In sevvandi/eventstream: Streaming Events and their Early Classification

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

eventstream

The goal of eventstream is to extract and classify events in contiguous spatio-temporal data streams of 2 or 3 dimensions. For details see [@eventstream].

Installation

You can install the development version of eventstream from github with:

#install.packages("devtools")
devtools::install_github("sevvandi/eventstream")

Generate data - 2D

This is an example of a data stream you can generate with eventstream.

library("eventstream")
library("ggplot2")
library("raster")
library("maps")

str <- gen_stream(3, sd=1)
zz <- str$data
dat <- as.data.frame(t(zz))
dat.x <- 1:dim(dat)[2]
dat.y <- 1:dim(dat)[1]
mesh.xy <- eventstream:::meshgrid(dat.x,dat.y)
xyz.dat <- cbind(as.vector(mesh.xy$x), as.vector(mesh.xy$y), as.vector(as.matrix(dat)) )
xyz.dat <- as.data.frame(xyz.dat)
colnames(xyz.dat) <- c("Time", "Location", "Value")
ggplot(xyz.dat, aes(Time, Location)) + geom_raster(aes(fill=Value)) +   scale_fill_gradientn(colours=topo.colors(12)) + theme_bw()

Extract events - 2D

The extracted events are plotted for the first 2 windows using a window size of 200 and a step size of 50.

zz2 <- zz[1:250,]
ftrs <- extract_event_ftrs(zz2, rolling=FALSE, win_size=200, step_size = 50, vis=TRUE)

Extract events - 3D

To extract 3D events we use the NO2 data from NASA's NEO website.

data(NO2_2019)
dim(NO2_2019)
ftrs_2019 <- extract_event_ftrs(NO2_2019, thres=0.97, epsilon = 2, miniPts = 20, win_size=4, step_size=1, rolling=TRUE, tt=1, vis=FALSE)
dim(ftrs_2019)
ftrs_2019[1, , ]

The features contain r dim(ftrs_2019)[1] events, r dim(ftrs_2019)[2] features, and r dim(ftrs_2019)[3] age brackets for the events.

First, let us visualize NO2 data for March 2019.

data(NO2_2019)
r <- raster(NO2_2019[1, ,],xmn=-179.5,xmx=179.5,ymn=-89.5,ymx=89.5,crs="+proj=longlat +datum=WGS84")
plot(r, legend=F, main="2019 March NO2 levels")
map("world",add=T, fill=FALSE, col="darkgrey")

Visualize 2D cross sections of 3D events

Next we extract 3D events from March - June 2019. Then we visualize 2D cross sections of these 3D events for March 2019.

data(NO2_2019)
output <- get_clusters_3d(NO2_2019, thres=0.97, epsilon = 2, miniPts = 20)
cluster.all <- output$clusters
xyz.high <- output$data
all_no2_clusters_march <- xyz.high[xyz.high[,1]==1,-1]
all_cluster_ids_march <- cluster.all$cluster[xyz.high[,1]==1]
cluster_ids_march <- all_cluster_ids_march[all_cluster_ids_march!=0]
no2_clusters_march <- all_no2_clusters_march[all_cluster_ids_march!=0,]

march_map <- matrix(0, nrow=180, ncol=360)
set.seed(123)
new_ids <- sample( length(unique(cluster_ids_march)),length(unique(cluster_ids_march)) )
new_cluster_ids <- cluster_ids_march
for(i in 1:length(unique(cluster_ids_march)) ) {
  new_cluster_ids[ cluster_ids_march== unique(cluster_ids_march)[i]] <- new_ids[i]
}

march_map[no2_clusters_march[,1:2]] <- new_cluster_ids
r <- raster(march_map,xmn=-179.5,xmx=179.5,ymn=-89.5,ymx=89.5,crs="+proj=longlat +datum=WGS84")
plot(r, legend=F)

map("world",add=T, fill=FALSE, col="darkgrey")