BIPPS (Bayesian Inference for Point Patterns in Space) is a spatial statistical modeling package for analyzing point patterns or count data across a spatial domain. It uses latent Gaussian processes to model the underlying intensity of a point process, making it particularly useful for applications in:

• Ecological data analysis
• Disease mapping
• Astronomical observations
• Any spatial point pattern or count data analysis

The package implements a fully Bayesian approach with efficient MCMC sampling, allowing for inference on complex spatial dependencies through a latent factor model structure.

• Model spatial point processes using latent Gaussian processes
• Support for multiple response types (multivariate spatial data)
• Configurable number of latent factors
• Support for multiple data replicates/images
• Parallel processing capabilities
• Various spatial correlation structures
• Support for different distributional families (Poisson, Negative Binomial)
• Implementation of spatial cross-correlation functions

You can install the development version of BIPPS from GitHub:

devtools::install_github("mkln/bipps")

For best performance, consider the following:

• OpenMP and BLAS Libraries: BIPPS works best with OpenMP and optimized BLAS libraries like OpenBLAS or Intel MKL.

• Intel MKL: Dirk Eddelbuettel has a guide on installing Intel MKL on Debian/Ubuntu systems. Add MKL_THREADING_LAYER=GNU to your ~/.Renviron file.

• AMD CPUs: When using AMD CPUs, it may be best to install intel-mkl-2019.5-075 and add MKL_DEBUG_CPU_TYPE=5 to your ~/.Renviron. More recent versions of Intel MKL have not been extensively tested.

• OpenBLAS: If using OpenBLAS, it's recommended to let OpenMP handle all parallelization when running BIPPS. This can be managed with the RhpcBLASctl package.

Here's a basic example of how to use BIPPS:

library(bipps)
library(dplyr)
library(ggplot2)

# Set random seed for reproducibility
set.seed(42)

# Create spatial grid
nx <- 30
ny <- 30
coords <- expand.grid(
  x = seq(0, 1, length.out = nx),
  y = seq(0, 1, length.out = ny)
)

# Example usage with synthetic data
# (See vignettes for complete examples)
results <- multi_bipps(
  y_list = y_list,         # Your response data (point counts)
  x_list = x_list,         # Your covariates
  coords = coords,         # Spatial coordinates
  k = 3,                   # Number of latent factors
  family = "poisson",      # Response distribution
  n_samples = 1000,        # MCMC samples to collect
  n_burn = 500,            # Burn-in iterations
  n_threads = 2,           # Parallel threads
  verbose = 10             # Progress reporting interval
)

# Access results
summary(results)

For more detailed examples and usage information, please refer to the package vignettes:

# Install package with vignettes
devtools::install_github("mkln/bipps", build_vignettes = TRUE)

# View available vignettes
vignette(package = "bipps")

BIPPS implements spatial models where:

• The data follow a point process with spatially varying intensity
• Spatial dependence is modeled through latent Gaussian processes
• A factor model structure allows for dimensionality reduction
• Spatial cross-correlations describe relationships between different response types

The spatial cross-correlation function between response types r and s is defined as:

$$ \rho_{rs}(\mathbf{h}; \mathbf{\theta}) = \frac{C_{rs}(\mathbf{h}; \mathbf{\theta})}{\sqrt{C_{rr}(\mathbf{0}; \mathbf{\theta})} \sqrt{C_{ss}(\mathbf{0}; \mathbf{\theta})}} $$

where:

$$C_{rs}(\mathbf{h}) = \sum_{j=1}^k \lambda_{rj}\lambda_{sj} \rho(\mathbf{h}; \varphi_j)$$

BIPPS is particularly useful for:

• Analyzing spatial point patterns in ecology
• Modeling disease incidence across geographical regions
• Studying stellar or galactic distributions
• Any application involving spatially distributed count data

Contributions to the package are welcome. Please feel free to submit issues or pull requests on GitHub.

If you use BIPPS in your research, please cite:

[Citation information]