NEW Deflation for CCA and LDA #54

Banana1530 · 2019-08-21T06:04:52Z

PR #46 is moved here.

pg_setting -> pg_settings moma_sparsity -> _moma_sparsity_type selection_scheme_str -> select_scheme_list selection_criterion* -> select_scheme*

Banana1530 · 2019-08-21T06:07:29Z

Merge into develop under the new branch scheme instead of master -DONE

Increase coverage of the new stuff - this is a big hit compared to the previous version -NOT YET

Pull the Shiny stuff out of this PR and just leave it as CCA and LDA - DONE. Pulled out.

Clean up the history - DONE.

michaelweylandt · 2019-08-21T18:53:35Z

README.Rmd


 ## Installation

 The newest version of the package can be installed from Github:

 ```{r, eval=FALSE}
 library(devtools)
-install_github("michaelweylandt/MoMA", ref="build")
+install_github("michaelweylandt/MoMA", ref = "build")


This should be changed to master now I think.

man/figures/moma-formula.svg

michaelweylandt · 2019-08-21T21:22:22Z

Is this PR supposed to add pkgdown websiting as well? If so, I think we'll need an adjustment in build_steps.R

michaelweylandt · 2019-08-21T21:24:56Z

R/moma_arguments.R

-#'   \item{\code{\link{spfusedlasso}}} TODO
-#'   \item{\code{\link{cluster}}} TODO
+#'   \item{\code{\link{moma_lasso}}}: sparsity
+#'   \item{\code{\link{moma_mcp}}}: adaptive sparsity


Maybe just "sparsity" or "non-convex sparsity" here? "Adaptive" makes me think of the adaptive lasso.

michaelweylandt · 2019-08-21T21:25:37Z

R/moma_arguments.R

+#'   \item{\code{\link{moma_grplasso}}}: group-wise sparsity
+#'   \item{\code{\link{moma_fusedlasso}}}: piece-wise constant
+#'   \item{\code{\link{moma_spfusedlasso}}}: sparsity and piece-wise constant
+#'   \item{\code{\link{moma_l1tf}}}: piece-wise quadratic


"Piecewise polynomial (default quadratic)" might be more accuracy.

michaelweylandt · 2019-08-21T21:26:06Z

R/moma_arguments.R

+#'   \item{\code{\link{moma_fusedlasso}}}: piece-wise constant
+#'   \item{\code{\link{moma_spfusedlasso}}}: sparsity and piece-wise constant
+#'   \item{\code{\link{moma_l1tf}}}: piece-wise quadratic
+#'   \item{\code{\link{moma_cluster}}}: structured sparsity


"Structured" is a bit vague - maybe "unordered fusion" (I contrast it with ordered fusion => fused lasso in my head).

michaelweylandt · 2019-08-21T21:26:43Z

R/moma_arguments.R

-#' @name moma_sparsity
+#' These functions specify the value of the \code{u_sparse,v_sparse} arguments in the
+#'  \code{moma_*pca} series of functions, and the \code{x_sparse,y_sparse} arguments
+#' in the \code{moma_*cca} and \code{moma_*lda} series of functions.


I hadn't picked up on the X vs U distinction for PCA vs multi-source methods. Nice touch.

michaelweylandt · 2019-08-21T21:28:39Z

R/moma_arguments.R

+#' @param lambda A vector containing penalty values
+#' @param select_scheme A char being either "b" (nested BIC search) or "g" (grid search).
+#'
+#' Selection of tuning parameters (four parameters in our case, \code{alpha_u}, \code{alpha_v}, \code{lambda_u}, \code{lambda_v}) is a problem in MoMA. Currently we provide


Let's not call this a problem - too negative.

Maybe something more like:

MoMA provides a flexible framework for regularized multivariate analysis with several tuning parameters for different forms of regularization. To assist the user in selecting these parameters ....

michaelweylandt · 2019-08-21T21:49:43Z

src/moma_expose.cpp

-    int selection_criterion_lambda_v = 0,
-    int max_bic_iter                 = 5,
-    int rank                         = 1)
+    int deflation_scheme       = 1,  // Defaults to 1 = PCA_Hotelling


Can we use DeflationScheme::PCA_Hotelling here instead of hard coding 1? It would make the code a bit easier to read. (Might need to change the type on deflation_scheme.)

Can Rcpp handle exporting a function with an enum-type argument?

Hmmm.... it looks like the answer is "sort of." The following works, but isn't totally type-safe. We could probably do a more general solution in the future. Issue?

#include "Rcpp.h" enum DeflationMethod { PCA = 0, PLS = 1, CCA = 2 }; namespace Rcpp { SEXP wrap(DeflationMethod dm){ Rcpp::IntegerVector dm_int = Rcpp::wrap(static_cast<int>(dm)); dm_int.attr("class") = "DeflationMethod"; return dm_int; } template <> DeflationMethod as(SEXP dm_sexp){ Rcpp::IntegerVector dm_iv(dm_sexp); int dm_int = dm_iv(0); DeflationMethod dm = static_cast<DeflationMethod>(dm_int); return dm; } } // [[Rcpp::plugins(cpp11)]] // [[Rcpp::export]] DeflationMethod make_pls(){ DeflationMethod pls = DeflationMethod::PLS; return pls; } // [[Rcpp::export]] void take_pls(DeflationMethod x){ Rcpp::Rcout << " x = " << x << std::endl; }

See new commits "Enum-fy selection scheme encoding on R side", and "Enum-fy selection scheme encoding on C++ side".

michaelweylandt · 2019-08-21T21:50:12Z

src/moma_expose.cpp

+               double EPS_inner,
+               long MAX_ITER_inner,
+               std::string solver,
+               int deflation_scheme,            // PCA = 1, CCA = 2, LDA = 3, PLS = 4


I don't think this comment is right.

michaelweylandt · 2019-08-21T21:53:38Z

README.Rmd

 )
 ```

 # MoMA: Modern Multivariate Analysis

-[![Travis-CI Build Status](https://api.travis-ci.com/michaelweylandt/MoMA.svg)](https://travis-ci.com/michaelweylandt/MoMA) [![License: GPL v2](https://img.shields.io/badge/License-GPL%20v2-brightgreen.svg)](https://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html) [![CRAN_Status_Badge](http://www.r-pkg.org/badges/version/MoMA)](https://cran.r-project.org/package=MoMA) [![Project Status: WIP – Initial development is in progress, but there has not yet been a stable, usable release suitable for the public.](http://www.repostatus.org/badges/latest/wip.svg)](http://www.repostatus.org/#wip)
+<!-- badges: start -->
+[![Travis-CI Build Status](https://api.travis-ci.com/michaelweylandt/MoMA.svg)](https://travis-ci.com/michaelweylandt/MoMA) 


I think these links need to refer to DataSlingers instead of michaelweylandt now.

Fixed. See commit "Update badges".

michaelweylandt · 2019-08-21T21:54:07Z

README.Rmd

+
+* `moma_spfusedlasso()`: Sparse fused LASSO.
+
+* `moma_l1tf()`: L-one trend filtering.


You should be able to use MathJax / LaTeX here: $\ell_1$ and have it render properly.

michaelweylandt · 2019-08-21T21:54:56Z

_pkgdown.yml

@@ -1,9 +1,72 @@
 template:
  params:
-    bootswatch: paper
+    package: tidytemplate


I don't think we want to use the tidyverse branding (unless I'm missing something).

Oh I just thought their website template looks better.

Their website says "Please don’t use it for your own package." What do you like about it? We might be able to copy/borrow aspects.

Update, it is now

template: params: bootswatch: flatly

See commit "Not use tidytemplate".

michaelweylandt · 2019-08-21T23:30:36Z

I hadn't given Travis permission to read this under the DataSlingers organization, but I just did. Hopefully the next commit in this branch will kick off a build properly.

codecov · 2019-08-22T07:44:43Z

Codecov Report

❗ No coverage uploaded for pull request base (develop@784871b). Click here to learn what that means.
The diff coverage is 76.12%.

@@            Coverage Diff             @@
##             develop      #54   +/-   ##
==========================================
  Coverage           ?   85.63%           
==========================================
  Files              ?       34           
  Lines              ?     3182           
  Branches           ?        0           
==========================================
  Hits               ?     2725           
  Misses             ?      457           
  Partials           ?        0

Impacted Files	Coverage Δ
src/moma.h	`100% <ø> (ø)`
R/moma_solve.R	`98.07% <ø> (ø)`
src/moma_solver.h	`100% <ø> (ø)`
src/moma_base.h	`100% <ø> (ø)`
src/moma_solver_BICsearch.cpp	`100% <100%> (ø)`
R/moma_arguments.R	`94.33% <100%> (ø)`
src/moma_solver.cpp	`94.73% <100%> (ø)`
R/moma_sflda.R	`59.8% <59.8%> (ø)`
R/moma_sfcca.R	`74.64% <74.64%> (ø)`
R/moma_expose.R	`91.66% <83.33%> (ø)`
... and 5 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 784871b...03923ee. Read the comment docs.

Banana1530 · 2019-08-22T09:45:55Z

src/moma.cpp

+    std::string i_solver,
+    DeflationScheme i_ds)
+    : MoMA(i_X_working.t() * i_Y_working,  // the matrix on which we find pSVD
+           i_lambda_u,


Here is why X_original cannot be zero.

What is this a reference to? Did I ask about zeros?

Oh sorry, it should be "here is why X_original cannot be constant".

Banana1530 · 2019-08-23T08:59:59Z

What changes are needed in the build_step.R?

Banana1530 · 2019-08-23T09:51:48Z

R/util.R

@@ -218,6 +281,19 @@ DEFLATION_SCHEME <- c(
    PCA_Projection = 6
 )

+# MUST be consistent with
+# `SelectionScheme` in moma_base.h
+SELECTION_SCHEME <- c(


SELECTION_SCHEME is not exposed to the user ~~and~~ but used internally.

michaelweylandt · 2019-08-23T14:45:40Z

Re: build_steps.R I think we need something to actually build the pkgdown site.

michaelweylandt · 2019-08-23T14:46:49Z

Merging now to get this in before GSoC ends.

Banana1530 added 21 commits August 21, 2019 13:59

Change of function name: reset -> set_penalty

01e1dd9

Extend C++ class MoMA to include CCA and LDA

6b71aa0

C++ interface for CCA and LDA

144b5c0

Remove redundancy in moma_sfpca.R

522f93a

Add SFCCA$initialize() and tests

cdd7f39

Add SFLDA$initialize() and tests

308cf0a

Update error message

7c63aa3

Add PCA deflation schemes on C++ side

2c4e5d7

Expose deflation_scheme to R

e206112

Add tests for PCA deflations

e11ace4

Add orthogonality tests

d66193a

_u/v -> _x/y

f25d143

Change test files' names

73f7d5a

Update test cases

4cfcc24

Change variables' names

f4a344e

pg_setting -> pg_settings moma_sparsity -> _moma_sparsity_type selection_scheme_str -> select_scheme_list selection_criterion* -> select_scheme*

Refactor SFPCA and update tests

bfc846b

Refactor SFLDA and SFCCA

a0d8b25

Add docs

62b55ca

Add vignettes

dcaecf5

a typo

1b49b8c

Give info: use sec-diff-mat as default.

e2b496e

Banana1530 mentioned this pull request Aug 21, 2019

Deflation for CCA and LDA #46

Closed

Banana1530 requested a review from michaelweylandt August 21, 2019 06:08

Banana1530 force-pushed the cca branch from 9776586 to e2b496e Compare August 21, 2019 15:33

Banana1530 mentioned this pull request Aug 21, 2019

GSoC (Google Summer of Code) 2019 Report #55

Closed

michaelweylandt reviewed Aug 21, 2019

View reviewed changes

Set knitr::opts_chunk$set(cache = FALSE)

26e77b0

Banana1530 commented Aug 22, 2019

View reviewed changes

Response

2b2d916

Banana1530 force-pushed the cca branch from b49817a to 2b2d916 Compare August 22, 2019 16:47

Update badges

8f3e2a3

michaelweylandt mentioned this pull request Aug 22, 2019

Portably Render MoMA Problem in README #58

Open

Enum-fy selection scheme encoding on C++ side

9a670cf

Enum-fy selection scheme encoding on R side

585beb7

Banana1530 mentioned this pull request Aug 23, 2019

Handle passing deflation scheme encoding between R and C++ #60

Open

Not use tidytemplate

c861f19

Banana1530 commented Aug 23, 2019

View reviewed changes

PCA_Schur_complement -> PCA_Schur_Complement

03923ee

michaelweylandt approved these changes Aug 23, 2019

View reviewed changes

michaelweylandt merged commit 5d02a2a into DataSlingers:develop Aug 23, 2019


		* `moma_spfusedlasso()`: Sparse fused LASSO.

		* `moma_l1tf()`: L-one trend filtering.

NEW Deflation for CCA and LDA #54

NEW Deflation for CCA and LDA #54

Conversation

Banana1530 commented Aug 21, 2019

Banana1530 commented Aug 21, 2019 • edited

Choose a reason for hiding this comment

michaelweylandt commented Aug 21, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Banana1530 Aug 23, 2019 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

michaelweylandt commented Aug 21, 2019

codecov bot commented Aug 22, 2019 • edited

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Banana1530 commented Aug 23, 2019

Banana1530 Aug 23, 2019 • edited

Choose a reason for hiding this comment

michaelweylandt commented Aug 23, 2019

michaelweylandt commented Aug 23, 2019

Banana1530 commented Aug 21, 2019 •

edited

Banana1530 Aug 23, 2019 •

edited

codecov bot commented Aug 22, 2019 •

edited

Banana1530 Aug 23, 2019 •

edited