-Added a flag use.mkl for users who have installed rpca-mkl, resulting in dramatic speedups over the default rpca-based version.
-When choosing k, we no longer use the the normal distribution to approximate the distribution of spacings and compute p-values. Instead, we set k to correspond to the largest spacing that is more than 6 standard deviations from the "noise" singular value spacings. This leads to effectively the same choice of k as the original version, but does not make the normality assumption (that is not true).
-Instead of thresholding based on the most negative value of each gene, we threshold based on the 0.001 quantile, which is more robust to outliers. By setting the quantile probability to 0, you will restore the original behavior of thresholding by the most negative value.
-Fixed bug which introduced NAs during the scaling process for some datasets
-Added a flag
use.mklfor users who have installed rpca-mkl, resulting in dramatic speedups over the default rpca-based version. -When choosingk, we no longer use the the normal distribution to approximate the distribution of spacings and compute p-values. Instead, we setkto correspond to the largest spacing that is more than 6 standard deviations from the "noise" singular value spacings. This leads to effectively the same choice ofkas the original version, but does not make the normality assumption (that is not true). -Instead of thresholding based on the most negative value of each gene, we threshold based on the 0.001 quantile, which is more robust to outliers. By setting the quantile probability to 0, you will restore the original behavior of thresholding by the most negative value. -Fixed bug which introduced NAs during the scaling process for some datasets