Skip to contents

fit_ml() supports both historical idiographic calls using outcome/predictors and the consolidated panel-model API using y/x. Calls that name y or x, or that pass a consolidated control such as time, scope, subgroup or tune, use the consolidated result contract; existing positional and outcome/predictors calls retain their original behaviour.

Usage

fit_idiographic_ml(
  data,
  outcome,
  predictors,
  id,
  day = NULL,
  beep = NULL,
  task = c("auto", "regression", "classification"),
  model = NULL,
  estimator = NULL,
  compare = c("both", "individual", "pooled"),
  test_prop = 0.2,
  min_train = 10L,
  min_test = 1L,
  lambda = 1,
  alpha = 0.5,
  k = 5L,
  n_components = NULL,
  max_iter = 100L,
  tol = 1e-06,
  standardize = TRUE,
  keep_fits = FALSE,
  ...
)

fit_individualized_ml(
  data,
  outcome,
  predictors,
  id,
  day = NULL,
  beep = NULL,
  task = c("auto", "regression", "classification"),
  model = NULL,
  estimator = NULL,
  compare = c("both", "individual", "pooled"),
  test_prop = 0.2,
  min_train = 10L,
  min_test = 1L,
  lambda = 1,
  alpha = 0.5,
  k = 5L,
  n_components = NULL,
  max_iter = 100L,
  tol = 1e-06,
  standardize = TRUE,
  keep_fits = FALSE,
  ...
)

fit_ml(
  data,
  outcome,
  predictors,
  id,
  day = NULL,
  beep = NULL,
  task = c("auto", "regression", "classification"),
  model = NULL,
  estimator = NULL,
  compare = c("both", "individual", "pooled"),
  test_prop = 0.2,
  min_train = 10L,
  min_test = 1L,
  lambda = 1,
  alpha = 0.5,
  k = 5L,
  n_components = NULL,
  max_iter = 100L,
  tol = 1e-06,
  standardize = TRUE,
  keep_fits = FALSE,
  ...,
  y,
  x
)

Arguments

data

A data frame or matrix.

outcome, predictors

Historical idiographic outcome and predictor arguments.

id

Person identifier column.

day, beep

Optional historical ordering columns.

task

Outcome task: "auto", "regression", or "classification".

model

One or more model names.

estimator

Optional implementation backend.

compare

Historical scope selector: "both", "individual", or "pooled".

test_prop

Proportion of each person's ordered rows used for testing.

min_train, min_test

Minimum training and test rows.

lambda, alpha

Penalized-model controls.

k

Number of neighbours for nearest-neighbour models.

n_components

Number of principal components for the historical API.

max_iter, tol

Iteration limit and convergence tolerance.

standardize

Use training-only predictor standardization?

keep_fits

Retain fitted backend objects?

...

Arguments passed to the selected implementation; in the consolidated interface, the controls listed in Details.

y, x

Consolidated outcome and predictor selectors. Supply both.

Value

For consolidated calls, an idiographic_fit (also of class idiostats_fit). Its tables are read with accessors: metrics() (one row per scope, model, estimator and person, plus .overall rows, with rmse, mae, bias and r_squared for regression), predictions() (one row per held-out occasion and model), coefs(), importance(), tuning() and failures().

For historical calls, an idioml_result. as.data.frame() returns its held-out predictions, summary() its metrics (one row per model_scope, model, estimator and person), coefs() its coefficients (one row per model_scope, model, person and term) and predict() predictions for new rows.

Details

Both interfaces evaluate prediction on an ordered hold-out: the last ceiling(test_prop * n) rows of each person are the test set and are never used for estimation. In the consolidated interface, tune = TRUE carves a further validation block (valid_prop) from the tail of the training rows, scores the candidate settings on it, and refits the winning setting on the training and validation rows before predicting the test rows.

Consolidated controls passed through ... are time (ordering column), scope ("pooled", "subgroup", "individual", "both" or "all"), subgroup (as in fit_lm()), valid_prop, ncomp (pcr), mtry and num_trees (forest), cost (svm), tune and grid (a named list of tuning grids, e.g. list(knn = list(k = c(3, 5)))). models() lists every model name with its estimator, package and tuned parameter. Native models need only base R. When an optional backend package is missing, an interactive session offers to install it after confirmation; otherwise the call stops with an error that names the package.

Examples

# Consolidated interface: pooled and person-specific models, tidy results.
fit <- fit_ml(srl, y = "effort", x = "efficacy:monitoring", id = "name",
              time = "day", scope = "both", model = c("mean", "ridge"))
metrics(fit, overall = TRUE)
#>        scope model estimator  subject subgroup    n     rmse      mae      bias
#> 1     pooled  mean    native .overall     .all 1150 26.88010 22.50530 0.9244965
#> 2 individual  mean    native .overall     .all 1150 22.62189 17.84407 0.9032171
#> 3     pooled ridge    native .overall     .all 1150 20.18719 16.28609 0.7323493
#> 4 individual ridge    native .overall     .all 1150 17.22461 12.86903 0.7960435
#>      r_squared
#> 1 -0.001184303
#> 2  0.290895607
#> 3  0.435318032
#> 4  0.588896681
best_model(fit)
#>        scope model estimator  subject subgroup    n     rmse      mae      bias
#> 4 individual ridge    native .overall     .all 1150 17.22461 12.86903 0.7960435
#>   r_squared
#> 4 0.5888967

# Tune the number of neighbours on a validation block before testing.
tuned <- fit_ml(srl, y = "effort", x = "efficacy:monitoring", id = "name",
                time = "day", scope = "pooled", model = "knn", tune = TRUE)
tuning(tuned)
#>    scope model estimator  subject subgroup parameter value   n     rmse
#> 1 pooled   knn    native .overall     .all         k     1 899 13.58116
#> 2 pooled   knn    native .overall     .all         k     3 899 16.51403
#> 3 pooled   knn    native .overall     .all         k     5 899 18.83511
#> 4 pooled   knn    native .overall     .all         k     7 899 19.82245
#> 5 pooled   knn    native .overall     .all         k    15 899 20.26371
#> 6 pooled   knn    native .overall     .all         k     9 899 20.26697
#>         mae accuracy brier rank
#> 1  5.140692       NA    NA    1
#> 2 10.832807       NA    NA    2
#> 3 14.141548       NA    NA    3
#> 4 15.301335       NA    NA    4
#> 5 15.803694       NA    NA    5
#> 6 15.785908       NA    NA    6

# Historical interface (outcome, predictors, id).
legacy <- fit_ml(srl, "effort", c("efficacy", "planning"), "name",
                 model = "ridge")
legacy
#> Idiographic Machine-Learning Result
#>   Task:          regression
#>   Algorithms:    1 (ridge)
#>   Outcome:       effort
#>   Predictors:    2 (efficacy, planning)
#>   Subjects:      36
#>   Predictions:   2300
#>   individual/ridge[native] RMSE:   17.8653
#>   pooled/ridge[native] RMSE:   20.8727
#>   Tables:        as.data.frame(x) | summary(x) | coefs(x)