A quick PMF estimation based on alternating Poisson regressions with tree-structured priors

Usage

asap_fit_pmf_larch(
  Y_,
  max_depth,
  max_iter = 100L,
  verbose = TRUE,
  a0 = 1,
  b0 = 1,
  do_log1p = FALSE,
  rseed = 1337L,
  svd_init = FALSE,
  do_degree_correction = FALSE,
  normalize_cols = FALSE,
  EPS = 1e-08,
  jitter = 1,
  NUM_THREADS = 0L
)

Arguments

Y_: non-negative data matrix (gene x sample)
max_depth: maximum depth of a perfect binary tree
max_iter: max number of optimization steps
verbose: verbosity
a0: gamma(a0, b0) default: a0 = 1
b0: gamma(a0, b0) default: b0 = 1
do_log1p: do log(1+y) transformation
rseed: random seed (default: 1337)
svd_init: initialize by SVD (default: FALSE)
normalize_cols: normalize columns by col_norm (default: FALSE)
EPS: (default: 1e-8)
jitter: (default: 1.0)

Value

a list that contains:

log.likelihood log-likelihood trace
theta loading (sample x factor)
log.theta log-loading (sample x factor)
log.theta.sd sd(log-loading) (sample x factor)
beta dictionary (gene x factor)
log.beta log dictionary (gene x factor)
log.beta.sd sd(log-dictionary) (gene x factor)