The Local Affine Multidimensional Projection (LAMP) technique is a versatile method to map multidimensional data into a visual space. LAMP bears the a number of interesting properties:
- LAMP has been designed to make use of a set of control points to drive the mapping process. If the position of the contol points change the mapping changes accordingly. However, you can also use LAMP without any control points (see control_points_free.py);
- LAMP is also able to handle streaming data (since the control points are fixed);
- It is stable, that is, small perturbations in the data result in small perturbations in the mapping.
PS. LAMP only supports Cartesian data.
The provided class Lamp code implements the LAMP technique described in the paper:
- P. Joia, D. Coimbra, J. Cuminato, F. Paulovich, L.G. Nonato. Local affine multidimensional projection. IEEE Transactions on Visualization and Computer Graphics, 17(12):2563-71, 2011.
Please cite/acknowledge the paper above when using the code in your work.
class Lamp(Xdata = None, control_points = None, weights = None,
label=False, scale=True, dim = 2)Parameters:
Xdata: N x K matrix where N is the number of instances and K is the dimension of the data (number of attributes)
control_points: M x (dim + 1) matrix where M is the number of control points and dim is the dimension of the visual space (default dim=2). The last column of the control_points matrix must contain the ids of the control points in the Xdata matrix
weights: N x M matrix where weights[i,j] is the weight between Xdata[i] and control_point[j]. Default (weights=None) is the inverse of the Euclidean distance
label: when True assumes the last column of Xdata as labels
scale: apply a transformation in the control points before accomplish the mapping. Produce better projections when the original and control points coordiantes have very different scales (default scale=True)
dim: dimension of the image space (default dim=2)
Examples:
- MDS_as_controlpoints.py:
- shows how to use MDS to generate control points. - streaming.py:
- example on how to use LAMP to project streaming data - control_points_free.py:
- shows how to use LAMP without specifying control points.