overview
In bnRep: A Repository of Bayesian Networks from the Academic Literature

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(bnRep)

The R package bnRep includes the largest repository of Bayesian networks, which were all collected from recent academic literature in a variety of fields! If you are using any Bayesian network from bnRep you should cite:

Leonelli, M (2024). "bnRep: A repository of Bayesian networks from the academic literature." arXiv preprint arXiv:2409.19158.

@Article{,
    title = {bn{R}ep: A repository of {B}ayesian networks from the academic literature},
    author = {Manuele Leonelli},
    journal = {arXiv preprint arXiv:2409.19158},
    year = {2024}
  }

Go to (link here) to explore the repository online!

Contribution

If you are interested in having your Bayesian network included in bnRep you must prepare three objects:

the Bayesian network as a bn.fit object (if not created with bnlearn you can always use import functions, such as read.bif());
an R file with the same name of the bn.fit object reporting the documentation of the Bayesian network;
a vector/excel file with the required details to include in the bnRep_summary object.

You can submit the required objects directly via github (e.g fork/pull), or via email.

If you struggle with any of these steps, please get in touch and I will try to help!

Overview

bnRep includes over 200 Bayesian networks from more than 150 academic publications. It includes discrete, Gaussian and conditional linear Gaussian Bayesian networks, all stored as appropriate bn.fit objects from bnlearn. They can be exported for use to other software (e.g. Phython libraries) using functions from bnlearn such as write.bif(). Recall that in order to plot the associated DAG, one must first convert it to a graph object with bn.net() from the bnlearn package.

Installation

# Install stable version from CRAN:
install.packages("bnRep")

# Or the development version from GitHub:
remotes::install_github("manueleleonelli/bnRep")

Usage

We will use the lawschool Bayesian network as an example. To load it in the environment simply call data(lawschool) and to then plot it (for instance using graphviz.plot from the bnlearn package)

library(bnRep)
library(bnlearn)
data("lawschool")
qgraph::qgraph(bn.net(lawschool))

Notice that the function bn.net function must be used in order to plot the network.

Exploring bnRep

bnRep includes two features to explore the Bayesian networks in the repository:

bnRep_summary: a dataframe with important details about each network in the repository.
bnRep_app: a Shiny app to interactively explore bnRep_summary and filter the networks according to various criteria. The app is also available online at (link here).

Here's the columns of bnRep_summary:

colnames(bnRep_summary)

An overview of the Bayesian networks in bnRep

The following plots show some summary statistics of the repository.

library(ggplot2)
library(dplyr)
library(stringr)
library(scales)
library(RColorBrewer)

# Assuming bnRep_summary is your data frame and Type is a factor column
# Creating the barplot with percentages on the Y-axis and labels on the bars
bnRep_summary %>%
  count(Type) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Type, y = perc, fill = Type)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 100)) +
  labs(title = "Bayesian networks by type",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")

bnRep_summary %>%
  count(Structure) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Structure, y = perc, fill = Structure)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 50)) +
  labs(title = "Bayesian networks by structure definition",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")

bnRep_summary %>%
  count(Probabilities) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Probabilities, y = perc, fill = Probabilities)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 60)) +
  labs(title = "Bayesian networks by probabilities definition",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")

bnRep_summary %>%
  count(Graph) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Graph, y = perc, fill = Graph)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 70)) +
  labs(title = "Bayesian networks by graph type",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")

custom_colors <- colorRampPalette(RColorBrewer::brewer.pal(9, "Pastel1"))(length(unique(bnRep_summary$Area)))


bnRep_summary %>%
  count(Area) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Area, y = perc, fill = Area)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 3.5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 25)) +
  labs(title = "Bayesian networks by academic area",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 60, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_manual(values = custom_colors)

bnRep_summar <- bnRep_summary
bnRep_summar <- bnRep_summar %>%
  mutate(Year = ifelse(Year <= 2019, "2019 and earlier", as.character(Year)))

# Convert Year into a factor for plotting
bnRep_summar$Year <- factor(bnRep_summar$Year, levels = c("2019 and earlier", sort(unique(bnRep_summar$Year[bnRep_summar$Year != "2019 and earlier"]))))


bnRep_summar %>%
  count(Year) %>%
  mutate(perc = n / sum(n) * 100) %>%
  ggplot(aes(x = Year, y = perc, fill = Year)) +
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = paste0(round(perc, 1), "%")), vjust = -0.5, size = 5, color = "black") +
  scale_y_continuous(labels = scales::percent_format(scale = 1), limits = c(0, 35)) +
  labs(title = "Bayesian networks by year",
       x = "Type",
       y = "Percentage") +
  theme_minimal(base_size = 15) +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1),
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")

unique_journals_df <- bnRep_summary %>%
  distinct(Reference, .keep_all = TRUE) %>%
  group_by(Journal) %>%
  filter(n() >= 3) %>%
  ungroup()

unique_journals_df <- unique_journals_df %>%
  mutate(Journal = stringr::str_replace_all(Journal, "\\s", "\n")) 

# Create a barplot with counts instead of percentages
unique_journals_df %>%
  count(Journal) %>%
  ggplot(aes(x = Journal, y = n, fill = Journal)) +  # y = n for counts
  geom_bar(stat = "identity", width = 0.7, color = "black", show.legend = FALSE) +
  geom_text(aes(label = n), vjust = -0.5, size = 5, color = "black") +  # Show counts on top of bars
  labs(title = "Bayesian Networks by Journal (having at least 3)",
       x = "Journal",
       y = "Count") +
  theme_minimal(base_size = 15) +
theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),
    axis.text.x = element_text(angle = 0, hjust = 0.5),  # No rotation, centered
    panel.grid.major = element_line(color = "gray80"),
    panel.grid.minor = element_blank()
  ) +
  scale_fill_brewer(palette = "Pastel1")+
  scale_y_continuous(limits = c(0, 7))

# Create the histogram with log10 scale on the x-axis
ggplot(bnRep_summary, aes(x = Nodes)) +
  geom_histogram(binwidth = 0.1, fill = "skyblue", color = "black", alpha = 0.7) +  # Customize the appearance
  scale_x_log10() +  # Set x-axis to log10 scale
  labs(title = "Distribution of Nodes (Log10 Scale)",
       x = "Number of Nodes (log10 scale)",
       y = "Count") +
  theme_minimal(base_size = 15) +  # Use a clean theme with larger base text size
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),  # Center the title
    axis.title.x = element_text(face = "bold"),  # Make x-axis title bold
    axis.title.y = element_text(face = "bold"),  # Make y-axis title bold
    panel.grid.major = element_line(color = "gray80"),  # Customize grid lines
    panel.grid.minor = element_blank(),  # Remove minor grid lines for a cleaner look
    axis.text.x = element_text(angle = 45, hjust = 1)  # Angle the x-axis labels for better readability
  ) +
  scale_fill_brewer(palette = "Set2")  # Use a custom color palette

ggplot(bnRep_summary, aes(x = Arcs)) +
  geom_histogram(binwidth = 0.1, fill = "skyblue", color = "black", alpha = 0.7) +  # Customize the appearance
  scale_x_log10() +  # Set x-axis to log10 scale
  labs(title = "Distribution of Arcs (Log10 Scale)",
       x = "Number of Arcs (log10 scale)",
       y = "Count") +
  theme_minimal(base_size = 15) +  # Use a clean theme with larger base text size
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold"),  # Center the title
    axis.title.x = element_text(face = "bold"),  # Make x-axis title bold
    axis.title.y = element_text(face = "bold"),  # Make y-axis title bold
    panel.grid.major = element_line(color = "gray80"),  # Customize grid lines
    panel.grid.minor = element_blank(),  # Remove minor grid lines for a cleaner look
    axis.text.x = element_text(angle = 45, hjust = 1)  # Angle the x-axis labels for better readability
  ) +
  scale_fill_brewer(palette = "Set2")