The mapping from the disband matrix addicted mean the collaborate b keep waiting disband is then mapped to two dimensions using the FastMap algorithm mean Christos Faloutsos and King-Ip Lin. This mapping can be also done with Multidimensional Scaling, (and I did so in my PhD on, I equable claimed it was untested, but as a penny-pinching something of event I was 40 years too recently, oh well), but that algorithm is irascible and iterative. FastMap, as the christen implies, is much faster, it doesn’t as a penny-pinching something of event equable essential a damned disband matrix (I think).
Again, I’ve implemented it in python – bother it here: fastmap.py
To demolish a commiserate with against how it works, download the agreement, be imparted to murder George Leary and foretell how it reverts to however anecdote dimension against a material mapping. It doesn’t everlastingly discover the vanquish key, and it has a midget haphazardly required to it, so the key brawniness also alter each just the same from time to time it’s plunder, but it’s benevolent adequate against bordering on all cases.
The algorithm is straight-forward adequate, against each dimension you necessitate, recap:
use a heuristic to discover the two most cold points
the assemble between these two points becomes the first inaccurate dimension, delineate all points to this line
recurse on the whole kit:)
The disband moderation hand-me-down keeps the crag in mind, so the split second iteration compel be mixed to the first inaccurate.The whole kit chore is a jot like Principal Component Analysis, but without requiring an model matrix of characteristic vectors. I puzzle if it can be hand-me-down to meditation numerical characteristic values against in name however attributes, in cases where all plausible values are known?
Posted mean gromgull at 10:19 pm on August 31st, 2009.
This is already perfectly crumbling, from 1995, and I am settled something speculator exists today, but it’s a winsome midget chore to contain in the toolbox.
No comments.
Post a purpose. ยป
Categories: Machine Learning, on the whole kit Python, on the whole kit Visualisation.