nxcurve

Dimensionality reduction ( DR ) is a data transformation process which provides a low-dimensional ( attribute or variable ) representation of high dimensional data sets. This with the following purposes: noise reduction, storage space reduction, data visualization, efficient data processing and to concentrate the important information in fewer variables than the original set. A performance visual measure in DM is topology preservation. Quality curves RNX, proposed by Lee and Verleysen, evaluates performance generating a graphical representation of topology preservation. Nowadays there is a tool for topology conservation evaluation of DM algorithms, developed also by Lee and Verleysen (2009) but this tool is implemented only in Matlab. Here a problem arises because Matlab is limited and cannot be implemented in other technologies. here, we are going to implement, in python, a software evaluation module of curves RNX in order to be used in other technologies.

Installation

Use the package manager pip to install nxcurve.

pip install nxcurve

Usage

from sklearn import manifold, datasets  # datasets
from nxcurve import quality_curve

n_comp = 2        # number of components to be reduced
n_nei = 20        # nearest neighbors
nsamples = 2000   # number of points (samples)

# Creating manifold 
X, color = datasets.make_swiss_roll(n_samples=nsamples)

# Performing dimensionality reduction
X_r, err = manifold.locally_linear_embedding(X, n_neighbors=n_nei, n_components=n_comp)

# Drawing RNX curve
quality_curve(X,X_r,n_nei,'r',True)

"""
    input: X original data, X_r reduced data, n_neighbors, option, graph
    output: _NX vector, area under the curve and plot if graph == True
    Any character in the following list generates a new figure: (opt)
    q: Q_NX(K)
    b: N_NX(K)
    r: R_NX(K)
"""

Features

• RNX curve and area under the curve
• QNX curve and area under the curve
• BNX curve and area under the curve

Development

• Grahp for the coranking matrix
• LCMC from a coranking matrix (local continuity meta criterion)
• Error Handling

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT