This repository contains concrete application examples for the micompr R package, which implements a procedure for comparing multivariate samples associated with different groups.
These examples are described in detail in the following reference:
- Fachada N, Rodrigues J, Lopes VV, Martins RC, Rosa AC. (2016) micompr: An R Package for Multivariate Independent Comparison of Observations. The R Journal 8(2):405–420. https://journal.r-project.org/archive/2016-2/fachada-rodrigues-lopes-etal.pdf
The replication of a simulation model in a new context highlights differences between the conceptual and implemented models, as well as inconsistencies in the conceptual model specification, promoting model verification, model validation and model credibility
In this example, provided in the pphpc.R script, micompr is used
for comparing the outputs of two implementations of the PPHPC agent-based
model. The compared output data is available at
https://zenodo.org/record/46848. Uncompress the data to a local folder, and
specify the folder in the
dir_data
variable within the script.
This example, provided in the sunspot.R script, uses the monthly sunspot data included with R, which contains the monthly numbers of sunspots from 1749 to the present day. The example aims to answer the following question: Were the solar cycles during the 1749–1859 interval significantly different from the more recent observations?
This example, provided in the saugeen.R script, uses the Saugeen River daily flow data included in the deseasonalize R package. This data consists of a time series of the rivers’ daily flow (m3/s) from 1915 to 1979. The example aims to answer the following question: is there any statistical difference between the flow dynamics during the 1915–1944 and 1950–1979 periods (perhaps due to climate change or some other factor)?
In this example we use the tools provided by the micompr package to study the PH2 database of dermoscopic images. This image database contains a total of 200 dermoscopic images of melanocytic lesions, including, from benign to more serious, 80 common nevi, 80 atypical nevi, and 40 melanomas. The goal is to verify if images of the three types of lesions form statistically distinguishable samples.
This example is provided in the derma.R script. However, the following pre-processing of the images was performed (using ImageMagick command-line utilities under Linux) before comparing them with micompr:
- Since each image comes in its own separate folder, we first copy all the images to the same folder:
# List folders (within the "PH2 Dataset images" folder) containing images
DIRS=`ls PH2\ Dataset\ images/`
# Create "images" folder
mkdir -p images
# Copy all images to the "images" folder
for DIR in $DIRS
do
cp PH2\ Dataset\ images/${DIR}/${DIR}_Dermoscopic_Image/${DIR}.* images
done
- These are 8-bit RGB color images, with a resolution of purportedly 768 × 560 pixels. The following command shows this is not the case, and that the image sizes vary between 761 × 570 and 769 × 577:
# List all PH2 images
IMGS=`ls images/`
# Check properties of all images
for IMG in $IMGS
do
identify images/${IMG} | cut -d " " -f 3,5
done
- As such, we resize all images to 760 × 570 prior before comparing them with micompr:
# List all PH2 images
IMGS=`ls images/`
# Create a folder for the resized images
mkdir -p images_resize
# Resize all images to 760 × 570
for IMG in $IMGS
do
convert images/${IMG} -resize 760x570\! images_resize/${IMG}
done
The folder containing the resized images is specified in the
imgfolder
variable within the derma.R script.