three_prop_cv: Estimate 3prop weights using double cross-validation
In miguelbiron/threepRop: R Implementation of the 3prop Label Propagation Algorithm

Description Usage Arguments Details Value References Examples

This function estimates the weights for the 3prop algorithm using a double cross-validation procedure.

1 2	three_prop_cv(M, y, R = 3L, n_folds = 3L, reg = 1e-09, method = "LDA")

`M`	normalized affinity matrix, as returned by `normalize_A()`, of size `N x N`. Must be of class `sparseMatrix`.
`y`	vector of labels, of length `N`. Must be of class `sparseVector`.
`R`	maximum length of the random walks to consider. The default 3 is the usual for 3prop.
`n_folds`	number of CV folds to use in both inner and outer CV loops.
`reg`	regularization parameter for LDA.
`method`	string for method to compute the coefficients. Can be "LDA" (default) or "Ridge". Both use the parameter `reg`.

By double cross-validation (CV), we mean that there are two CV loops, one nested within the other. The first loop deals with calculating the random walk features, while the second loop deals with estimating the coefficients alpha associated to those features.

Additionally, the function returns computes the area under the ROC curve (AUROC) for every fold in the outer CV loop.

A list with two elements: a matrix of size R x n_folds, containing the weights estimated for each (outer) CV iteration, and a vector of length n_folds containing the AUROC for each such iteration.

Mostafavi, S., Goldenberg, A., & Morris, Q. (2012). Labeling nodes using three degrees of propagation. PloS one, 7(12), e51947.

sim_SBM = simulate_simple_SBM(N = 2500L, p_1 = 0.2, D = 0.04, R = 0.25)
M = normalize_A(sim_SBM$A, "asym")
three_prop_cv(M=M, y=sim_SBM$y)
three_prop_cv(M=M, y=sim_SBM$y, method = "Ridge")