Model Targets

Retrieve the model-specific targets of a design together with their meaning. Cross-validation folds are built from cv_labels; the estimator sees the targets returned here. The two are identical for ordinary classification designs but differ when a design carries matrix-valued or continuous targets (feature prediction, feature sets, feature RSA).

Usage

model_targets(design, partition = c("train", "test"), ...)

# S3 method for class 'mvpa_design'
model_targets(design, partition = c("train", "test"), ...)

# S3 method for class 'feature_rsa_design'
model_targets(design, partition = c("train", "test"), ...)

# S3 method for class 'feature_sets_design'
model_targets(design, partition = c("train", "test"), ...)

Arguments

design

A design object such as mvpa_design, feature_sets_design, or feature_rsa_design.

partition

Either "train" or "test". Requesting the test partition of a design without test targets returns NULL.

...

Additional arguments passed to methods.

Value

A model_targets list as described above, or NULL when the requested partition has no targets.

Details

model_targets completes the cv_labels / targets split introduced in mvpa_design. It never changes what y_train returns: y_train() keeps returning the cross-validation labels for backward compatibility.

The returned object is a list of class model_targets with:

values

A factor, numeric vector, or numeric matrix with one row (or element) per observation of the requested partition.

observation_ids

Integer row identifiers, aligned with values.

response_ids

Character identifiers for the responses: factor levels, matrix column names, or NULL for an unnamed scalar response.

response_groups

Optional grouping of responses (for example the feature-set membership of a feature_sets_design), or NULL.

row_weights

Optional numeric observation weights, or NULL.

type

One of "categorical", "continuous", or "matrix".

partition

The requested partition.

See also

mvpa_design, y_train

Examples

des <- mvpa_design(data.frame(cond = rep(c("a", "b"), 10)), y_train = ~ cond)
model_targets(des)$type
#> [1] "categorical"

feats <- matrix(rnorm(20 * 3), 20, 3, dimnames = list(NULL, c("f1", "f2", "f3")))
des2 <- mvpa_design(data.frame(id = 1:20), cv_labels = 1:20, targets = feats)
dim(model_targets(des2)$values)
#> [1] 20  3