Manual
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 8
Package ‘akmeans’
February 20, 2019
Type Package
Title akmeans: 'Anchored' kmeans for Longitudinal Data
Version 0.1.0
Date 2019-02-06
Author Monsuru Adepeju [cre, aut], Samuel Langton [aut], Jon Bannister [aut]
Maintainer Monsuru Adepeju <monsuurg2010@gmail.com>
Description Advances an akmeans clustering technque and a stability-
based quality criterion for longitudinal data. Also, contains
functions for useful for the analysis of longitudinal data.
License GPL-2
Encoding UTF-8
LazyData TRUE
Imports kml, devtools, Hmisc, ggplot2, rgdal, base, utils, reshape2, later
Suggests knitr,
rmarkdown
RoxygenNote 6.1.1
VignetteBuilder knitr
Rtopics documented:
akmeans_clust........................................ 2
alphaLabel.......................................... 2
assault_data......................................... 3
gm_crime_data ....................................... 3
lpm_centroids........................................ 3
missingV_filler ....................................... 4
outlierDetect ........................................ 5
plot_clust .......................................... 6
props ............................................ 6
qpm_centroids........................................ 7
whiteSpaces......................................... 7
Index 8
1
2alphaLabel
akmeans_clust akmeans_clust
Description
This function group trajectories based on a given list of initial centroids
Usage
akmeans_clust(traj, id_field = FALSE, init_method = "lpm", n_clusters = 3)
Arguments
traj A matrix or data.frame with each row representing the trajectory of observations
of a unique location. The columns show the observations at consecutive time
steps.
id_field Whether the first column is a unique (id) field. Default: FALSE
init_method initialisation method. Specifying a method to determine the initial centroids for
clustering. Default: "lpm" - linear partitioning medoids @seealso lpm_centroids]
n_clusters number of clusters to generate. Default: 3: (minimum value)
Details
Given a list of trajectories represented in a matrix or data.frame, and a method for choosing initial
cluster centroids (e.g. lpm_centroids), a list of clusters is generated after a limited number of
iterations. traj <- assault_data print(traj) result <- akmeans_clust(traj, id_field = TRUE, init_method
= "lpm", n_clusters = 3) plot_clust(result)
Value
The original (traj) data with cluster label appended
alphaLabel Numerics ids to alphabetical ids
Description
Function to transform a list of numeric ids to alphabetic ids
Usage
alphaLabel(x)
Arguments
xA vector of numeric ids
assault_data 3
assault_data Sample crime (assault) dataset
Description
Simulated crime dataset with missing values.
Usage
assault_data
Format
A matrix
gm_crime_data Sample crime dataset
Description
Crime dataset of greater Manchester crime data aggregated at the LSOA geographical level data
(Source: data.police.uk)
Usage
gm_crime_data
Format
A matrix
lpm_centroids Linear Partition Medoids (LPM) Centroids
Description
This function to create the initial centroids based on linear partitioning medoids (lpm) initialisation
(Adepeju et al. 2019, submitted)
Usage
lpm_centroids(dat, id_field2 = FALSE, n_centroids = 3)
Arguments
dat A matrix or data.frame with each row representing the trajectory of observations
of a unique location. The columns show the observation at consecutive time
steps.
id_field2 Whether the first column is a unique (id) field. default: FALSE
n_centroids Number of initial (linear) centroids to generate based on lpm technique
4missingV_filler
Value
l_centroids
References
Adepeju M, Langton S, Bannister J. (2019). akmeans: Anchored k-means: A longitudinal clustering
technique for measuring long-term inequality in the exposure to crime at the micro-area levels
(submitted).
missingV_filler Data imputing for longitudinal data
Description
This function fills up any missing entries (NA, Inf, 0) in a matrix or dataframe using a value
derived using a chosen method.
Usage
missingV_filler(traj, id_field = FALSE, method = 2, replace_with = 1, fill_zeros = FALSE)
Arguments
traj A matrix or data.frame with each row representing the trajectory of a unique
location. The columns show the observations at consecutive time steps.
id_field Whether the first column is a unique (id) field. default: FALSE
method Method for calculating the missing values. Available options: 1: arithmetic, 2:
regression. default: 1
replace_with How to calculate the missing value. For arithmetic method: replace_with
options are: 1: Mean value of column, 2: Minimum value of column, 3: Max-
imum value of column, 4: Mean value of row, 5: Minimum value of row, or
6: Maximum value of row. For regression method: the only available option
for replace_with is: 1: linear. That is, use a linear regression to interpolate or
extrapolate the missing data values. Note: only the missing data points derive
their new values from the regression line while the rest of the data points retain
their original values. Trajectories with only one observation will be removed.
fill_zeros Whether to consider zeros (0) as missing values. Default: FALSE. Only available
for 2:regression method.
Details
Given a matrix or data.frame with some missing values represented by (NA, Inf, 0), the func-
tion missingV_filler determines the missing values using either the arithmetic or regression
method.
Value
A data.frame with missing values (NA, Inf, 0) filled up
outlierDetect 5
Examples
traj <- assault_data
print(traj)
missingV_filler(traj, id_field = TRUE, method = 2, replace_with = 1, fill_zeros = FALSE)
outlierDetect Outlier detection in longitudinal or repeated observations
Description
Detect outlier in a longitudinal or repeated data. This function identify the outlier observations ac-
cording to a specified method. A matrix, ’outlier_mat’, is created with entries ’TRUE’ or ’FALSE’
indicating whether or not an observation is an outlier. The final list of outlier trajectories is de-
termined by the ’hortz_tolerance’ parameter i.e. how many observation in a trajectory exceed the
’threshold’ value.
Usage
outlierDetect(dat, id_field = FALSE, method = "quantile",
threshold = 0.95, hortz_tolerance = 1, replace_with = "Mean_row")
Arguments
dat A matrix or data.frame with each row representing the trajectory of observations
of a unique location. The columns show the observation at consecutive time
steps.
id_field Whether the first column is a unique (id) field. [default: FALSE]
method Specify the method for identifying the outlier. Available methods: (1) "quantile"
(2) "manual" - a user-defined value
threshold Value in which an observation must exceed in order to be flagged as outlier.
Depending on the method specified: (1) for "quantile" method, enter a numeric
vector of probabilities with values in [0,1], (2) for "Manual" method: a user-
specified value.
hortz_tolerance
Specifying the number of observations of a trajectory that have to exceed the cut-
off ’threshold’ value in order for the trajectory to be flagged as outlier. [default:
1]
replace_with Value to replace the outlier observation with. Values to replace with [Values:
"Mean_col" or "Mean_row"]. The default is "Mean_row", meaning to imput the
average values of the field in which the observation is located.
Value
dat_
6props
plot_clust To plot the clusters
Description
To plot the clusters
Usage
plot_clust(data_clusters_list, id_field = TRUE)
Arguments
data_clusters_list
A data.frame of clusters from akmeans_clust, in which the last column repre-
sents alphabetical cluster ids (labels)
id_field Whether the first column is a unique (id) field. [default: TRUE]
Value
data_clusters_list
props Function to convert counts or rates to proportion
Description
Function to convert counts or rates to proportion
Usage
props(rates, id_field = FALSE)
Arguments
rates A matrix or data.frame with each row representing the trajectory of observations
of a unique location. The columns show the observation at consecutive time
steps.
id_field Whether the first column is a unique (id) field. [default: FALSE]
Value
props
qpm_centroids 7
qpm_centroids Quadratic Partition Medoids (QPM) Centroids
Description
Quadratic Partition Medoids (QPM) Centroids
Usage
qpm_centroids(dat, n_centroids = 3, id_field = FALSE)
Arguments
dat A matrix or data.frame with each row representing the trajectory of observations
of a unique location. The columns show the observation at consecutive time
steps.
n_centroids Number of initial (quadratic) centroids to generate based on the qpm method
(See attached Vignette)
id_field Whether the first column is a unique (id) field. [default: FALSE]
Value
q_centroids
whiteSpaces Function to remove whitespaces in data entries
Description
Function to remove whitespaces in data entries
Usage
whiteSpaces(dat, head = TRUE)
Arguments
dat A matrix or data.frame
head If column names exist
Value
dat_Cleaned
