Ideas for refining the structure of FFTrees objects

[ndphillips]

@hneth In the context of #224 I realized that the structure of FFTrees objects is making development challenging - such as knowing how to modularize the plot.FFTrees() function.

The current code to create FFTrees objects is contained here:

FFTrees/R/fftrees_create.R

Lines 765 to 852 in 71b9da0

	# Create x (as list):
	x <- list(
	# Names of criterion vs. cues:
	criterion_name = criterion_name,
	cue_names = cue_names,
	# Formula:
	formula = formula, # original formula
	# Tree info:
	trees = list(
	n = NULL,
	best = NULL,
	definitions = NULL,
	inwords = NULL,
	stats = NULL,
	level_stats = NULL,
	decisions = list(
	train = list(),
	test = list()
	)
	),
	# Raw training data:
	data = list(
	train = data,
	test = data.test
	),
	# Store parameters (as list):
	params = list(
	algorithm = algorithm,
	#
	goal = goal,
	goal.chase = goal.chase,
	goal.threshold = goal.threshold,
	#
	max.levels = max.levels,
	numthresh.method = numthresh.method,
	numthresh.n = numthresh.n,
	repeat.cues = repeat.cues,
	stopping.rule = stopping.rule,
	stopping.par = stopping.par,
	#
	sens.w = sens.w,
	#
	cost.outcomes = cost.outcomes,
	cost.cues = cost.cues,
	#
	main = main,
	decision.labels = decision.labels,
	#
	my.goal = my.goal,
	my.goal.fun = my.goal.fun,
	my.tree = my.tree,
	#
	quiet = quiet
	),
	# One row per algorithm competition:
	competition = list(
	train = data.frame(
	algorithm = NA,
	n = NA,
	hi = NA, fa = NA, mi = NA, cr = NA,
	sens = NA, spec = NA, far = NA,
	ppv = NA, npv = NA,
	acc = NA, bacc = NA,
	cost = NA, cost_dec = NA, cost_cue = NA
	),
	test = data.frame(
	algorithm = NA,
	n = NA,
	hi = NA, fa = NA, mi = NA, cr = NA,
	sens = NA, spec = NA, far = NA,
	ppv = NA, npv = NA,
	acc = NA, bacc = NA,
	cost = NA, cost_dec = NA, cost_cue = NA
	),
	models = list(lr = NULL, cart = NULL, rf = NULL, svm = NULL)
	) # competition.
	) # x.

Here are the core issues I see:

• Inconsistent and confusing naming
  • Ex) How do the definitions, inwords, stats, level_stats, and decisions objects in FFTrees.relate to each other?
• Information at different levels of abstraction aren't consistently stored
  • Ex) Why are tree definitions stored at the tree level but not the node level? Why aren't overall tree accuracy stats located close to the tree level accuracy stats?
• Inconsistent storage locations
  • Ex) Why are criterion_name, cue_names and formula stored at the same level as trees, data and params? Could these be stored in a list such as metadata?

To solve these issues, I'm drafted an object design doc at https://github.com/ndphillips/FFTrees/wiki/%5B80%25%5D-FFTrees-Object-Design. I'm eager for feedback

[hneth]

Just realized that my comment on #224 should better have been posted here — sorry!

[ndphillips]

No worries, let's carry on the discussion here