klustakwik2 0.2.6

Installation instructions

Install Python using the Anaconda distribution. You will need to install the packages numpy, scipy, cython and nose. For Windows, Python 2.7 might be a better option than 3.x.

On all platforms, KlustaKwik can be installed using pip install klustakwik2 (from source) or conda install -c kwikteam klustakwik2 (precompiled binary). The default installation options are as follows:

• Linux: Multithreading on by default.

• Windows: Multithreading on by default if MSVC installed, otherwise off. Note that under the Anaconda distribution, multithreading will be off by default, but this can be easily resolved if you have MSVC installed, see below.

• Mac: Multithreading off by default.

To override these options, install from source (see below).

Usage

To cluster a pair of files name.fet.n, name.fmask.n run the command:

kk2_legacy name n

This will generate a name.klg.n and name.clu.n file. Note that the first time you run it, it will generate a whole lot of warnings and compiler output: ignore this, it is normal.

You can specify additional options to this script. The major ones are explained below:

• max_iterations=1000: the maximum number of full iterations to carry out before giving up.

• max_possible_clusters=1000: the maximum number of clusters to output at the end. At the moment, this option limits RAM usage but in a future version there will be no such limit.

• drop_last_n_features=0: don’t use the last N features of the fet file for clustering.

• save_clu_every=None: how often (in minutes) to save a temporary .clu file.

• num_starting_clusters=500: how many initial clusters to use (should probably be higher than the number of final clusters you expect to find).

• start_from_clu=None: if specified, start the clustering from a previously saved .clu file.

• use_mua_cluster=True: whether or not to use “black hole” MUA cluster variant of the algorithm. This is designed to improve clustering on noisier data sets, but you can turn it off if you are having problems with it. Note that if this is on, cluster 2 will be the MUA cluster, and the first ‘normal’ cluster will be cluster 3.