Reconstruction from compressed representation

  |   Source
Reconstruction from compressed representation
  • PCA compression thousand into hundred dimensional features
  • If there's a compressed algorithm, there should be uncompressed algorithm to give data compression back to its original value
  • With this, we can uncompressed  the data reduction earlier, Ex hundred into thousand dimensional features(original_

  • The graph above shows how we first make z1 projection line and give each example to be projected
  • So how do we return the compressed back to original x?
  • Left, original data, compressed then the right give the x approx
  • Also called Reconstruction from the construct representation
  • Quiz below , max retain should be equal to one. if more, than the data retain have over-variance over the original data.

  • These method shows us how we can convert the data reduction matrix z back to original x(or its original approximation)
  • NEXT, mechanic how to use PCA well , how to choose k for reducing matrix z