Skip to content

Latest commit

 

History

History
72 lines (48 loc) · 2.96 KB

updates.MD

File metadata and controls

72 lines (48 loc) · 2.96 KB

Updates

11/26/18 (version 0.1.4)

  • fixed problems with package versions (I think they didn't update in previous pushes)
  • bug fix: wedin/random direction percentiles switched to correct singular value scale
  • Problems with indiv_rank =0 and pca scores projection.
  • bug fix: setting store_full = False now works
  • setting joint_rank/individual_ranks = 0 now works
  • bug fix: centers now computed correctly
  • got rid of nticks for scree plots

11/08/18 (version 0.1.3)

  • new diagnostic plots
  • fixed some visualization bugs

11/05/18 (version 0.1.2)

  • bug fix: observation names incorrectly extracted when using np.arrays as input
  • speed up: only calculate top singular value for random direction bound

11/01/18 (version 0.1.1)

  • bug fix: AJIVE.common and PCA.from_precomputed now correctly extract observation names

10/29/18 (version 0.1.0)

Major changes to both frontend and backend

  • cleaned up backend code to be more readable
  • renamed main object JIVE -> AJIVE
  • frontend now follows sklearn API more closely
  • fixed weird joint rank estimation bug
  • added jive.PCA.PCA object
  • added a number of new visualization tools
  • modified documentation to follow new API
  • TODOs left: finish documentation, testing, some minor issues noted in the code, sparse capabilities, put on pip

1/23/18 (version 0.0.2)

  • changes to Jive object API

    • block specific/common estimates now called via jive.get_block_specific_estimates() and jive.get_common_joint_space_estimate()
    • now only one wedin bound resampling procedure (e.g. removed basis resampling)
    • removed jive.compute_wedin_bound

  • added random direction bound (e.g. jive.sample_random_direction_bounds())

  • jive block returns samples from random direction (before just returned point estimate).

    • samples can be accessed via jive.wedin_sv_samples and jive.random_sv_samples

  • added diagnostic plot (e.g. jive.plot_joint_diagnostic())

  • added diagnostic demo

10/19/17 (version 0.0.1)

  • changes to Jive object API

    • Jive() no longer creates the scree plot on initialization
    • the results of SVD (i.e. U, D, V) are now called scores, sv, loadings

  • added sparse functionality

    • Jive() now works for both sparse and dense matrices (user can input a mix)
    • Jive() will not do the full SVD for sparse matrices -- the user has to specify the rank of the initial SVD (via the init_svd_ranks argument which is a list). The initial SVD rank should be larger than the initial signal rank.
    • The block specific joint/initial space estimation present a problem for large, sparse matrices. If S is a sparse matrix then UU^TS will be a dense matrix the same size as S. To prevent creating and attempting to compute the SVD of very large sparse matrices we implemented some functions based on scipy.linalg.sparse.LinearOperator that will do lazy evaluation for UU^TS and (I - UU^T)S (see lazymatpy module).

10/15/20 (version 0.0.2)

  • minor clean ups
  • this package has moved homes and this version will no longer be maintained