Uses the DISCOURSE algorithmic framework to simulate multiple data sets that produce target ANOVA F-statistics under a specified factorial design given input parameters.
Usage
parallel_aov(
parallel_start = 3,
return_best_solution = FALSE,
N,
levels,
target_group_means,
target_f_list,
integer,
range,
formula,
factor_type,
subgroup_sizes = NULL,
df_effects = NULL,
tolerance = 1e-08,
typeSS = 3,
max_iter = 1000,
init_temp = 1,
cooling_rate = NULL,
max_step = 0.2,
max_starts = 1,
checkGrim = FALSE,
min_decimals = 1,
progress_mode = "console"
)Arguments
- parallel_start
Number of independent runs (parallel or sequential) to simulate data sets.
- return_best_solution
Logical; return only the best run if TRUE.
- N
Integer. Total number of subjects (sum of `subgroup_sizes`).
- levels
Integer vector. Number of factor levels per factor in the design.
- target_group_means
Numeric vector. Desired means for each group in the design.
- target_f_list
List with components:
- F
Numeric vector of target F-statistics.
- effect
Character vector of effect names matching `F`.
- contrast
Optional character formula for contrasts.
- contrast_method
Optional character specifying contrast method.
- integer
Logical. If TRUE, candidate values are treated as integers, if FALSE treated as continous values.
- range
Numeric vector of length 2. Lower and upper bounds for candidate means.
- formula
Formula or character. Model formula used to compute F-values (e.g., `y ~ A + B + A*B`).
- factor_type
Character vector. Type of each factor (`"between"` or `"within"`) matching length of `levels`.
- subgroup_sizes
Numeric vector. Optional sizes of each between-subjects group for unbalanced designs; length must equal product of `levels` for between factors.
- df_effects
Numeric vector. Degrees of freedom of the model effects. Default is `NULL`.
- tolerance
Numeric. Error tolerance for convergence; stops early if best error < `tolerance`. Default `1e-6`.
- typeSS
Integer. Type of sums-of-squares for ANOVA (2 or 3). Default is 3.
- max_iter
Integer. Maximum iterations per restart. Default is 1e3.
- init_temp
Numeric. Initial temperature for annealing. Default is 1.
- cooling_rate
Numeric. Cooling rate per iteration (between 0 and 1); if NULL, calculated automatically as `(init_temp-10)/init_temp`.
- max_step
Numeric. Maximum move size as proportion of `range`. Default is 0.2.
- max_starts
Integer. Number of annealing restarts. Default is 1.
- checkGrim
Logical. If TRUE and `integer = TRUE`, perform GRIM checks on `target_group_means`. Default is FALSE.
- min_decimals
Integer. Minimum number of decimal places for target values (including trailing zeros). Default `1`.
- progress_mode
Character. Either "console" or "shiny" for progress handler. Default `console`.
Value
A list of multiple `discourse.object`s, each containing:
- best_error
Numeric. Minimum error (RMSE) achieved.
- data
Data frame of optimized outcome values (and grouping variables).
- inputs
List of all input arguments.
- track_error
Numeric vector of best error at each iteration.
- grim
List of the GRIM results.
Examples
if (FALSE) { # \dontrun{
# Balanced 2x2 design
parallel_aov(
parallel_start = 7,
return_best_solution = FALSE,
N = 40,
levels = c(2, 2),
target_group_means = c(1, 2, 3, 4),
target_f_list = list(effect = c("A", "B"),
F = c(5.6, 8.3), ),
formula = y ~ A + B + A*B,
factor_type = c("between", "between"),
range = c(0, 5),
integer = FALSE,
max_iter = 1000,
max_starts = 3
)
} # }