dbnlearn-package {dbnlearn}

Dynamic Bayesian Network Structure Learning, Parameter Learning and Forecasting

Description

Dynamic Bayesian Network Structure Learning, Parameter Learning and Forecasting. This package implements a model of Gaussian Dynamic Bayesian Networks with temporal windows, based on collections of linear regressors for Gaussian nodes. The package allows learning the structure of univariate time series, learning parameters and forecasting.

Details

Package:	dbnlearn-package
Type:	Package
Version:	0.1.0
Date:	2020-07-17
License:	MIT + file LICENSE

Author(s)

Robson Fernandes
Institute of Mathematical and Computer Sciences
University of Sao Paulo - ICMC-USP

Maintainer: Robson Fernandes [e-mail]

References

Koller D, Friedman N (2009). Probabilistic Graphical Models: Principles and Techniques. MIT Press.

Korb K, Nicholson AE (2010). Bayesian Artificial Intelligence. Chapman & Hall/CRC, 2nd edition.

Pearl J (1988). Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan Kaufmann.

Nagarajan R, Scutari M, Lebre S (2013). Bayesian Networks in R with Applications in Systems Biology. Springer.

Examples

library(dbnlearn)
library(bnviewer)
library(ggplot2)

#Time Series AirPassengers
ts <- AirPassengers

#Time Series Preprocessing with time window = 12
X.ts = dbn.preprocessing(ts, window = 12)

#Define 70% Train and 30% Test Data Set
percent = 0.7
n = nrow(X.ts)

trainIndex <- seq_len(length.out = floor(x = percent * n))
X.ts.train <- X.ts[trainIndex,]
X.ts.test <- X.ts[-trainIndex,]

#Dynamic Bayesian Network Structure Learning
ts.learning = dbn.learn(X.ts.train)

#Viewer Dynamic Bayesian Network
viewer(ts.learning,
       edges.smooth = TRUE,
       bayesianNetwork.height = "400px",
       node.colors = list(background = "#f4bafd",
                          border = "#2b7ce9",
                          highlight = list(background = "#97c2fc",
                                           border = "#2b7ce9")),
       bayesianNetwork.layout = "layout_with_sugiyama")


#Dynamic Bayesian Network Fit
ts.fit = dbn.fit(ts.learning, X.ts.train)


#Predict values
prediction = dbn.predict(ts.fit, X.ts.test)


#Plot Real vs Predict
real = X.ts.test[, "X_t"]
prediction = prediction

df.validation = data.frame(list(real = real, prediction = prediction))

ggplot(df.validation, aes(seq(1:nrow(df.validation)))) +
  geom_line(aes(y = real, colour="real")) +
  geom_line(aes(y = prediction, colour="prediction")) +
  scale_color_manual(values = c(
    'real' = 'deepskyblue',
    'prediction' = 'maroon1')) +
  labs(title = "Dynamic Bayesian Network",
      subtitle = "AirPassengers Time Series",
      colour = "Legend",
      x = "Time Index",
      y = "Values") + theme_minimal()