Statistical tests for evaluating unimodality, in particular:

• Hartigan's dip test (equivalent to excess mass test).
• Silverman's bandwidth test.
• Calibrated versions of the above.
• Estimation of excess mass modes.

Added flow cytometry interface, for integrated preprocessing of flow cytometry data. See tests/test_fc_interface.py

K. Johnsson and M. Fontes (2017): What is a “unimodal” cell population? Using statistical tests as criteria for unimodality in automated gating and quality control. Cytometry, 91. doi:10.1002/cyto.a.23173

Hartigan and Hartigan (1985): The dip test of unimodality. The Annals of Statistics. 13(1).

Silverman (1981): Using kernel density estimates to investigate multimodality. Journal of the Royal Statistical Society. Series B. 43(1).

M.-Y. Cheng and P. Hall (1999): Mode testing in difficult cases. The Annals of Statistics. 27(4).

Müller and Sawitski (1991): Excess Mass Estimates and Tests for Multimodality. JASA. 86(415).

Cheng and Hall (1998): On mode testing and empirical approximations to distributions. Statistics & Probability Letters 39.

Basic usage of the package is showcased in the test files in the directory tests/. Each test file can be run directly or with mpi, for example

mpirun -n N python test_modality.py

where N is the number of cores to be used.

After the package has been installed (see below) it can be used from R through the R package XRPython.

NB! In your R session, before XRPython is loaded you have to set the PYTHONPATH environment variable so that it points to the directory where your Python packages are installed (e.g. "/usr/local/lib/python2.7/site-packages"). This can be done with:

Sys.setenv(PYTHONPATH="`path to packages`")

If you don't know where your packages are installed, start a Python session and write:

import os
print os.environ['PYTHONPATH']

Some basic usage is showcased in the file tests/test_R.R.

The package has the following dependencies:

• Python 2.7 or Python 3.6, as well as packages listed in setup.py.
• OpenMPI

rpy2 is necessary for the uncalibrated version of Hartigan's dip test, as well as R and the R package diptest (see Installation).

OpenMPI can be installed by:

sudo apt-get update
sudo apt-get install -y libopenmpi-dev openmpi-bin

Python and OpenMPI can be installed using homebrew. First install homebrew following the instructions at http://brew.sh. Then install Python and OpenMPI:

brew install python, openmpi

Download modality-1.1.tar.gz at

https://github.com/kjohnsson/modality/blob/master/dist/modality-1.1.tar.gz

and unpack it. In the terminal, go to the extracted directory modality-1.1 and write

pip install .

If installation for only one user is wanted, you can add the option --user.

During installation, tablulated p-values for Hartigan's dip test are obtained from the R package 'diptest'. If the package fails to load, rpy2 tries to install it. Another way to install the 'diptest' R package is to write "install.packages('diptest')" within R.

When using parallel execution with MPI, it is recommended to add the following code to your script, so that the process is aborted if only one thread fails.

from mpi4py import MPI
import traceback
import sys

if MPI.COMM_WORLD.Get_size() > 1:
    # Ensure that all threads are terminated if one fails

    def mpiexceptabort(type, value, tb):
        traceback.print_exception(type, value, tb)
        MPI.COMM_WORLD.Abort(1)

    sys.excepthook = mpiexceptabort