This repository contains an attempt to replicate some of the results that were achieved in
In particular, it is attempted to replicate the results that are illustrated in Figure 4.
Another reason why this repository was created is that the implementation of the Jump Method algorithm on Python cannot be found easily on the internet (I have not found any).
Also, this repository contains a Python class jumpmethod.py
which in turn
contains two functions: distortions
and jumps
that calculate vectors of
distortions and jumps for a given number of clusters to check.
To use this:
jm = JumpsMethod(data)
jm.distortions()
jm.jumps()
number_cluster = jm.number_clusters()
Finally, there is also a JupyterNotebook Simulations (Figure 4).ipynb
which
was created for illustrative explanations why and how the replication can be
achieved.
- Add the Transformed distortion curves to the class (easy: this is just a cumulate of Jumps).
- Add the description of the algorithm to the README.
- Replicate the Iris results (SJ, p. 12).
- Replicate the bootstrap results (SJ, pp. 13--15).
- Performance check (may be improvements).