Creates a list of advanced parameters for the task_hatsa function.
Usage
task_hatsa_opts(
lambda_blend_value = 0.15,
k_gev_dims = 10,
row_augmentation = TRUE,
residualize_condition_anchors = FALSE,
omega_weights = NULL,
omega_mode = c("fixed", "adaptive"),
reliability_scores_list = NULL,
scale_omega_trace = TRUE,
alpha_laplacian = 0.93,
degree_type_laplacian = c("abs", "positive", "signed"),
k_conn_pos = 10,
k_conn_neg = 10,
k_conn_task_pos = 10,
k_conn_task_neg = 10,
similarity_method_task = "pearson",
W_task_helper_func = NULL,
n_refine = 5,
check_redundancy = TRUE,
redundancy_threshold = 0.45,
residualize_k_conn_proj = 64,
residualize_k_conn_labels = 10,
gev_lambda_max = 0.8,
gev_epsilon_reg = 1e-06,
parcel_names = NULL
)Arguments
- lambda_blend_value
Numeric `lambda` in `[0,1]`. Weight for `L_task` in blend. Default 0.15.
- k_gev_dims
Integer, requested dimension for GEV patches. Default 10. Used if `task_method == "gev_patch"`.
- row_augmentation
Logical. If `TRUE`, add projected task features to anchor matrices for GPA refinement. Requires suitable `task_data_list`. Default `TRUE` if suitable data provided.
- residualize_condition_anchors
Logical. If `TRUE` and `row_augmentation` is `TRUE`, residualize projected task anchors against parcel anchors. Default `FALSE`.
- omega_weights
List specifying fixed weights for weighted Procrustes (e.g., `list(parcel = 1.0, condition = 0.5)`). Used if `row_augmentation=TRUE` and `omega_mode == "fixed"`. Defaults handled by `solve_procrustes_rotation_weighted`.
- omega_mode
Character string: `"fixed"` or `"adaptive"`. Controls weighting in GPA. Default `"fixed"`.
- reliability_scores_list
List (parallel to `subject_data_list`), each element a numeric vector of reliability scores (e.g., R^2) for task data (length `C`). Used if `omega_mode == "adaptive"`.
- scale_omega_trace
Logical. Whether to rescale weights in weighted GPA so trace equals total anchors. Default `TRUE`.
- alpha_laplacian
Numeric, laziness parameter for graph Laplacians (`L = I - alpha D^-1 W`). Default 0.93.
- degree_type_laplacian
Character string (`"abs"`, `"positive"`, `"signed"`). Type of degree calculation for Laplacian. Default `"abs"`.
- k_conn_pos
Integer >= 0. k-NN sparsification for positive edges in `W_conn`.
- k_conn_neg
Integer >= 0. k-NN sparsification for negative edges in `W_conn`.
- k_conn_task_pos
Integer >= 0. k-NN sparsification for positive edges in `W_task`.
- k_conn_task_neg
Integer >= 0. k-NN sparsification for negative edges in `W_task`.
- similarity_method_task
Character string or function. Method to compute similarity for `W_task` (e.g., "pearson", "spearman"). Default "pearson".
- W_task_helper_func
Function. The specific function to compute `W_task` (e.g., `compute_W_task_from_activations`, `compute_W_task_from_encoding`). If `NULL`, attempts to infer based on `task_data_list` structure (currently assumes activations `C x Vp`). Default `NULL`.
- n_refine
Integer >= 0. Number of GPA refinement iterations.
- check_redundancy
Logical. If `TRUE`, check correlation between `W_conn` and `W_task`. Default `TRUE`.
- redundancy_threshold
Numeric. Spearman rho threshold for triggering `W_task` residualization. Default 0.45.
- residualize_k_conn_proj
Integer. Number of `L_conn` eigenvectors to project `W_task` out of. Default 64.
- residualize_k_conn_labels
Integer. k-NN value for re-sparsifying `W_task_res` after residualization. Default 10.
- gev_lambda_max
Numeric. Max GEV eigenvalue `lambda` to retain for patches. Default 0.8.
- gev_epsilon_reg
Numeric. Small regularization for `L_conn` in GEV. Default 1e-6.
- parcel_names
Optional character vector of parcel names. If `NULL`, names like "P1", "P2"... are generated.