Builds the neighbour kernels and the random-effect design for continuous
intensities, then fits them with the identity-link engine. The Gaussian
counterpart of pace_fit_streaming(); see the notes at the top of this file
for what differs and why they are kept apart.
Usage
pace_fit_streaming_gaussian(
Y,
df,
types = NULL,
celltype_col,
image_col,
coord_cols = c("x", "y"),
h_bio = 30,
h_tech = 5,
eps = NULL,
condition_col = NULL,
kernel_per_image = FALSE,
image_re = c("none", "intercept", "slopes", "condition_slopes"),
drop_sparse_neff = 30,
within_image = TRUE,
n_iter = 32L,
threads = 4L,
chunk_size = 128L,
tau_shrinkage = "adaptive",
early_stop_tol = 0.02,
min_iter = 12L,
tau_max = 100,
verbose = TRUE
)Arguments
- Y, df, types, celltype_col, image_col, coord_cols
Intensities, the model frame, the cell types and the columns naming them.
- h_bio, h_tech, eps
Kernel bandwidths and the neighbour radius.
- condition_col, kernel_per_image, image_re, drop_sparse_neff, within_image
Design options, as
pace_fit_streaming()takes them.- n_iter, threads, chunk_size, tau_shrinkage, early_stop_tol, min_iter, tau_max
Fitting control, as
pace_fit_streaming()takes them.- verbose
Print the fitting trace.