata = data.frame(x = x_line))
fitted_line <- l_layer_line(p, x = x_line, y = y_line, color = "red")

#
# And now would like to have the line change whenever points were moved.
# (This would change values of the states "xTemp" and/or "yTemp"
#  in the plot.)
# 
# Updating the least-squares line would be effected by a function like

updateLine <- function(myPlot, ls_line = NULL) {
    
    if (!is.null(ls_line) &  ls_line %in% l_layer_ids(myPlot)) {
        # we'll update it.
        xnew <- myPlot['xTemp']
        if (length(xnew) == 0) {xnew <- myPlot['x']}
        
        # For y
        ynew <- myPlot['yTemp']
        if (length(ynew) == 0) {ynew <- myPlot['y']}
        
        # New fit
        new_fit <- lm( y ~ x, data = data.frame(x = xnew, y = ynew))
        x_line <- extendrange(xnew)
        y_line <- predict(new_fit, newdata = data.frame(x = x_line))
        
        # configure the least-squares line
        l_configure(ls_line, x = x_line, y = y_line)
    }
    
    # Update the tcl language's event handler
    tcl('update', 'idletasks')
}

#  The final step is simply

l_bind_state(p, 
             c("xTemp", "yTemp"),  
             function() {updateLine(p, fitted_line)})
#
# Now go move some points!
# 
# 
# There is no end to the possibilities!!
# 
# 
################################################################################