Bootstrap: Bootstrapping of Distance Matrix

In stevexniu/FuseNet: Non-Parametric Network Fusion

Bootstrapping of Distance Matrix

Description

Bootstrapping of Distance Matrix

Usage

Bootstrap(
  data,
  dist_mat_null,
  k = 10,
  kernel = c("gaussian", "euclidean"),
  normalization = c("cosine", "lognorm", "none"),
  normalize_factor = 10000,
  pca_dims = 0,
  norm_type = c("l1", "l2"),
  n_iters = 100,
  ratio = 0.05,
  t = 0,
  calc_perturb_mat = FALSE,
  n_cores = NULL,
  zero_percent = 0.7,
  ...
)

Arguments

data	An M x d matrix or data.frame with M rows of data points and d columns of features.
dist_mat_null	An M x M distance matrix calculated from the original data (null).
k	Number of nearest neighbors. Default is 10. See details from nn2.
kernel	Kernel distance used: gaussian, gaussian distance kernel. See details EuclideanDist. euclidean, euclidean distance kernel. See details GaussianDist.
normalization	Normalization method used: #' cosine, cosine normalization. See details Normalization. lognorm, log normalization. See details Normalization. none, normalization is not performed.
normalize_factor	Normalize factor used in log normalization. Default is 10000. See details Normalization.
pca_dims	Number of dimensions used. Default is 0 and PCA is not performed.
norm_type	Type of norm used: l1, L1-like norm. See details L1Norm. l2, L1-like norm. See details L2Norm.
n_iters	Number of bootstrapping iterations. Default is 100.
ratio	Fraction of features to be downsampled in the original data matrix. Default is 0.05 aka 5%.
t	Matrix power used for the distance matrix. Default is 0 and powering is not performed. See MatrixPower for details.
calc_perturb_mat	Whether to calculate the perturb matrix. Default is FALSE.
n_cores	Number of cores used. Default is to use all existing cores. See details makeCluster.
zero_percent	Zero-entry percentage threshold. If the number of zeros in the returned matrices is above this number, a sparse matrix will be returned. Default is 0.7 aka 70%.
...	Additional parameters pass to makeCluster.

Value

Returns a list with entries:

• feature_weight, n x d binary matrix with n rows of bootstrap iterations and d columns of features where 0 means feature not sampled and 1 means sampled.

• sample_weight, n x M matrix with n rows of bootstrap iterations and M columns of data points where each entry represents weight.

• perturb_mat, d x M matrix with d rows of features and M columns of data points where each entry represents the relative importance of a feature to a data point.

• dist_mat, M x M distance matrix.

stevexniu/FuseNet documentation built on May 16, 2022, 12:23 p.m.