pca_impl

  • Available in: PCA
  • Hyperparameter: no

Description

The pca_impl parameter allows you to specify PCA implementations for Singular-Value Decomposition (SVD) or Eigenvalue Decomposition (EVD), using either the Matrix Toolkit Java (MTJ) libary or the Java Matrix (JAMA) library.

Available options include:

  • mtj_evd_densematrix: Eigenvalue decompositions for dense matrix using MTJ
  • mtj_evd_symmmatrix: Eigenvalue decompositions for symmetric matrix using MTJ (default)
  • mtj_svd_densematrix: Singular-value decompositions for dense matrix using MTJ
  • jama: Eigenvalue decompositions for dense matrix using JAMA

Example

  • r
  • python
library(h2o)
h2o.init()

# Load the US Arrests dataset
arrestsH2O = h2o.importFile("https://s3.amazonaws.com/h2o-public-test-data/smalldata/pca_test/USArrests.csv")

# Train using the JAMA PCA implementation option
model <- h2o.prcomp(training_frame=arrestsH2O, k=4, pca_impl="JAMA", seed=1234)

# View the importance of components
model@model$importance
Importance of components:
                              pc1       pc2      pc3      pc4
Standard deviation     202.723056 27.832264 6.523048 2.581365
Proportion of Variance   0.980347  0.018479 0.001015 0.000159
Cumulative Proportion    0.980347  0.998826 0.999841 1.000000

# View the eigenvectors
model@model$eigenvectors
Rotation:
               pc1       pc2       pc3       pc4
Murder   -0.042392 -0.016163  0.065884  0.996795
Assault  -0.943957 -0.320686 -0.066552 -0.040946
UrbanPop -0.308428  0.938459 -0.154967  0.012343
Rape     -0.109637  0.127257  0.983471 -0.067603