1. CoGAPS
  2. Commit 2cdb0256a0675cee0896dab037fe9fe505b0fffb
Browse code

new R Interface working

Tom Sherman authored on 02/07/2018 00:58:55
Showing24 changed files
DESCRIPTION
History View file @2cdb0256a
... ... 
@@ -51,7 +51,6 @@ Collate:
51 51 
     'CoGAPS.R'
52 52 
     'GWCoGAPS.R'
53 53 
     'RcppExports.R'
54 
-    'SMatrix.R'
55 54 
     'binaryA.R'
56 55 
     'calcCoGAPSStat.R'
57 56 
     'calcGeneGSStat.R'
NAMESPACE
History View file @2cdb0256a
... ... 
@@ -1,6 +1,6 @@
1 1 
 # Generated by roxygen2: do not edit by hand
2 2
3 
-export()
3 
+export(CoGAPS)
4 4 
 export(GWCoGAPS)
5 5 
 export(GWCoGapsFromCheckpoint)
6 6 
 export(MergeResultsWithSCE)
R/CoGAPS.R
History View file @2cdb0256a
... ... 
@@ -1,10 +1,11 @@
1 1 
 #' @include class-CogapsParams.R
2 2 
 NULL
3 3
4 
-#' CoGAPS
5 
-#' @name CoGAPS Matrix Factorization Algorithm
4 
+#' CoGAPS Matrix Factorization Algorithm
5 
+#' @export
6 6 
 #' @docType methods
7 7 
 #' @rdname CoGAPS-methods
8 
+#'
8 9 
 #' @description calls the C++ MCMC code and performs Bayesian
9 10 
 #' matrix factorization returning the two matrices that reconstruct
10 11 
 #' the data matrix
... ... 
@@ -15,34 +16,52 @@ NULL
15 16 
 #' @param uncertainty uncertainty matrix (same supported types as data)
16 17 
 #' @param fixedMatrix data for fixing the values of either the A or P matrix;
17 18 
 #'  used in conjuction with whichMatrixFixed (see CogapsParams)
18 
-#' @param checkpointFile name of the checkpoint file
19 
+#' @param checkpointInFile name of the checkpoint file
19 20 
 #' @param ... keeps backwards compatibility with arguments from older versions
20 21 
 #' @return CogapsResult object
21 22 
 #' @examples
22 23 
 #' # Running from R object
23 24 
 #' data(GIST)
24 25 
 #' resultA <- CoGAPS(GIST.D)
26 
+#'
25 27 
 #' # Running from file name
26 28 
 #' gist_path <- system.file("extdata/GIST.mtx", package="CoGAPS")
27 29 
 #' resultB <- CoGAPS(gist_path)
30 
+#'
28 31 
 #' Setting Parameters
29 32 
 #' params <- new("CogapsParams")
30 33 
 #' params <- setParam(params, "nPatterns", 5)
31 34 
 #' resultC <- CoGAPS(GIST.D, params)
32 35 
 #' @importFrom methods new
33 
-#' @export
34 36 
 setGeneric("CoGAPS", function(data, params=new("CogapsParams"),
35 
-uncertainty=NULL, fixedMatrix=NULL, checkpointFile=NULL, ...)
37 
+uncertainty=NULL, fixedMatrix=matrix(0), checkpointInFile="", ...)
36 38 
 {
37 39 
     # parse parameters from ...
38 40 
     params <- parseOldParams(params, list(...))
39 41 
     params <- parseDirectParams(params, list(...))
40 42
43 
+    # check if fixed matrix is set so is whichMatrixFixed
44 
+    if (!all(fixedMatrix == matrix(0)) & (!params@whichMatrixFixed %in% c('A', 'P')))
45 
+        stop("fixedMatrix passed, but whichMatrixFixed not set")
46 
+
47 
+    # check if uncertainty is null
48 
+    if (is.null(uncertainty) & class(data) == "character")
49 
+        uncertainty <- ""
50 
+    else if (is.null(uncertainty))
51 
+        uncertainty <- matrix(0)
52 
+
53 
+    # check if we're running from a checkpoint
54 
+    if (nchar(checkpointInFile) > 0)
55 
+    {
56 
+        message(paste("Running CoGAPS from a checkpoint, all parameters",
57 
+            "besides data and uncertainty will be ignored"))
58 
+    }
59 
+
41 60 
     # call method
42 61 
     gapsReturnList <- standardGeneric("CoGAPS")
43 62
44 63 
     # convert list to CogapsResult object
45 
-    return(CogapsResult(
64 
+    return(new("CogapsResult",
46 65 
         Amean       = gapsReturnList$Amean,
47 66 
         Asd         = gapsReturnList$Asd,
48 67 
         Pmean       = gapsReturnList$Pmean,
... ... 
@@ -50,114 +69,91 @@ uncertainty=NULL, fixedMatrix=NULL, checkpointFile=NULL, ...)
50 69 
         seed        = gapsReturnList$seed,
51 70 
         meanChiSq   = gapsReturnList$meanChiSq,
52 71 
         diagnostics = gapsReturnList$diagnostics
53 
-    ))
72 
+    ))
54 73 
 })
55 74
56 75 
 #' @rdname CoGAPS-methods
57 76 
 #' @aliases CoGAPS
58 77 
 #' @importFrom tools file_ext
59 78 
 setMethod("CoGAPS", signature(data="character", params="CogapsParams"),
60 
-function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
79 
+function(data, params, uncertainty, fixedMatrix, checkpointInFile, ...)
61 80 
 {
62 81 
     # check file extension
63 82 
     if (!(file_ext(data) %in% c("tsv", "csv", "mtx")))
64 83 
         stop("unsupported file extension for data")
65 84
66 85 
     # check uncertainty matrix
67 
-    if (!is.null(uncertainty))
68 
-    {
69 
-        if (class(uncertainty) != "character")
70 
-            stop("uncertainty must be same data type as data (file name)")
71 
-        if (!(file_ext(uncertainty) %in% c("tsv", "csv", "mtx")))
72 
-            stop("unsupported file extension for uncertainty")
73 
-    }
86 
+    if (class(uncertainty) != "character")
87 
+        stop("uncertainty must be same data type as data (file name)")
88 
+    if (nchar(uncertainty) > 0 & !(file_ext(uncertainty) %in% c("tsv", "csv", "mtx")))
89 
+        stop("unsupported file extension for uncertainty")
74 90
75 91 
     # call C++ function
76 
-    cogaps_cpp_from_file(data, uncertainty, params@nPatterns,
77 
-        params@nIterations, params@outputFrequency, params@seed, params@alphaA,
78 
-        params@alphaP, params@maxGibbsMassA, params@maxGibbsMassP,
79 
-        params@messages, params@singleCell, params@checkpointOutFile,
80 
-        params@nCores)
92 
+    cogaps_cpp_from_file(data, params, uncertainty, fixedMatrix, checkpointInFile)
81 93 
 })
82 94
83 95 
 #' @rdname CoGAPS-methods
84 96 
 #' @aliases CoGAPS
85 97 
 setMethod("CoGAPS", signature(data="matrix", params="CogapsParams"),
86 
-function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
98 
+function(data, params, uncertainty, fixedMatrix, checkpointInFile, ...)
87 99 
 {
88 100 
     # check matrix
89 
-    if (!is.null(uncertainty) & class(uncertainty) != "matrix")
101 
+    if (class(uncertainty) != "matrix")
90 102 
         stop("uncertainty must be same data type as data (matrix)")
91 
-    checkDataMatrix(data, uncertainty)
103 
+    checkDataMatrix(data, uncertainty, params)
92 104
93 105 
     # call C++ function
94 
-    cogaps_cpp(data, uncertainty, params@nPatterns,
95 
-        params@nIterations, params@outputFrequency, params@seed, params@alphaA,
96 
-        params@alphaP, params@maxGibbsMassA, params@maxGibbsMassP,
97 
-        params@messages, params@singleCell, params@checkpointOutFile,
98 
-        params@nCores)
106 
+    cogaps_cpp(data, params, uncertainty, fixedMatrix, checkpointInFile)
99 107 
 })
100 108
101 109 
 #' @rdname CoGAPS-methods
102 110 
 #' @aliases CoGAPS
103 111 
 setMethod("CoGAPS", signature(data="data.frame", params="CogapsParams"),
104 
-function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
112 
+function(data, params, uncertainty, fixedMatrix, checkpointInFile, ...)
105 113 
 {
106 114 
     # check matrix
107 
-    if (!is.null(uncertainty) & class(uncertainty) != "matrix")
115 
+    if (class(uncertainty) != "matrix")
108 116 
         stop("uncertainty must be matrix when data is a data.frame")
109 
-    checkDataMatrix(data.matrix(data), uncertainty)
117 
+    checkDataMatrix(data.matrix(data), uncertainty, params)
110 118
111 119 
     # call C++ function
112 
-    cogaps_cpp(data.matrix(data), uncertainty, params@nPatterns,
113 
-        params@nIterations, params@outputFrequency, params@seed, params@alphaA,
114 
-        params@alphaP, params@maxGibbsMassA, params@maxGibbsMassP,
115 
-        params@messages, params@singleCell, params@checkpointOutFile,
116 
-        params@nCores)
120 
+    cogaps_cpp(data.matrix(data), params, uncertainty, fixedMatrix, checkpointInFile)
117 121 
 })
118 122
119 123 
 #' @rdname CoGAPS-methods
120 124 
 #' @aliases CoGAPS
121 125 
 #' @importClassesFrom SummarizedExperiment SummarizedExperiment
122 126 
 setMethod("CoGAPS", signature(data="SummarizedExperiment", params="CogapsParams"),
123 
-function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
127 
+function(data, params, uncertainty, fixedMatrix, checkpointInFile, ...)
124 128 
 {
125 129 
     # extract count matrix
126 130 
     countMatrix = assay(data, "counts")
127 131
128 132 
     # check matrix
129 
-    if (!is.null(uncertainty) & class(uncertainty) != "matrix")
133 
+    if (class(uncertainty) != "matrix")
130 134 
         stop("uncertainty must be matrix when data is a SummarizedExperiment")
131 
-    checkDataMatrix(countMatrix, uncertainty)
135 
+    checkDataMatrix(countMatrix, uncertainty, params)
132 136
133 137 
     # call C++ function
134 
-    cogaps_cpp(countMatrix, uncertainty, params@nPatterns,
135 
-        params@nIterations, params@outputFrequency, params@seed, params@alphaA,
136 
-        params@alphaP, params@maxGibbsMassA, params@maxGibbsMassP,
137 
-        params@messages, params@singleCell, params@checkpointOutFile,
138 
-        params@nCores)
138 
+    cogaps_cpp(countMatrix, params, uncertainty, fixedMatrix, checkpointInFile)
139 139 
 })
140 140
141 141 
 #' @rdname CoGAPS-methods
142 142 
 #' @aliases CoGAPS
143 143 
 #' @importClassesFrom SingleCellExperiment SingleCellExperiment
144 144 
 setMethod("CoGAPS", signature(data="SingleCellExperiment", params="CogapsParams"),
145 
-function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
145 
+function(data, params, uncertainty, fixedMatrix, checkpointInFile, ...)
146 146 
 {
147 147 
     # extract count matrix
148 148 
     countMatrix = assay(data, "counts")
149 149
150 150 
     # check matrix
151 
-    if (!is.null(uncertainty) & class(uncertainty) != "matrix")
151 
+    if (class(uncertainty) != "matrix")
152 152 
         stop("uncertainty must be matrix when data is a SingleCellExperiment")
153 
-    checkDataMatrix(countMatrix, uncertainty)
153 
+    checkDataMatrix(countMatrix, uncertainty, params)
154 154
155 155 
     # call C++ function
156 
-    cogaps_cpp(countMatrix, uncertainty, params@nPatterns,
157 
-        params@nIterations, params@outputFrequency, params@seed, params@alphaA,
158 
-        params@alphaP, params@maxGibbsMassA, params@maxGibbsMassP,
159 
-        params@messages, params@singleCell, params@checkpointOutFile,
160 
-        params@nCores)
156 
+    cogaps_cpp(countMatrix, params, uncertainty, fixedMatrix, checkpointInFile)
161 157 
 })
162 158
163 159 
 #' Information About Package Compilation
... ... 
@@ -171,4 +167,17 @@ function(data, params, uncertainty, fixedMatrix, checkpointFile, ...)
171 167 
 buildReport <- function()
172 168 
 {
173 169 
     getBuildReport_cpp()
170 
+}
171 
+
172 
+#' Check that provided data is valid
173 
+#'
174 
+#' @param data data matrix
175 
+#' @param uncertainty uncertainty matrix
176 
+#' @return throws an error if data has problems
177 
+checkDataMatrix <- function(data, uncertainty, params)
178 
+{
179 
+    if (sum(data < 0) > 0 | sum(uncertainty < 0) > 0)
180 
+        stop("negative values in data and/or uncertainty matrix")
181 
+    if (nrow(data) == params@nPatterns || ncol(data) == params@nPatterns)
182 
+        stop("nPatterns must be less than dimensions of data")
174 183 
 }
175 184 
\ No newline at end of file
R/GWCoGAPS.R
History View file @2cdb0256a
... ... 
@@ -149,7 +149,7 @@ runInitialPhase <- function(simulationName, allDataSets, nFactor, ...)
149 149
150 150 
         # run CoGAPS without any fixed patterns
151 151 
         cptFileName <- paste(simulationName, "_initial_cpt_", i, ".out", sep="")
152 
-        CoGAPS(sampleD, sampleS, nFactor=nFactor, seed=nut[i],
152 
+        CoGAPS(sampleD, uncertainty=sampleS, nFactor=nFactor, seed=nut[i],
153 153 
             checkpointFile=cptFileName, ...)
154 154 
     }
155 155 
     return(initialResult)
... ... 
@@ -202,7 +202,7 @@ runFinalPhase <- function(simulationName, allDataSets, consensusPatterns, nCores
202 202
203 203 
         # run CoGAPS with fixed patterns
204 204 
         cptFileName <- paste(simulationName, "_final_cpt_", i, ".out", sep="")
205 
-        CoGAPS(sampleD, sampleS, fixedPatterns=consensusPatterns,
205 
+        CoGAPS(sampleD, uncertainty=sampleS, fixedMatrix=consensusPatterns,
206 206 
             nFactor=nFactorFinal, seed=nut[i], checkpointFile=cptFileName,
207 207 
             whichMatrixFixed='P', ...)
208 208
R/RcppExports.R
History View file @2cdb0256a
... ... 
@@ -1,12 +1,12 @@
1 1 
 # Generated by using Rcpp::compileAttributes() -> do not edit by hand
2 2 
 # Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
3 3
4 
-cogaps_cpp_from_file <- function(data, unc, nPatterns, maxIterations, outputFrequency, seed, alphaA, alphaP, maxGibbsMassA, maxGibbsMassP, messages, singleCell, checkpointOutFile, nCores) {
5 
-    .Call('_CoGAPS_cogaps_cpp_from_file', PACKAGE = 'CoGAPS', data, unc, nPatterns, maxIterations, outputFrequency, seed, alphaA, alphaP, maxGibbsMassA, maxGibbsMassP, messages, singleCell, checkpointOutFile, nCores)
4 
+cogaps_cpp_from_file <- function(data, params, unc, fixedMatrix, checkpointInFile) {
5 
+    .Call('_CoGAPS_cogaps_cpp_from_file', PACKAGE = 'CoGAPS', data, params, unc, fixedMatrix, checkpointInFile)
6 6 
 }
7 7
8 
-cogaps_cpp <- function(data, unc, nPatterns, maxIterations, outputFrequency, seed, alphaA, alphaP, maxGibbsMassA, maxGibbsMassP, messages, singleCell, checkpointOutFile, nCores) {
9 
-    .Call('_CoGAPS_cogaps_cpp', PACKAGE = 'CoGAPS', data, unc, nPatterns, maxIterations, outputFrequency, seed, alphaA, alphaP, maxGibbsMassA, maxGibbsMassP, messages, singleCell, checkpointOutFile, nCores)
8 
+cogaps_cpp <- function(data, params, unc, fixedMatrix, checkpointInFile) {
9 
+    .Call('_CoGAPS_cogaps_cpp', PACKAGE = 'CoGAPS', data, params, unc, fixedMatrix, checkpointInFile)
10 10 
 }
11 11
12 12 
 getBuildReport_cpp <- function() {
R/SMatrix.R
History View file @2cdb0256a
13 13 
deleted file mode 100644
... ... 
@@ -1,75 +0,0 @@
1 
-CoGAPS <- function(D, S=NULL, nFactor=7, nEquil=1000, nSample=1000, nOutputs=1000,
2 
-                   nSnapshots=0, alphaA=0.01, alphaP=0.01, maxGibbmassA=100, maxGibbmassP=100,
3 
-                   seed=-1, messages=TRUE, singleCellRNASeq=FALSE, whichMatrixFixed='N',
4 
-                   fixedPatterns=matrix(0), checkpointInterval=0,
5 
-                   checkpointFile="gaps_checkpoint.out", nCores=1, ...)
6 
-{
7 
-  #returns a default uncertainty matrix of 0.1*D dataset if it's greater than 0.1
8 
-  if(!is.null(S))
9 
-    S <- pmax(0.1*D, 0.1)
10 
-
11 
-  # get v2 arguments
12 
-  oldArgs <- list(...)
13 
-  if (!is.null(oldArgs$nOutR))
14 
-    nOutputs <- oldArgs$nOutR
15 
-  if (!is.null(oldArgs$max_gibbmass_paraA))
16 
-    maxGibbmassA <- oldArgs$max_gibbmass_paraA
17 
-  if (!is.null(oldArgs$max_gibbmass_paraP))
18 
-    maxGibbmassP <- oldArgs$max_gibbmass_paraP
19 
-  if (!is.null(oldArgs$sampleSnapshots) & is.null(oldArgs$numSnapshots))
20 
-    nSnapshots <- 100
21 
-  if (!is.null(oldArgs$sampleSnapshots) & !is.null(oldArgs$numSnapshots))
22 
-    nSnapshots <- oldArgs$numSnapshots
23 
-  if (missing(D) & !is.null(oldArgs$data))
24 
-    D <- oldArgs$data
25 
-  if (missing(S) & !is.null(oldArgs$unc))
26 
-    S <- oldArgs$unc
27 
-
28 
-  # get pump arguments - hidden for now from user
29 
-  pumpThreshold <- "unique"
30 
-  nPumpSamples <- 0
31 
-  if (!is.null(list(...)$pumpThreshold))
32 
-    pumpThreshold <- list(...)$pumpThreshold
33 
-  if (!is.null(list(...)$nPumpSamples))
34 
-    pumpThreshold <- list(...)$nPumpSamples
35 
-
36 
-  # check arguments
37 
-  if (class(D) != "matrix" | class(S) != "matrix")
38 
-    stop('D and S must be matrices')
39 
-  if (any(D < 0) | any(S < 0))
40 
-    stop('D and S matrix must be non-negative')
41 
-  if (nrow(D) != nrow(S) | ncol(D) != ncol(S))
42 
-    stop('D and S matrix have different dimensions')
43 
-  if (whichMatrixFixed == 'A' & nrow(fixedPatterns) != nrow(D))
44 
-    stop('invalid number of rows for fixedPatterns')
45 
-  if (whichMatrixFixed == 'A' & ncol(fixedPatterns) > nFactor)
46 
-    stop('invalid number of columns for fixedPatterns')
47 
-  if (whichMatrixFixed == 'P' & nrow(fixedPatterns) > nFactor)
48 
-    stop('invalid number of rows for fixedPatterns')
49 
-  if (whichMatrixFixed == 'P' & ncol(fixedPatterns) != ncol(D))
50 
-    stop('invalid number of columns for fixedPatterns')
51 
-  thresholdEnum <- c("unique", "cut")
52 
-
53 
-  # get seed
54 
-  if (seed < 0)
55 
-  {
56 
-    # TODO get time in milliseconds
57 
-    seed <- 0
58 
-  }
59 
-
60 
-  # run algorithm with call to C++ code
61 
-  result <- cogaps_cpp(D, S, nFactor, nEquil, nEquil/10, nSample, nOutputs,
62 
-                       nSnapshots, alphaA, alphaP, maxGibbmassA, maxGibbmassP, seed, messages,
63 
-                       singleCellRNASeq, whichMatrixFixed, fixedPatterns, checkpointInterval,
64 
-                       checkpointFile, which(thresholdEnum==pumpThreshold), nPumpSamples,
65 
-                       nCores)
66 
-
67 
-  # label matrices and return list
68 
-  patternNames <- paste('Patt', 1:nFactor, sep='')
69 
-  rownames(result$Amean) <- rownames(result$Asd) <- rownames(D)
70 
-  colnames(result$Amean) <- colnames(result$Asd) <- patternNames
71 
-  rownames(result$Pmean) <- rownames(result$Psd) <- patternNames
72 
-  colnames(result$Pmean) <- colnames(result$Psd) <- colnames(D)
73 
-  return(v2CoGAPS(result, ...)) # backwards compatible with v2
74 
-}
75 
-
R/class-CogapsParams.R
History View file @2cdb0256a
... ... 
@@ -12,7 +12,7 @@ setClass("CogapsParams", slots = c(
12 12 
     maxGibbsMassP = "numeric",
13 13 
     seed = "numeric",
14 14 
     messages = "logical",
15 
-    singleCellRNASeq = "logical",
15 
+    singleCell = "logical",
16 16 
     whichMatrixFixed = "character",
17 17 
     checkpointInterval = "numeric",
18 18 
     checkpointOutFile = "character",
... ... 
@@ -36,7 +36,7 @@ setMethod("initialize", "CogapsParams",
36 36 
         .Object@maxGibbsMassP <- 100
37 37 
         .Object@seed <- getMilliseconds(as.POSIXlt(Sys.time()))
38 38 
         .Object@messages <- TRUE
39 
-        .Object@singleCellRNASeq <- FALSE
39 
+        .Object@singleCell <- FALSE
40 40 
         .Object@whichMatrixFixed <- "N"
41 41 
         .Object@checkpointInterval <- 0
42 42 
         .Object@checkpointOutFile <- "gaps_checkpoint.out"
... ... 
@@ -125,9 +125,9 @@ setGeneric("parseDirectParams", function(object, args)
125 125 
 setMethod("setParam", signature(object="CogapsParams"),
126 126 
     function(object, whichParam, value)
127 127 
     {
128 
-        slot(params, whichParam) <- value
129 
-        validObject(params)
130 
-        return(params)
128 
+        slot(object, whichParam) <- value
129 
+        validObject(object)
130 
+        return(object)
131 131 
     }
132 132 
 )
133 133
... ... 
@@ -136,7 +136,7 @@ setMethod("setParam", signature(object="CogapsParams"),
136 136 
 setMethod("getParam", signature(object="CogapsParams"),
137 137 
     function(object, whichParam)
138 138 
     {
139 
-        slot(params, whichParam)
139 
+        slot(object, whichParam)
140 140 
     }
141 141 
 )
142 142
... ... 
@@ -145,25 +145,40 @@ setMethod("getParam", signature(object="CogapsParams"),
145 145 
 setMethod("parseOldParams", signature(object="CogapsParams"),
146 146 
     function(object, oldArgs)
147 147 
     {
148 
-        if (!is.null(oldArgs$nFactor))
149 
-            params@nPatterns <- oldArgs$nFactor
150 
-        if (!is.null(oldArgs$nEquil))
151 
-            params@nIterations <- oldArgs$nEquil
152 
-        if (!is.null(oldArgs$nSample))
153 
-            params@nIterations <- oldArgs$nSample
154 
-        if (!is.null(oldArgs$nOutputs))
155 
-            params@outputFrequency <- nOutputs
156 
-        if (!is.null(oldArgs$maxGibbmassA))
157 
-            params@maxGibbsMassA <- oldArgs$maxGibbmassA
158 
-        if (!is.null(oldArgs$maxGibbmassP))
159 
-            params@maxGibbsMassA <- oldArgs$maxGibbmassP
148 
+        helper <- function(arg, params, newArg)
149 
+        {
150 
+            if (!is.null(oldArgs[[arg]]))
151 
+            {
152 
+                warning(paste("parameter", arg, "is deprecated, it will still",
153 
+                    "work, but setting", newArg, "in the params object is the",
154 
+                    "preferred method"))
155 
+                params <- setParam(params, newArg, oldArgs[[arg]])
156 
+                oldArgs[[arg]] <- NULL
157 
+            }
158 
+            return(params)
159 
+        }
160 160
161 
-        if (!is.null(oldArgs$nSnapshots))
161 
+        object <- helper("nFactor", object, "nPatterns")
162 
+        object <- helper("nIter", object, "nIterations")
163 
+        object <- helper("nEquil", object, "nIterations")
164 
+        object <- helper("nSample", object, "nIterations")
165 
+        object <- helper("nOutR", object, "outputFrequency")
166 
+        object <- helper("nOutput", object, "outputFrequency")
167 
+        object <- helper("maxGibbmassA", object, "maxGibbsMassA")
168 
+        object <- helper("max_gibbmass_paraA", object, "maxGibbsMassA")
169 
+        object <- helper("maxGibbmassP", object, "maxGibbsMassP")
170 
+        object <- helper("max_gibbmass_paraP", object, "maxGibbsMassP")
171 
+        object <- helper("checkpointFile", object, "checkpointOutFile")
172 
+        object <- helper("singleCellRNASeq", object, "singleCell")
173 
+
174 
+        if (!is.null(oldArgs$nSnapshots) | !is.null(oldArgs$sampleSnapshots) | !is.null(oldArgs$numSnapshots))
162 175 
             warning("snapshots not currently supported in release build")
163 176 
         if (!is.null(oldArgs$fixedPatterns))
164 177 
             stop("pass fixed matrix in with 'fixedMatrix' argument")
165 
-
166 
-        return(params)
178 
+        if (!is.null(oldArgs$S))
179 
+            stop("pass uncertainty matrix in with 'uncertainty', not 'S'")
180 
+
181 
+        return(object)
167 182 
     }
168 183 
 )
169 184
... ... 
@@ -174,11 +189,11 @@ setMethod("parseDirectParams", signature(object="CogapsParams"),
174 189 
     {
175 190 
         for (s in slotNames(object))
176 191 
         {
177 
-            if (!is.null(args[s]))
192 
+            if (!is.null(args[[s]]))
178 193 
             {
179 
-                params <- setParam(object, s, args[s])
194 
+                object <- setParam(object, s, args[[s]])
180 195 
             }
181 196 
         }
182 
-        return(params)
197 
+        return(object)
183 198 
     }
184 199 
 )
185 200 
\ No newline at end of file
R/scCoGAPS.R
History View file @2cdb0256a
... ... 
@@ -188,7 +188,7 @@ sc_runFinalPhase <- function(simulationName, allDataSets, consensusAs, nCores, .
188 188
189 189 
         # run CoGAPS with fixed patterns
190 190 
         cptFileName <- paste(simulationName, "_final_cpt_", i, ".out", sep="")
191 
-        CoGAPS(sampleD, sampleS, fixedPatterns=consensusAs,
191 
+        CoGAPS(sampleD, uncertainty=sampleS, fixedMatrix=consensusAs,
192 192 
             nFactor=nFactorFinal, seed=nut[i], checkpointFile=cptFileName,
193 193 
             whichMatrixFixed='A', singleCellRNASeq=TRUE, ...)
194 194
man/CoGAPS-methods.Rd
History View file @2cdb0256a
... ... 
@@ -1,33 +1,34 @@
1 1 
 % Generated by roxygen2: do not edit by hand
2 2 
 % Please edit documentation in R/CoGAPS.R
3 3 
 \docType{methods}
4 
-\name{CoGAPS Matrix Factorization Algorithm}
4 
+\name{CoGAPS}
5 5 
 \alias{CoGAPS}
6 
-\alias{CoGAPS Matrix Factorization Algorithm}
7 6 
 \alias{CoGAPS,SingleCellExperiment,CogapsParams-method}
8 7 
 \alias{CoGAPS,SummarizedExperiment,CogapsParams-method}
9 8 
 \alias{CoGAPS,character,CogapsParams-method}
10 9 
 \alias{CoGAPS,data.frame,CogapsParams-method}
11 10 
 \alias{CoGAPS,matrix,CogapsParams-method}
12 
-\title{CoGAPS}
13 
-\format{An object of class \code{NULL} of length 0.}
11 
+\title{CoGAPS Matrix Factorization Algorithm}
14 12 
 \usage{
13 
+CoGAPS(data, params = new("CogapsParams"), uncertainty = NULL,
14 
+  fixedMatrix = matrix(0), checkpointInFile = "", ...)
15 
+
15 16 
 \S4method{CoGAPS}{character,CogapsParams}(data, params = new("CogapsParams"),
16 
-  uncertainty = NULL, fixedMatrix = NULL, checkpointFile = NULL, ...)
17 
+  uncertainty = NULL, fixedMatrix = matrix(0), checkpointInFile = "", ...)
17 18
18 19 
 \S4method{CoGAPS}{matrix,CogapsParams}(data, params = new("CogapsParams"),
19 
-  uncertainty = NULL, fixedMatrix = NULL, checkpointFile = NULL, ...)
20 
+  uncertainty = NULL, fixedMatrix = matrix(0), checkpointInFile = "", ...)
20 21
21 22 
 \S4method{CoGAPS}{data.frame,CogapsParams}(data, params = new("CogapsParams"),
22 
-  uncertainty = NULL, fixedMatrix = NULL, checkpointFile = NULL, ...)
23 
+  uncertainty = NULL, fixedMatrix = matrix(0), checkpointInFile = "", ...)
23 24
24 25 
 \S4method{CoGAPS}{SummarizedExperiment,CogapsParams}(data,
25 
-  params = new("CogapsParams"), uncertainty = NULL, fixedMatrix = NULL,
26 
-  checkpointFile = NULL, ...)
26 
+  params = new("CogapsParams"), uncertainty = NULL,
27 
+  fixedMatrix = matrix(0), checkpointInFile = "", ...)
27 28
28 29 
 \S4method{CoGAPS}{SingleCellExperiment,CogapsParams}(data,
29 
-  params = new("CogapsParams"), uncertainty = NULL, fixedMatrix = NULL,
30 
-  checkpointFile = NULL, ...)
30 
+  params = new("CogapsParams"), uncertainty = NULL,
31 
+  fixedMatrix = matrix(0), checkpointInFile = "", ...)
31 32 
 }
32 33 
 \arguments{
33 34 
 \item{data}{File name or R object (see details for supported types)}
... ... 
@@ -39,7 +40,7 @@
39 40 
 \item{fixedMatrix}{data for fixing the values of either the A or P matrix;
40 41 
 used in conjuction with whichMatrixFixed (see CogapsParams)}
41 42
42 
-\item{checkpointFile}{name of the checkpoint file}
43 
+\item{checkpointInFile}{name of the checkpoint file}
43 44
44 45 
 \item{...}{keeps backwards compatibility with arguments from older versions}
45 46 
 }
... ... 
@@ -59,13 +60,14 @@ Currently, raw count matrices are the only supported R object. For
59 60 
 # Running from R object
60 61 
 data(GIST)
61 62 
 resultA <- CoGAPS(GIST.D)
63 
+
62 64 
 # Running from file name
63 65 
 gist_path <- system.file("extdata/GIST.mtx", package="CoGAPS")
64 66 
 resultB <- CoGAPS(gist_path)
67 
+
65 68 
 Setting Parameters
66 69 
 params <- new("CogapsParams")
67 70 
 params <- setParam(params, "nPatterns", 5)
68 71 
 resultC <- CoGAPS(GIST.D, params)
69 72 
 }
70 
-\keyword{datasets}
71 73
man/checkDataMatrix.Rd
History View file @2cdb0256a
72 74 
new file mode 100644
... ... 
@@ -0,0 +1,20 @@
1 
+% Generated by roxygen2: do not edit by hand
2 
+% Please edit documentation in R/CoGAPS.R
3 
+\name{checkDataMatrix}
4 
+\alias{checkDataMatrix}
5 
+\title{Check that provided data is valid}
6 
+\usage{
7 
+checkDataMatrix(data, uncertainty, params)
8 
+}
9 
+\arguments{
10 
+\item{data}{data matrix}
11 
+
12 
+\item{uncertainty}{uncertainty matrix}
13 
+}
14 
+\value{
15 
+throws an error if data has problems
16 
+}
17 
+\description{
18 
+Check that provided data is valid
19 
+}
20 
+
src/Archive.h
History View file @2cdb0256a
... ... 
@@ -59,6 +59,7 @@ public:
59 59 
     // don't have C++11 and don't want to add another dependency on boost,
60 60 
     // so no template tricks
61 61
62 
+    friend Archive& operator<<(Archive &ar, char val)     { return writeToArchive(ar, val); }
62 63 
     friend Archive& operator<<(Archive &ar, bool val)     { return writeToArchive(ar, val); }
63 64 
     friend Archive& operator<<(Archive &ar, int val)      { return writeToArchive(ar, val); }
64 65 
     friend Archive& operator<<(Archive &ar, unsigned val) { return writeToArchive(ar, val); }
... ... 
@@ -68,6 +69,7 @@ public:
68 69 
     friend Archive& operator<<(Archive &ar, double val)   { return writeToArchive(ar, val); }
69 70 
     friend Archive& operator<<(Archive &ar, boost::random::mt11213b val) { return writeToArchive(ar, val); }
70 71
72 
+    friend Archive& operator>>(Archive &ar, char &val)     { return readFromArchive(ar, val); }
71 73 
     friend Archive& operator>>(Archive &ar, bool &val)     { return readFromArchive(ar, val); }
72 74 
     friend Archive& operator>>(Archive &ar, int &val)      { return readFromArchive(ar, val); }
73 75 
     friend Archive& operator>>(Archive &ar, unsigned &val) { return readFromArchive(ar, val); }
... ... 
@@ -76,28 +78,6 @@ public:
76 78 
     friend Archive& operator>>(Archive &ar, float &val)    { return readFromArchive(ar, val); }
77 79 
     friend Archive& operator>>(Archive &ar, double &val)   { return readFromArchive(ar, val); }
78 80 
     friend Archive& operator>>(Archive &ar, boost::random::mt11213b &val) { return readFromArchive(ar, val); }
79 
-
80 
-/*
81 
-    friend Archive& operator>>(Archive &ar, uint64_t &val)
82 
-    {
83 
-        return readPrimitiveFromArchive(ar, val);
84 
-    }
85 
-
86 
-    template<typename T>
87 
-    friend Archive& operator<<(Archive &ar, T val)
88 
-    {
89 
-        ar.mStream.write(reinterpret_cast<char*>(&val), sizeof(T)); // NOLINT
90 
-        return ar;
91 
-    }
92 
-
93 
-
94 
-    template<typename T>
95 
-    friend Archive& operator>>(Archive &ar, T &val)
96 
-    {
97 
-        ar.mStream.read(reinterpret_cast<char*>(&val), sizeof(T)); // NOLINT
98 
-        return ar;
99 
-    }
100 
-*/
101 81 
 };
102 82
103 83 
 #endif
src/Cogaps.cpp
History View file @2cdb0256a
... ... 
@@ -32,73 +32,91 @@ static Rcpp::NumericMatrix createRMatrix(const Matrix &mat)
32 32 
     return rmat;
33 33 
 }
34 34
35 
-static bool nonNullUncertainty(const RowMatrix &mat)
35 
+static bool isNull(const RowMatrix &mat)
36 36 
 {
37 
-    return mat.nRow() > 1 || mat.nCol() > 1;
37 
+    return mat.nRow() == 1 && mat.nCol() == 1;
38 38 
 }
39 39
40 
-static bool nonNullUncertainty(const std::string &path)
40 
+static bool isNull(const std::string &path)
41 41 
 {
42 
-    return !path.empty();
42 
+    return path.empty();
43 43 
 }
44 44
45 45 
 template <class T>
46 
-static Rcpp::List cogapsRun(const T &data, const T &unc, unsigned nPatterns,
47 
-unsigned maxIter, unsigned outputFrequency, unsigned seed, float alphaA,
48 
-float alphaP, float maxGibbsMassA, float maxGibbsMassP, bool messages,
49 
-bool singleCell, unsigned nCores)
46 
+static Rcpp::List cogapsRun(const T &data, const Rcpp::S4 &params, const T &unc,
47 
+const RowMatrix &fixedMatrix, const std::string &checkpointInFile)
50 48 
 {
51 
-    GapsDispatcher dispatcher(seed);
49 
+    GapsDispatcher dispatcher;
52 50
53 
-    dispatcher.setNumPatterns(nPatterns);
54 
-    dispatcher.setMaxIterations(maxIter);
55 
-    dispatcher.setOutputFrequency(outputFrequency);
56 
-
57 
-    dispatcher.setAlpha(alphaA, alphaP);
58 
-    dispatcher.setMaxGibbsMass(maxGibbsMassA, maxGibbsMassP);
51 
+    // check if we're initializing with a checkpoint or not
52 
+    if (!isNull(checkpointInFile))
53 
+    {
54 
+        dispatcher.initialize(data, checkpointInFile);
55 
+    }
56 
+    else
57 
+    {
58 
+        dispatcher.initialize(data, params.slot("nPatterns"), params.slot("seed"));
59 59
60 
-    dispatcher.printMessages(messages);
61 
-    dispatcher.singleCell(singleCell);
62 
-    dispatcher.setNumCoresPerSet(nCores);
63 
-
64 
-    dispatcher.loadData(data);
60 
+        // set optional parameters
61 
+        dispatcher.setMaxIterations(params.slot("nIterations"));
62 
+        dispatcher.setOutputFrequency(params.slot("outputFrequency"));
63 
+
64 
+        dispatcher.setSparsity(params.slot("alphaA"), params.slot("alphaP"),
65 
+            params.slot("singleCell"));
66 
+
67 
+        dispatcher.setMaxGibbsMass(params.slot("maxGibbsMassA"),
68 
+            params.slot("maxGibbsMassP"));
65 69
66 
-    if (nonNullUncertainty(unc))
70 
+        dispatcher.printMessages(params.slot("messages"));
71 
+
72 
+        // check if running with a fixed matrix
73 
+        if (!isNull(fixedMatrix))
74 
+        {
75 
+            dispatcher.setFixedMatrix(params.slot("whichMatrixFixed"), fixedMatrix);
76 
+        }
77 
+    }
78 
+
79 
+    // set the uncertainty matrix
80 
+    if (!isNull(unc))
67 81 
     {
68 82 
         dispatcher.setUncertainty(unc);
69 83 
     }
84 
+
85 
+    // set parameters that aren't saved in the checkpoint
86 
+    dispatcher.setNumCoresPerSet(params.slot("nCores"));
87 
+    dispatcher.setCheckpointInterval(params.slot("checkpointInterval"));
88 
+    dispatcher.setCheckpointOutFile(params.slot("checkpointOutFile"));
70 89
90 
+    // run the dispatcher and return the GapsResult in an R list
71 91 
     GapsResult result(dispatcher.run());
92 
+    GAPS_ASSERT(result.meanChiSq > 0.f);
72 93 
     return Rcpp::List::create(
73 94 
         Rcpp::Named("Amean") = createRMatrix(result.Amean),
74 95 
         Rcpp::Named("Pmean") = createRMatrix(result.Pmean),
75 96 
         Rcpp::Named("Asd") = createRMatrix(result.Asd),
76 
-        Rcpp::Named("Psd") = createRMatrix(result.Psd)
97 
+        Rcpp::Named("Psd") = createRMatrix(result.Psd),
98 
+        Rcpp::Named("seed") = result.seed,
99 
+        Rcpp::Named("meanChiSq") = result.meanChiSq,
100 
+        Rcpp::Named("diagnostics") = Rcpp::List::create()
77 101 
     );
78 102 
 }
79 103
80 104 
 // [[Rcpp::export]]
81 
-Rcpp::List cogaps_cpp_from_file(const std::string &data, const std::string &unc,
82 
-unsigned nPatterns, unsigned maxIterations, unsigned outputFrequency,
83 
-uint32_t seed, float alphaA, float alphaP, float maxGibbsMassA,
84 
-float maxGibbsMassP, bool messages, bool singleCell,
85 
-const std::string &checkpointOutFile, unsigned nCores)
105 
+Rcpp::List cogaps_cpp_from_file(const std::string &data, const Rcpp::S4 &params,
106 
+const std::string &unc, const Rcpp::NumericMatrix &fixedMatrix,
107 
+const std::string &checkpointInFile)
86 108 
 {
87 
-    return cogapsRun(data, unc, nPatterns, maxIterations, outputFrequency, seed,
88 
-        alphaA, alphaP, maxGibbsMassA, maxGibbsMassP, messages, singleCell,
89 
-        nCores);
109 
+    return cogapsRun(data, params, unc, convertRMatrix(fixedMatrix),
110 
+        checkpointInFile);
90 111 
 }
91 112
92 113 
 // [[Rcpp::export]]
93 
-Rcpp::List cogaps_cpp(const Rcpp::NumericMatrix &data,
94 
-const Rcpp::NumericMatrix &unc, unsigned nPatterns, unsigned maxIterations,
95 
-unsigned outputFrequency, uint32_t seed, float alphaA, float alphaP,
96 
-float maxGibbsMassA, float maxGibbsMassP, bool messages, bool singleCell,
97 
-const std::string &checkpointOutFile, unsigned nCores)
114 
+Rcpp::List cogaps_cpp(const Rcpp::NumericMatrix &data, const Rcpp::S4 &params,
115 
+const Rcpp::NumericMatrix &unc, const Rcpp::NumericMatrix &fixedMatrix,
116 
+const std::string &checkpointInFile)
98 117 
 {
99 
-    return cogapsRun(convertRMatrix(data), convertRMatrix(unc), nPatterns,
100 
-        maxIterations, outputFrequency, seed, alphaA, alphaP, maxGibbsMassA,
101 
-        maxGibbsMassP, messages, singleCell, nCores);
118 
+    return cogapsRun(convertRMatrix(data), params, convertRMatrix(unc),
119 
+        convertRMatrix(fixedMatrix), checkpointInFile);
102 120 
 }
103 121
104 122 
 // [[Rcpp::export]]
src/GapsDispatcher.cpp
History View file @2cdb0256a
... ... 
@@ -6,37 +6,72 @@
6 6 
     #include <omp.h>
7 7 
 #endif
8 8
9 
-/*
10 
-static std::vector< std::vector<unsigned> > sampleIndices(unsigned n, unsigned nSets)
9 
+GapsDispatcher::GapsDispatcher() : mSeed(0), mNumPatterns(3),
10 
+    mMaxIterations(1000), mNumCoresPerSet(1), mPrintMessages(true),
11 
+    mCheckpointsCreated(0), mPhase('C'), mCheckpointInterval(0),
12 
+    mCheckpointOutFile("gaps_checkpoint.out"), mInitialized(false)
13 
+{}
14 
+
15 
+GapsDispatcher::~GapsDispatcher()
11 16 
 {
12 
-    unsigned setSize = n / nSets;
13 
-    std::vector< std::vector<unsigned> > sampleIndices;
14 
-    std::vector<unsigned> toBeSampled;
15 
-    for (unsigned i = 0; i < n; ++i)
17 
+    for (unsigned i = 0; i < mRunners.size(); ++i)
16 18 
     {
17 
-        toBeSampled.push_back(i);
19 
+        delete mRunners[i];
18 20 
     }
21 
+}
19 22
20 
-    for (unsigned i = 0; i < (nSets - 1); ++i)
21 
-    {
22 
-        sampleIndices.push_back(gaps::random::sample(toBeSampled, setSize));
23 
-    }
23 
+void GapsDispatcher::setMaxIterations(unsigned n)
24 
+{
25 
+    mMaxIterations = n;
26 
+}
24 27
25 
-    GAPS_ASSERT(!toBeSampled.empty());
28 
+void GapsDispatcher::printMessages(bool print)
29 
+{
30 
+    mPrintMessages = print;
31 
+    mRunners[0]->printMessages(print);
32 
+}
26 33
27 
-    sampleIndices.push_back(toBeSampled);
28 
-    return sampleIndices;
34 
+void GapsDispatcher::setOutputFrequency(unsigned n)
35 
+{
36 
+    mRunners[0]->setOutputFrequency(n);
29 37 
 }
30 
-*/
31 38
32 
-void GapsDispatcher::runOneCycle(unsigned k)
39 
+void GapsDispatcher::setSparsity(float alphaA, float alphaP, bool singleCell)
40 
+{
41 
+    mRunners[0]->setSparsity(alphaA, alphaP, singleCell);
42 
+}
43 
+
44 
+void GapsDispatcher::setMaxGibbsMass(float maxA, float maxP)
45 
+{
46 
+    mRunners[0]->setMaxGibbsMass(maxA, maxP);
47 
+}
48 
+
49 
+void GapsDispatcher::setFixedMatrix(char which, const RowMatrix &mat)
50 
+{
51 
+    mRunners[0]->setFixedMatrix(which, mat);
52 
+}
53 
+
54 
+void GapsDispatcher::setNumCoresPerSet(unsigned n)
55 
+{
56 
+    mNumCoresPerSet = n;
57 
+}
58 
+
59 
+void GapsDispatcher::setCheckpointInterval(unsigned n)
33 60 
 {
34 
-    GAPS_ASSERT(mDataIsLoaded);
35 
-    mRunners[0]->run(k, mOutputFrequency, mPrintMessages, mNumCoresPerSet);
61 
+    mCheckpointInterval = n;
62 
+}
63 
+
64 
+void GapsDispatcher::setCheckpointOutFile(const std::string &path)
65 
+{
66 
+    mCheckpointOutFile = path;
36 67 
 }
37 68
38 69 
 GapsResult GapsDispatcher::run()
39 70 
 {
71 
+    GAPS_ASSERT(mInitialized);
72 
+    GAPS_ASSERT(mPhase == 'C' || mPhase == 'S');
73 
+
74 
+    // calculate appropiate number of cores if compiled with openmp
40 75 
     #ifdef __GAPS_OPENMP__
41 76 
     if (mPrintMessages)
42 77 
     {
... ... 
@@ -47,55 +82,81 @@ GapsResult GapsDispatcher::run()
47 82 
     }
48 83 
     #endif
49 84
50 
-    GapsResult result(mRunners[0]->nRow(), mRunners[0]->nCol());
51 
-    GAPS_ASSERT(mDataIsLoaded);
52 
-
53 
-    if (mPrintMessages)
85 
+    // this switch allows for the algorithm to be interruptable
86 
+    switch (mPhase)
54 87 
     {
55 
-        gaps_printf("Calibration Phase\n");
88 
+        case 'C':
89 
+            if (mPrintMessages)
90 
+            {
91 
+                gaps_printf("Calibration Phase\n");
92 
+            }
93 
+            runOneCycle(mMaxIterations);
94 
+            mPhase = 'S';
95 
+
96 
+        case 'S':
97 
+            if (mPrintMessages)
98 
+            {
99 
+                gaps_printf("Sampling Phase\n");
100 
+            }
101 
+            mRunners[0]->startSampling();
102 
+            runOneCycle(mMaxIterations);
103 
+            break;
56 104 
     }
57 
-    runOneCycle(mMaxIterations);
58 105
59 
-    if (mPrintMessages)
60 
-    {
61 
-        gaps_printf("Sampling Phase\n");
62 
-    }
63 
-    mRunners[0]->startSampling();
64 
-    runOneCycle(mMaxIterations);
65 
-
66 
-    result.Amean = mRunners[0]->AMean();
67 
-    result.Pmean = mRunners[0]->PMean();
68 
-    result.Asd = mRunners[0]->AStd();
69 
-    result.Psd = mRunners[0]->PStd();
70 
-    result.seed = mSeed;
106 
+    // extract useful information from runners
107 
+    GapsResult result(mRunners[0]->nRow(), mRunners[0]->nCol(), mSeed);
108 
+    result.Amean = mRunners[0]->Amean();
109 
+    result.Pmean = mRunners[0]->Pmean();
110 
+    result.Asd = mRunners[0]->Asd();
111 
+    result.Psd = mRunners[0]->Psd();
112 
+    result.meanChiSq = mRunners[0]->meanChiSq();
71 113 
     return result;
72 114 
 }
73 115
74 
-void GapsDispatcher::loadData(const RowMatrix &D)
116 
+void GapsDispatcher::createCheckpoint() const
75 117 
 {
76 
-    mRunners.push_back(new GapsRunner(D, mNumPatterns, mAlphaA,
77 
-        mAlphaP, mMaxGibbsMassA, mMaxGibbsMassP, mSingleCell));
78 
-    mDataIsLoaded = true;
118 
+    Archive ar(mCheckpointOutFile, ARCHIVE_WRITE);
119 
+
120 
+    gaps::random::save(ar);
121 
+    ar << mSeed << mNumPatterns << mMaxIterations << mPrintMessages <<
122 
+        mCheckpointsCreated << mPhase;
123 
+
124 
+    ar << *mRunners[0];
79 125 
 }
80 126
81 
-void GapsDispatcher::setUncertainty(const RowMatrix &S)
127 
+void GapsDispatcher::runOneCycle(unsigned k)
82 128 
 {
83 
-    GAPS_ASSERT(mDataIsLoaded);
84 
-    mRunners[0]->setUncertainty(S);
129 
+    unsigned nCheckpoints = mCheckpointInterval > 0 ? k / mCheckpointInterval : 0;
130 
+    while (mCheckpointsCreated < nCheckpoints)
131 
+    {
132 
+        mRunners[0]->run(mCheckpointInterval, mNumCoresPerSet);
133 
+        createCheckpoint();
134 
+        ++mCheckpointsCreated;
135 
+    }
136 
+    mRunners[0]->run(k - mCheckpointInterval * mCheckpointsCreated, mNumCoresPerSet);
137 
+    mCheckpointsCreated = 0; // reset checkpoint count for next cycle
85 138 
 }
86 139
87 
-void GapsDispatcher::loadData(const std::string &pathToData)
140 
+/*
141 
+static std::vector< std::vector<unsigned> > sampleIndices(unsigned n, unsigned nSets)
88 142 
 {
89 
-    gaps_printf("Loading Data...");
90 
-    gaps_flush();
91 
-    mRunners.push_back(new GapsRunner(pathToData, mNumPatterns, mAlphaA,
92 
-        mAlphaP, mMaxGibbsMassA, mMaxGibbsMassP, mSingleCell));
93 
-    mDataIsLoaded = true;
94 
-    gaps_printf("Done!\n");
143 
+    unsigned setSize = n / nSets;
144 
+    std::vector< std::vector<unsigned> > sampleIndices;
145 
+    std::vector<unsigned> toBeSampled;
146 
+    for (unsigned i = 0; i < n; ++i)
147 
+    {
148 
+        toBeSampled.push_back(i);
149 
+    }
150 
+
151 
+    for (unsigned i = 0; i < (nSets - 1); ++i)
152 
+    {
153 
+        sampleIndices.push_back(gaps::random::sample(toBeSampled, setSize));
154 
+    }
155 
+
156 
+    GAPS_ASSERT(!toBeSampled.empty());
157 
+
158 
+    sampleIndices.push_back(toBeSampled);
159 
+    return sampleIndices;
95 160 
 }
161 
+*/
96 162
97 
-void GapsDispatcher::setUncertainty(const std::string &pathToMatrix)
98 
-{
99 
-    GAPS_ASSERT(mDataIsLoaded);
100 
-    mRunners[0]->setUncertainty(pathToMatrix);
101 
-}
102 163 
\ No newline at end of file
src/GapsDispatcher.h
History View file @2cdb0256a
... ... 
@@ -16,12 +16,14 @@ struct GapsResult
16 16 
     float meanChiSq;
17 17 
     uint32_t seed;
18 18
19 
-    GapsResult(unsigned nrow, unsigned ncol) : Amean(nrow, ncol),
20 
-        Asd(nrow, ncol), Pmean(nrow, ncol), Psd(nrow, ncol), meanChiSq(0.f),
21 
-        seed(0)
19 
+    GapsResult(unsigned nrow, unsigned ncol, uint32_t rngSeed) :
20 
+        Amean(nrow, ncol), Asd(nrow, ncol), Pmean(nrow, ncol), Psd(nrow, ncol),
21 
+        meanChiSq(0.f), seed(rngSeed)
22 22 
     {}
23 23
24 24 
     void writeCsv(const std::string &path);
25 
+    void writeTsv(const std::string &path);
26 
+    void writeGct(const std::string &path);
25 27 
 };
26 28
27 29 
 // should be agnostic to external caller (R/Python/CLI)
... ... 
@@ -29,86 +31,99 @@ class GapsDispatcher
29 31 
 {
30 32 
 private:
31 33
32 
-    unsigned mNumPatterns;
33 
-    unsigned mMaxIterations;
34 
-    unsigned mOutputFrequency;
35 34 
     uint32_t mSeed;
35 
+    unsigned mNumPatterns;
36 36
37 
-    float mAlphaA;
38 
-    float mAlphaP;
39 
-    float mMaxGibbsMassA;
40 
-    float mMaxGibbsMassP;
41 
-
37 
+    unsigned mMaxIterations;
38 
+    unsigned mNumCoresPerSet;
42 39 
     bool mPrintMessages;
43 
-    bool mSingleCell;
44 40
45 
-    bool mDataIsLoaded;
41 
+    unsigned mCheckpointsCreated;
42 
+    char mPhase; // 'C' for calibration, 'S' for sample
46 43
47 
-    unsigned mNumCoresPerSet;
44 
+    unsigned mCheckpointInterval;
45 
+    std::string mCheckpointOutFile;
48 46
47 
+    bool mInitialized;
49 48 
     std::vector<GapsRunner*> mRunners;
50 49
51 
-    char mFixedMatrix;
52 
-
53 50 
     void runOneCycle(unsigned k);
51 
+    void createCheckpoint() const;
54 52
55 53 
     GapsDispatcher(const GapsDispatcher &p); // don't allow copies
56 54 
     GapsDispatcher& operator=(const GapsDispatcher &p); // don't allow copies
57 55
56 
+    template <class DataType>
57 
+    void loadData(const DataType &data);
58 
+
58 59 
 public:
59 60
60 
-    explicit GapsDispatcher(uint32_t seed=0) : mNumPatterns(3),
61 
-        mMaxIterations(1000), mOutputFrequency(250), mSeed(seed), mAlphaA(0.01),
62 
-        mAlphaP(0.01), mMaxGibbsMassA(100.f), mMaxGibbsMassP(100.f),
63 
-        mPrintMessages(true), mSingleCell(false), mDataIsLoaded(false),
64 
-        mNumCoresPerSet(1), mFixedMatrix('N')
65 
-    {
66 
-        gaps::random::setSeed(mSeed);
67 
-    }
68 
-
69 
-    ~GapsDispatcher()
70 
-    {
71 
-        for (unsigned i = 0; i < mRunners.size(); ++i)
72 
-        {
73 
-            delete mRunners[i];
74 
-        }
75 
-    }
76 
-
77 
-    void setNumPatterns(unsigned n)     { mNumPatterns = n; }
78 
-    void setMaxIterations(unsigned n)   { mMaxIterations = n; }
79 
-    void setOutputFrequency(unsigned n) { mOutputFrequency = n; }
80 
-    void setNumCoresPerSet(unsigned n)  { mNumCoresPerSet = n; }
81 
-
82 
-    void printMessages(bool print) { mPrintMessages = print; }
83 
-    void singleCell(bool sc) { mSingleCell = sc; }
84 
-
85 
-    void setAlpha(float alphaA, float alphaP)
86 
-    {
87 
-        mAlphaA = alphaA;
88 
-        mAlphaP = alphaP;
89 
-    }
90 
-
91 
-    void setMaxGibbsMass(float maxA, float maxP)
92 
-    {
93 
-        mMaxGibbsMassA = maxA;
94 
-        mMaxGibbsMassP = maxP;
95 
-    }
96 
-
97 
-    void setFixedMatrix(char which, const RowMatrix &mat)
98 
-    {
99 
-        mFixedMatrix = which;
100 
-        mRunners[0]->setFixedMatrix(which, mat);
101 
-    }
102 
-
103 
-    void loadCheckpointFile(const std::string &pathToCptFile);
61 
+    GapsDispatcher();
62 
+    ~GapsDispatcher();
63 
+
64 
+    template <class DataType>
65 
+    void initialize(const DataType &data, unsigned nPatterns, uint32_t seed=0);
66 
+
67 
+    template <class DataType>
68 
+    void initialize(const DataType &data, const std::string &cptFile);
104 69
105 
-    void setUncertainty(const RowMatrix &S);
106 
-    void setUncertainty(const std::string &pathToMatrix);
70 
+    template <class DataType>
71 
+    void setUncertainty(const DataType &unc);
107 72
108 
-    void loadData(const RowMatrix &D);
109 
-    void loadData(const std::string &pathToData);
73 
+    void setMaxIterations(unsigned n);
74 
+    void printMessages(bool print);
75 
+    void setOutputFrequency(unsigned n);
76 
+    void setSparsity(float alphaA, float alphaP, bool singleCell);
77 
+    void setMaxGibbsMass(float maxA, float maxP);
78 
+    void setFixedMatrix(char which, const RowMatrix &mat);
79 
+
80 
+    void setNumCoresPerSet(unsigned n);
81 
+    void setCheckpointInterval(unsigned n);
82 
+    void setCheckpointOutFile(const std::string &path);
110 83
111 84 
     GapsResult run();
112 85 
 };
113 86
87 
+template <class DataType>
88 
+void GapsDispatcher::initialize(const DataType &data, unsigned nPatterns,
89 
+uint32_t seed)
90 
+{
91 
+    mSeed = seed;
92 
+    mNumPatterns = nPatterns;
93 
+    gaps::random::setSeed(mSeed);
94 
+
95 
+    loadData(data);
96 
+    mInitialized = true;
97 
+}
98 
+
99 
+template <class DataType>
100 
+void GapsDispatcher::initialize(const DataType &data, const std::string &cptFile)
101 
+{
102 
+    Archive ar(cptFile, ARCHIVE_READ);
103 
+    gaps::random::load(ar);
104 
+
105 
+    ar >> mSeed >> mNumPatterns >> mMaxIterations >> mPrintMessages >>
106 
+        mCheckpointsCreated >> mPhase;
107 
+
108 
+    loadData(data);
109 
+    ar >> *mRunners[0];
110 
+    mInitialized = true;
111 
+}
112 
+
113 
+template <class DataType>
114 
+void GapsDispatcher::setUncertainty(const DataType &unc)
115 
+{
116 
+    GAPS_ASSERT(mInitialized);
117 
+    mRunners[0]->setUncertainty(unc);
118 
+}
119 
+
120 
+template <class DataType>
121 
+void GapsDispatcher::loadData(const DataType &data)
122 
+{
123 
+    gaps_printf("Loading Data...");
124 
+    gaps_flush();
125 
+    mRunners.push_back(new GapsRunner(data, mNumPatterns));
126 
+    gaps_printf("Done!\n");
127 
+}
128 
+
114 129 
 #endif // __COGAPS_GAPS_DISPATCHER_H__
115 130 
\ No newline at end of file
src/GapsRunner.cpp
History View file @2cdb0256a
... ... 
@@ -8,64 +8,51 @@
8 8 
 #include <Rcpp.h>
9 9 
 #endif
10 10
11 
-#include <algorithm>
11 
+void GapsRunner::printMessages(bool print)
12 
+{
13 
+    mPrintMessages = print;
14 
+}
12 15
13 
-GapsRunner::GapsRunner(const RowMatrix &data, unsigned nPatterns, float alphaA,
14 
-float alphaP, float maxGibbsMassA, float maxGibbsMassP, bool singleCell)
15 
-    :
16 
-mNumRows(data.nRow()), mNumCols(data.nCol()),
17 
-mASampler(data, nPatterns, alphaA, maxGibbsMassA, singleCell),
18 
-mPSampler(data, nPatterns, alphaP, maxGibbsMassP, singleCell),
19 
-mStatistics(data.nRow(), data.nCol(), nPatterns),
20 
-mSamplePhase(false), mNumUpdatesA(0), mNumUpdatesP(0)
16 
+void GapsRunner::setOutputFrequency(unsigned n)
21 17 
 {
22 
-    mASampler.sync(mPSampler);
23 
-    mPSampler.sync(mASampler);
18 
+    mOutputFrequency = n;
24 19 
 }
25 20
26 
-GapsRunner::GapsRunner(const std::string &pathToData, unsigned nPatterns,
27 
-float alphaA, float alphaP, float maxGibbsMassA, float maxGibbsMassP,
28 
-bool singleCell)
29 
-    :
30 
-mNumRows(FileParser(pathToData).nRow()), mNumCols(FileParser(pathToData).nCol()),
31 
-mASampler(pathToData, nPatterns, alphaA, maxGibbsMassA, singleCell),
32 
-mPSampler(pathToData, nPatterns, alphaP, maxGibbsMassP, singleCell),
33 
-mStatistics(mNumRows, mNumCols, nPatterns),
34 
-mSamplePhase(false), mNumUpdatesA(0), mNumUpdatesP(0)
21 
+void GapsRunner::setSparsity(float alphaA, float alphaP, bool singleCell)
35 22 
 {
36 
-    mASampler.sync(mPSampler);
37 
-    mPSampler.sync(mASampler);
23 
+    mASampler.setSparsity(alphaA, singleCell);
24 
+    mPSampler.setSparsity(alphaP, singleCell);
25 
+}
26 
+
27 
+void GapsRunner::setMaxGibbsMass(float maxA, float maxP)
28 
+{
29 
+    mASampler.setMaxGibbsMass(maxA);
30 
+    mPSampler.setMaxGibbsMass(maxP);
38 31 
 }
39 32
40 33 
 void GapsRunner::setFixedMatrix(char which, const RowMatrix &mat)
41 34 
 {
35 
+    mFixedMatrix = which;
42 36 
     if (which == 'A')
43 37 
     {
44 38 
         mASampler.setMatrix(ColMatrix(mat));
39 
+        mASampler.recalculateAPMatrix();
40 
+        mPSampler.sync(mASampler);
45 41 
     }
46 42 
     else if (which == 'P')
47 43 
     {
48 44 
         mPSampler.setMatrix(mat);
45 
+        mPSampler.recalculateAPMatrix();
46 
+        mASampler.sync(mPSampler);
49 47 
     }
50 48 
 }
51 49
52 
-void GapsRunner::setUncertainty(const std::string &pathToMatrix)
50 
+void GapsRunner::startSampling()
53 51 
 {
54 
-    mASampler.setUncertainty(pathToMatrix);
55 
-    mPSampler.setUncertainty(pathToMatrix);
52 
+    mSamplePhase = true;
56 53 
 }
57 54
58 
-void GapsRunner::displayStatus(unsigned outFreq, unsigned current, unsigned total)
59 
-{
60 
-    if (outFreq > 0 && ((current + 1) % outFreq) == 0)
61 
-    {
62 
-        gaps_printf("%d of %d, Atoms:%lu(%lu) Chi2 = %.2f\n", current + 1,
63 
-            total, mASampler.nAtoms(), mPSampler.nAtoms(), mASampler.chi2());
64 
-    }
65 
-}
66 
-
67 
-void GapsRunner::run(unsigned nIter, unsigned outputFreq, bool printMessages,
68 
-unsigned nCores)
55 
+void GapsRunner::run(unsigned nIter, unsigned nCores)
69 56 
 {
70 57 
     for (unsigned i = 0; i < nIter; ++i)
71 58 
     {
... ... 
@@ -85,9 +72,9 @@ unsigned nCores)
85 72 
         unsigned nP = gaps::random::poisson(std::max(mPSampler.nAtoms(), 10ul));
86 73 
         updateSampler(nA, nP, nCores);
87 74
88 
-        if (printMessages)
75 
+        if (mPrintMessages)
89 76 
         {
90 
-            displayStatus(outputFreq, i, nIter);
77 
+            displayStatus(i, nIter);
91 78 
         }
92 79
93 80 
         if (mSamplePhase)
... ... 
@@ -97,19 +84,87 @@ unsigned nCores)
97 84 
     }
98 85 
 }
99 86
87 
+unsigned GapsRunner::nRow() const
88 
+{
89 
+    return mASampler.dataRows();
90 
+}
91 
+
92 
+unsigned GapsRunner::nCol() const
93 
+{
94 
+    return mASampler.dataCols();
95 
+}
96 
+
97 
+ColMatrix GapsRunner::Amean() const
98 
+{
99 
+    return mStatistics.Amean();
100 
+}
101 
+
102 
+RowMatrix GapsRunner::Pmean() const
103 
+{
104 
+    return mStatistics.Pmean();
105 
+}
106 
+
107 
+ColMatrix GapsRunner::Asd() const
108 
+{
109 
+    return mStatistics.Asd();
110 
+}
111 
+
112 
+RowMatrix GapsRunner::Psd() const
113 
+{
114 
+    return mStatistics.Psd();
115 
+}
116 
+
117 
+float GapsRunner::meanChiSq() const
118 
+{
119 
+    return mStatistics.meanChiSq(mASampler);
120 
+}
121 
+
122 
+Archive& operator<<(Archive &ar, GapsRunner &runner)
123 
+{
124 
+    ar << runner.mASampler << runner.mPSampler << runner.mStatistics <<
125 
+        runner.mPrintMessages << runner.mOutputFrequency <<
126 
+        runner.mFixedMatrix << runner.mSamplePhase << runner.mNumUpdatesA <<
127 
+        runner.mNumUpdatesP;
128 
+    return ar;
129 
+}
130 
+
131 
+Archive& operator>>(Archive &ar, GapsRunner &runner)
132 
+{
133 
+    ar >> runner.mASampler >> runner.mPSampler >> runner.mStatistics >>
134 
+        runner.mPrintMessages >> runner.mOutputFrequency >>
135 
+        runner.mFixedMatrix >> runner.mSamplePhase >> runner.mNumUpdatesA >>
136 
+        runner.mNumUpdatesP;
137 
+    return ar;
138 
+}
139 
+
100 140 
 void GapsRunner::updateSampler(unsigned nA, unsigned nP, unsigned nCores)
101 141 
 {
102 
-    mNumUpdatesA += nA;
103 
-    mASampler.update(nA, nCores);
104 
-    mPSampler.sync(mASampler);
142 
+    if (mFixedMatrix != 'A')
143 
+    {
144 
+        mNumUpdatesA += nA;
145 
+        mASampler.update(nA, nCores);
146 
+        if (mFixedMatrix != 'P')
147 
+        {
148 
+            mPSampler.sync(mASampler);
149 
+        }
150 
+    }
105 151
106 
-    mNumUpdatesP += nP;
107 
-    mPSampler.update(nP, nCores);
108 
-    mASampler.sync(mPSampler);
152 
+    if (mFixedMatrix != 'P')
153 
+    {
154 
+        mNumUpdatesP += nP;
155 
+        mPSampler.update(nP, nCores);
156 
+        if (mFixedMatrix != 'A')
157 
+        {
158 
+            mASampler.sync(mPSampler);
159 
+        }
160 
+    }
109 161 
 }
110 162
111 
-void GapsRunner::setUncertainty(const RowMatrix &S)
163 
+void GapsRunner::displayStatus(unsigned current, unsigned total)
112 164 
 {
113 
-    mASampler.setUncertainty(S);
114 
-    mPSampler.setUncertainty(S);
165 
+    if (mOutputFrequency > 0 && ((current + 1) % mOutputFrequency) == 0)
166 
+    {
167 
+        gaps_printf("%d of %d, Atoms:%lu(%lu) Chi2 = %.2f\n", current + 1,
168 
+            total, mASampler.nAtoms(), mPSampler.nAtoms(), mASampler.chi2());
169 
+    }
115 170 
 }
src/GapsRunner.h
History View file @2cdb0256a
... ... 
@@ -9,50 +9,71 @@
9 9 
 class GapsRunner
10 10 
 {
11 11 
 private:
12 
-
13 
-    unsigned mNumRows;
14 
-    unsigned mNumCols;
15 
-
12 
+
16 13 
     AmplitudeGibbsSampler mASampler;
17 14 
     PatternGibbsSampler mPSampler;
18 15 
     GapsStatistics mStatistics;
19 16
17 
+    bool mPrintMessages;
18 
+    unsigned mOutputFrequency;
19 
+    char mFixedMatrix;
20 20 
     bool mSamplePhase;
21 21
22 22 
     unsigned mNumUpdatesA;
23 23 
     unsigned mNumUpdatesP;
24 24
25 25 
     void updateSampler(unsigned nA, unsigned nP, unsigned nCores);
26 
+    void displayStatus(unsigned current, unsigned total);
26 27
27 28 
 public:
28 29
29 
-    GapsRunner(const RowMatrix &data, unsigned nPatterns, float alphaA,
30 
-        float alphaP, float maxGibbsMassA, float maxGibbsMassP,
31 
-        bool singleCell);
30 
+    template <class DataType>
31 
+    GapsRunner(const DataType &data, unsigned nPatterns);
32 32
33 
-    GapsRunner(const std::string &pathToData, unsigned nPatterns, float alphaA,
34 
-        float alphaP, float maxGibbsMassA, float maxGibbsMassP,
35 
-        bool singleCell);
36 
-
37 
-    void run(unsigned nIter, unsigned outputFreq, bool printMessages,
38 
-        unsigned nCores);
33 
+    template <class DataType>
34 
+    void setUncertainty(const DataType &unc);
39 35
40 
-    unsigned nRow() const { return mNumRows; }
41 
-    unsigned nCol() const { return mNumCols; }
36 
+    void printMessages(bool print);
37 
+    void setOutputFrequency(unsigned n);
38 
+    void setSparsity(float alphaA, float alphaP, bool singleCell);
39 
+    void setMaxGibbsMass(float maxA, float maxP);
42 40
43 
-    void setUncertainty(const RowMatrix &S);
44 
-    void setUncertainty(const std::string &pathToMatrix);
41 
+    void setFixedMatrix(char which, const RowMatrix &mat);
45 42
46 
-    void startSampling() { mSamplePhase = true; }
43 
+    void startSampling();
47 44
48 
-    void displayStatus(unsigned outFreq, unsigned current, unsigned total);
45 
+    void run(unsigned nIter, unsigned nCores);
49 46
50 
-    void setFixedMatrix(char which, const RowMatrix &mat);
47 
+    unsigned nRow() const;
48 
+    unsigned nCol() const;
51 49
52 
-    ColMatrix AMean() const { return mStatistics.AMean(); }
53 
-    RowMatrix PMean() const { return mStatistics.PMean(); }
54 
-    ColMatrix AStd() const { return mStatistics.AStd(); }
55 
-    RowMatrix PStd() const { return mStatistics.PStd(); }
50 
+    ColMatrix Amean() const;
51 
+    RowMatrix Pmean() const;
52 
+    ColMatrix Asd() const;
53 
+    RowMatrix Psd() const;
54 
+    float meanChiSq() const;
55 
+
56 
+    // serialization
57 
+    friend Archive& operator<<(Archive &ar, GapsRunner &runner);
58 
+    friend Archive& operator>>(Archive &ar, GapsRunner &runner);
56 59 
 };
57 60
61 
+template <class DataType>
62 
+GapsRunner::GapsRunner(const DataType &data, unsigned nPatterns)
63 
+    :
64 
+mASampler(data, nPatterns), mPSampler(data, nPatterns),
65 
+mStatistics(mASampler.dataRows(), mPSampler.dataCols(), nPatterns),
66 
+mSamplePhase(false), mNumUpdatesA(0), mNumUpdatesP(0), mFixedMatrix('N')
67 
+{
68 
+    mASampler.sync(mPSampler);
69 
+    mPSampler.sync(mASampler);
70 
+}
71 
+
72 
+template <class DataType>
73 
+void GapsRunner::setUncertainty(const DataType &unc)
74 
+{
75 
+    mASampler.setUncertainty(unc);
76 
+    mPSampler.setUncertainty(unc);
77 
+}
78 
+
58 79 
 #endif // __COGAPS_GAPS_RUNNER_H__
59 80 
\ No newline at end of file
src/GapsStatistics.cpp
History View file @2cdb0256a
... ... 
@@ -1,11 +1,13 @@
1 1 
 #include "GapsStatistics.h"
2 2 
 #include "math/Algorithms.h"
3 3
4 
-GapsStatistics::GapsStatistics(unsigned nRow, unsigned nCol, unsigned nFactor, PumpThreshold t)
5 
-    : mAMeanMatrix(nRow, nFactor), mAStdMatrix(nRow, nFactor),
6 
-        mPMeanMatrix(nFactor, nCol), mPStdMatrix(nFactor, nCol),
7 
-        mStatUpdates(0), mNumPatterns(nFactor), mPumpMatrix(nRow, nCol),
8 
-        mPumpThreshold(t), mPumpStatUpdates(0)
4 
+GapsStatistics::GapsStatistics(unsigned nRow, unsigned nCol, unsigned nPatterns,
5 
+PumpThreshold t)
6 
+    :
7 
+mAMeanMatrix(nRow, nPatterns), mAStdMatrix(nRow, nPatterns),
8 
+mPMeanMatrix(nPatterns, nCol), mPStdMatrix(nPatterns, nCol),
9 
+mStatUpdates(0), mNumPatterns(nPatterns), mPumpMatrix(nRow, nCol),
10 
+mPumpThreshold(t), mPumpStatUpdates(0)
9 11 
 {}
10 12
11 13 
 void GapsStatistics::update(const AmplitudeGibbsSampler &ASampler,
... ... 
@@ -29,6 +31,37 @@ const PatternGibbsSampler &PSampler)
29 31 
     }
30 32 
 }
31 33
34 
+ColMatrix GapsStatistics::Amean() const
35 
+{
36 
+    return mAMeanMatrix / mStatUpdates;
37 
+}
38 
+
39 
+ColMatrix GapsStatistics::Asd() const
40 
+{
41 
+    return gaps::algo::computeStdDev(mAStdMatrix, mAMeanMatrix,
42 
+        mStatUpdates);
43 
+}
44 
+
45 
+RowMatrix GapsStatistics::Pmean() const
46 
+{
47 
+    return mPMeanMatrix / mStatUpdates;
48 
+}
49 
+
50 
+RowMatrix GapsStatistics::Psd() const
51 
+{
52 
+    return gaps::algo::computeStdDev(mPStdMatrix, mPMeanMatrix,
53 
+        mStatUpdates);
54 
+}
55 
+
56 
+float GapsStatistics::meanChiSq(const AmplitudeGibbsSampler &ASampler) const
57 
+{
58 
+    ColMatrix A = mAMeanMatrix / mStatUpdates;
59 
+    RowMatrix P = mPMeanMatrix / mStatUpdates;
60 
+    RowMatrix M(gaps::algo::matrixMultiplication(A, P));
61 
+    return 2.f * gaps::algo::loglikelihood(ASampler.mDMatrix, ASampler.mSMatrix,
62 
+        M);
63 
+}
64 
+
32 65 
 static unsigned geneThreshold(const ColMatrix &rankMatrix, unsigned pat)
33 66 
 {
34 67 
     float cutRank = rankMatrix.nRow();
... ... 
@@ -127,37 +160,6 @@ const PatternGibbsSampler &PSampler)
127 160 
     patternMarkers(ASampler.mMatrix, PSampler.mMatrix, mPumpMatrix);
128 161 
 }
129 162
130 
-ColMatrix GapsStatistics::AMean() const
131 
-{
132 
-    return mAMeanMatrix / mStatUpdates;
133 
-}
134 
-
135 
-ColMatrix GapsStatistics::AStd() const
136 
-{
137 
-    return gaps::algo::computeStdDev(mAStdMatrix, mAMeanMatrix,
138 
-        mStatUpdates);
139 
-}
140 
-
141 
-RowMatrix GapsStatistics::PMean() const
142 
-{
143 
-    return mPMeanMatrix / mStatUpdates;
144 
-}
145 
-
146 
-RowMatrix GapsStatistics::PStd() const
147 
-{
148 
-    return gaps::algo::computeStdDev(mPStdMatrix, mPMeanMatrix,
149 
-        mStatUpdates);
150 
-}
151 
-
152 
-float GapsStatistics::meanChiSq(const AmplitudeGibbsSampler &ASampler) const
153 
-{
154 
-    ColMatrix A = mAMeanMatrix / mStatUpdates;
155 
-    RowMatrix P = mPMeanMatrix / mStatUpdates;
156 
-    RowMatrix M(gaps::algo::matrixMultiplication(A, P));
157 
-    return 2.f * gaps::algo::loglikelihood(ASampler.mDMatrix, ASampler.mSMatrix,
158 
-        M);
159 
-}
160 
-
161 163 
 RowMatrix GapsStatistics::pumpMatrix() const
162 164 
 {
163 165 
     unsigned denom = mPumpStatUpdates != 0 ? mPumpStatUpdates : 1.f;
src/GapsStatistics.h
History View file @2cdb0256a
... ... 
@@ -28,24 +28,25 @@ private:
28 28
29 29 
 public:
30 30
31 
-    GapsStatistics(unsigned nRow, unsigned nCol, unsigned nFactor, PumpThreshold t=PUMP_CUT);
32 
-
33 
-    ColMatrix AMean() const;
34 
-    RowMatrix PMean() const;
35 
-    ColMatrix AStd() const;
36 
-    RowMatrix PStd() const;
37 
-
38 
-    float meanChiSq(const AmplitudeGibbsSampler &ASampler) const;
31 
+    GapsStatistics(unsigned nRow, unsigned nCol, unsigned nPatterns,
32 
+        PumpThreshold t=PUMP_CUT);
39 33
40 34 
     void update(const AmplitudeGibbsSampler &ASampler,
41 35 
         const PatternGibbsSampler &PSampler);
42 36
37 
+    ColMatrix Amean() const;
38 
+    RowMatrix Pmean() const;
39 
+    ColMatrix Asd() const;
40 
+    RowMatrix Psd() const;
41 
+
42 
+    float meanChiSq(const AmplitudeGibbsSampler &ASampler) const;
43 
+
44 
+    // PUMP statistics
43 45 
     void updatePump(const AmplitudeGibbsSampler &ASampler,
44 46 
         const PatternGibbsSampler &PSampler);
45 47
46 48 
     RowMatrix pumpMatrix() const;
47 49 
     RowMatrix meanPattern();
48 
-
49 50 
     void patternMarkers(ColMatrix normedA, RowMatrix normedP, ColMatrix &statMatrix);
50 51
51 52 
     // serialization
src/GibbsSampler.cpp
History View file @2cdb0256a
... ... 
@@ -3,21 +3,15 @@
3 3
4 4 
 /******************** AmplitudeGibbsSampler Implementation ********************/
5 5
6 
-AmplitudeGibbsSampler::AmplitudeGibbsSampler(const RowMatrix &D,
7 
-unsigned nFactor, float alpha, float maxGibbsMass, bool singleCell)
8 
-    :
9 
-GibbsSampler(D, D.nRow(), nFactor, nFactor, alpha, maxGibbsMass, singleCell)
6 
+void AmplitudeGibbsSampler::sync(PatternGibbsSampler &sampler)
10 7 
 {
11 
-    mQueue.setDimensionSize(mBinSize, mNumCols);
8 
+    mOtherMatrix = &(sampler.mMatrix);
9 
+    mAPMatrix = sampler.mAPMatrix;
12 10 
 }
13 11
14 
-AmplitudeGibbsSampler::AmplitudeGibbsSampler(const std::string &pathToData,
15 
-unsigned nFactor, float alpha, float maxGibbsMass, bool singleCell)
16 
-    :
17 
-GibbsSampler(pathToData, FileParser(pathToData).nRow(), nFactor, nFactor, alpha,
18 
-maxGibbsMass, singleCell)
12 
+void AmplitudeGibbsSampler::recalculateAPMatrix()
19 13 
 {
20 
-    mQueue.setDimensionSize(mBinSize, mNumCols);
14 
+    mAPMatrix = gaps::algo::matrixMultiplication(mMatrix, *mOtherMatrix);
21 15 
 }
22 16
23 17 
 unsigned AmplitudeGibbsSampler::getRow(uint64_t pos) const
... ... 
@@ -41,12 +35,6 @@ bool AmplitudeGibbsSampler::canUseGibbs(unsigned r1, unsigned c1, unsigned r2, u
41 35 
     return canUseGibbs(r1, c1) || canUseGibbs(r2, c2);
42 36 
 }
43 37
44 
-void AmplitudeGibbsSampler::sync(PatternGibbsSampler &sampler)
45 
-{
46 
-    mOtherMatrix = &(sampler.mMatrix);
47 
-    mAPMatrix = sampler.mAPMatrix;
48 
-}
49 
-
50 38 
 void AmplitudeGibbsSampler::updateAPMatrix(unsigned row, unsigned col, float delta)
51 39 
 {
52 40 
     const float *other = mOtherMatrix->rowPtr(col);
... ... 
@@ -79,8 +67,9 @@ AlphaParameters AmplitudeGibbsSampler::alphaParameters(unsigned row, unsigned co
79 67 
 AlphaParameters AmplitudeGibbsSampler::alphaParameters(unsigned r1, unsigned c1,
80 68 
 unsigned r2, unsigned c2)
81 69 
 {
82 
-    if (r1 == r2)
70 
+    if (r1 == r2) // TODO should this ever happen
83 71 
     {
72 
+        GAPS_ASSERT(false);
84 73 
         return gaps::algo::alphaParameters(mDMatrix.nCol(), mDMatrix.rowPtr(r1),
85 74 
             mSMatrix.rowPtr(r1), mAPMatrix.rowPtr(r1), mOtherMatrix->rowPtr(c1),
86 75 
             mOtherMatrix->rowPtr(c2));
... ... 
@@ -98,8 +87,9 @@ float AmplitudeGibbsSampler::computeDeltaLL(unsigned row, unsigned col, float ma
98 87 
 float AmplitudeGibbsSampler::computeDeltaLL(unsigned r1, unsigned c1, float m1,
99 88 
 unsigned r2, unsigned c2, float m2)
100 89 
 {
101 
-    if (r1 == r2)
90 
+    if (r1 == r2) // TODO should this ever happen
102 91 
     {
92 
+        GAPS_ASSERT(false);
103 93 
         return gaps::algo::deltaLL(mDMatrix.nCol(), mDMatrix.rowPtr(r1),
104 94 
             mSMatrix.rowPtr(r1), mAPMatrix.rowPtr(r1), mOtherMatrix->rowPtr(c1),
105 95 
             m1, mOtherMatrix->rowPtr(c2), m2);
... ... 
@@ -109,21 +99,15 @@ unsigned r2, unsigned c2, float m2)
109 99
110 100 
 /********************* PatternGibbsSampler Implementation *********************/
111 101
112 
-PatternGibbsSampler::PatternGibbsSampler(const RowMatrix &D,
113 
-unsigned nFactor, float alpha, float maxGibbsMass, bool singleCell)
114 
-    :
115 
-GibbsSampler(D, nFactor, D.nCol(), nFactor, alpha, maxGibbsMass, singleCell)
102 
+void PatternGibbsSampler::recalculateAPMatrix()
116 103 
 {
117 
-    mQueue.setDimensionSize(mBinSize, mNumRows);
104 
+    mAPMatrix = gaps::algo::matrixMultiplication(*mOtherMatrix, mMatrix);
118 105 
 }
119 106
120 
-PatternGibbsSampler::PatternGibbsSampler(const std::string &pathToData,
121 
-unsigned nFactor, float alpha, float maxGibbsMass, bool singleCell)
122 
-    :
123 
-GibbsSampler(pathToData, nFactor, FileParser(pathToData).nCol(), nFactor, alpha,
124 
-maxGibbsMass, singleCell)
107 
+void PatternGibbsSampler::sync(AmplitudeGibbsSampler &sampler)
125 108 
 {
126 
-    mQueue.setDimensionSize(mBinSize , mNumRows);
109 
+    mOtherMatrix = &(sampler.mMatrix);
110 
+    mAPMatrix = sampler.mAPMatrix;
127 111 
 }
128 112
129 113 
 unsigned PatternGibbsSampler::getRow(uint64_t pos) const
... ... 
@@ -147,12 +131,6 @@ bool PatternGibbsSampler::canUseGibbs(unsigned r1, unsigned c1, unsigned r2, uns
147 131 
     return canUseGibbs(r1, c1) || canUseGibbs(r2, c2);
148 132 
 }
149 133
150 
-void PatternGibbsSampler::sync(AmplitudeGibbsSampler &sampler)
151 
-{
152 
-    mOtherMatrix = &(sampler.mMatrix);
153 
-    mAPMatrix = sampler.mAPMatrix;
154 
-}
155 
-
156 134 
 void PatternGibbsSampler::updateAPMatrix(unsigned row, unsigned col, float delta)
157 135 
 {
158 136 
     const float *other = mOtherMatrix->colPtr(row);
... ... 
@@ -185,8 +163,9 @@ AlphaParameters PatternGibbsSampler::alphaParameters(unsigned row, unsigned col)
185 163 
 AlphaParameters PatternGibbsSampler::alphaParameters(unsigned r1, unsigned c1,
186 164 
 unsigned r2, unsigned c2)
187 165 
 {
188 
-    if (c1 == c2)
166 
+    if (c1 == c2) // TODO should this ever happen
189 167 
     {
168 
+        GAPS_ASSERT(false);
190 169 
         return gaps::algo::alphaParameters(mDMatrix.nRow(), mDMatrix.colPtr(c1),
191 170 
             mSMatrix.colPtr(c1), mAPMatrix.colPtr(c1), mOtherMatrix->colPtr(r1),
192 171 
             mOtherMatrix->colPtr(r2));
... ... 
@@ -204,8 +183,9 @@ float PatternGibbsSampler::computeDeltaLL(unsigned row, unsigned col, float mass
204 183 
 float PatternGibbsSampler::computeDeltaLL(unsigned r1, unsigned c1, float m1,
205 184 
 unsigned r2, unsigned c2, float m2)
206 185 
 {
207 
-    if (c1 == c2)
186 
+    if (c1 == c2) // TODO should this ever happen
208 187 
     {
188 
+        GAPS_ASSERT(false);
209 189 
         return gaps::algo::deltaLL(mDMatrix.nRow(), mDMatrix.colPtr(c1),
210 190 
             mSMatrix.colPtr(c1), mAPMatrix.colPtr(c1), mOtherMatrix->colPtr(r1),
211 191 
             m1, mOtherMatrix->colPtr(r2), m2);
src/GibbsSampler.h
History View file @2cdb0256a
... ... 
@@ -11,13 +11,12 @@
11 11
12 12 
 #include <algorithm>
13 13
14 
-// forward declarations needed for friend classes
14 
+// forward declarations needed for friend classes/functions
15 
+
15 16 
 class AmplitudeGibbsSampler;
16 17 
 class PatternGibbsSampler;
17 18 
 class GapsStatistics;
18 19
19 
-/************************** GIBBS SAMPLER INTERFACE **************************/
20 
-
21 20 
 template <class T, class MatA, class MatB>
22 21 
 class GibbsSampler;
23 22
... ... 
@@ -27,6 +26,8 @@ Archive& operator<<(Archive &ar, GibbsSampler<T, MatA, MatB> &samp);
27 26 
 template <class T, class MatA, class MatB>
28 27 
 Archive& operator>>(Archive &ar, GibbsSampler<T, MatA, MatB> &samp);
29 28
29 
+/*************************** GIBBS SAMPLER INTERFACE **************************/
30 
+
30 31 
 template <class T, class MatA, class MatB>
31 32 
 class GibbsSampler
32 33 
 {
... ... 
@@ -36,12 +37,13 @@ private:
36 37
37 38 
 protected:
38 39
39 
-    MatA mMatrix;
40 
-    MatB* mOtherMatrix;
41 40 
     MatB mDMatrix;
42 41 
     MatB mSMatrix;
43 42 
     MatB mAPMatrix;
44 43
44 
+    MatA mMatrix;
45 
+    MatB* mOtherMatrix;
46 
+
45 47 
     ProposalQueue mQueue;
46 48 
     AtomicDomain mDomain;
47 49
... ... 
@@ -60,6 +62,9 @@ protected:
60 62
61 63 
     void processProposal(const AtomicProposal &prop);
62 64
65 
+    void addMass(uint64_t pos, float mass, unsigned row, unsigned col);
66 
+    void removeMass(uint64_t pos, float mass, unsigned row, unsigned col);
67 
+
63 68 
     void birth(uint64_t pos, unsigned row, unsigned col);
64 69 
     void death(uint64_t pos, float mass, unsigned row, unsigned col);
65 70 
     void move(uint64_t src, float mass, uint64_t dest, unsigned r1, unsigned c1,
... ... 
@@ -67,40 +72,43 @@ protected:
67 72 
     void exchange(uint64_t p1, float m1, uint64_t p2, float m2, unsigned r1,
68 73 
         unsigned c1, unsigned r2, unsigned c2);
69 74
70 
-    float gibbsMass(unsigned row, unsigned col, float mass);
71 
-    float gibbsMass(unsigned r1, unsigned c1, float m1, unsigned r2,
72 
-        unsigned c2, float m2);
73 
-
74 
-    void addMass(uint64_t pos, float mass, unsigned row, unsigned col);
75 
-    void removeMass(uint64_t pos, float mass, unsigned row, unsigned col);
76 75 
     bool updateAtomMass(uint64_t pos, float mass, float delta);
77 
-
78 76 
     void acceptExchange(uint64_t p1, float m1, float d1, uint64_t p2, float m2,
79 77 
         float d2, unsigned r1, unsigned c1, unsigned r2, unsigned c2);
80 78
79 
+    float gibbsMass(unsigned row, unsigned col, float mass);
80 
+    float gibbsMass(unsigned r1, unsigned c1, float m1, unsigned r2,
81 
+        unsigned c2, float m2);
82 
+
81 83 
 public:
82 84
83 
-    GibbsSampler(const RowMatrix &D, unsigned nrow, unsigned ncol,
84 
-        unsigned nPatterns, float alpha, float maxGibbsMass, bool singleCell);
85 
+    template <class DataType>
86 
+    GibbsSampler(const DataType &data, bool amp, unsigned nPatterns);
85 87
86 
-    GibbsSampler(const std::string &pathToData, unsigned nrow, unsigned ncol,
87 
-        unsigned nPatterns, float alpha, float maxGibbsMass, bool singleCell);
88 
+    template <class DataType>
89 
+    void setUncertainty(const DataType &unc);
90 
+
91 
+    void setSparsity(float alpha, bool singleCell);
92 
+    void setMaxGibbsMass(float max);
93 
+    void setAnnealingTemp(float temp);
88 94
89 
-    void setUncertainty(const RowMatrix &S);
90 
-    void setUncertainty(const std::string &path);
95 
+    void setMatrix(const MatA &mat);
91 96
92 97 
     void update(unsigned nSteps, unsigned nCores);
93 
-    void setAnnealingTemp(float temp);
94 
-    float getAvgQueue() const;
95 
-
98 
+
99 
+    unsigned dataRows() const;
100 
+    unsigned dataCols() const;
101 
+
96 102 
     float chi2() const;
97 103 
     uint64_t nAtoms() const;
98 104
99 
-    void setMatrix(const MatA &mat);
105 
+    #ifdef GAPS_DEBUG
106 
+    float getAvgQueue() const;
107 
+    #endif
100 108
101 109 
     // serialization
102 
-    friend Archive& operator<< <T, MatA, MatB> (Archive &ar, GibbsSampler &samp);
103 
-    friend Archive& operator>> <T, MatA, MatB> (Archive &ar, GibbsSampler &samp);
110 
+    friend Archive& operator<< <T, MatA, MatB> (Archive &ar, GibbsSampler &sampler);
111 
+    friend Archive& operator>> <T, MatA, MatB> (Archive &ar, GibbsSampler &sampler);
104 112 
 };
105 113
106 114 
 class AmplitudeGibbsSampler : public GibbsSampler<AmplitudeGibbsSampler, ColMatrix, RowMatrix>
... ... 
@@ -116,16 +124,6 @@ private:
116 124 
     bool canUseGibbs(unsigned r1, unsigned c1, unsigned r2, unsigned c2) const;
117 125 
     void updateAPMatrix(unsigned row, unsigned col, float delta);
118 126
119 
-public:
120 
-
121 
-    AmplitudeGibbsSampler(const RowMatrix &D, unsigned nFactor,
122 
-        float alpha=0.f, float maxGibbsMass=0.f, bool singleCell=false);
123 
-
124 
-    AmplitudeGibbsSampler(const std::string &pathToData, unsigned nFactor,
125 
-    float alpha=0.f, float maxGibbsMass=0.f, bool singleCell=false);
126 
-
127 
-    void sync(PatternGibbsSampler &sampler);
128 
-
129 127 
     AlphaParameters alphaParameters(unsigned row, unsigned col);
130 128 
     AlphaParameters alphaParameters(unsigned r1, unsigned c1, unsigned r2,
131 129 
         unsigned c2);
... ... 
@@ -133,6 +131,14 @@ public:
133 131 
     float computeDeltaLL(unsigned row, unsigned col, float mass);
134 132 
     float computeDeltaLL(unsigned r1, unsigned c1, float m1, unsigned r2,
135 133 
         unsigned c2, float m2);
134 
+
135 
+public:
136 
+
137 
+    template <class DataType>
138 
+    AmplitudeGibbsSampler(const DataType &data, unsigned nPatterns);
139 
+
140 
+    void sync(PatternGibbsSampler &sampler);
141 
+    void recalculateAPMatrix();
136 142 
 };
137 143
138 144 
 class PatternGibbsSampler : public GibbsSampler<PatternGibbsSampler, RowMatrix, ColMatrix>
... ... 
@@ -148,16 +154,6 @@ private:
148 154 
     bool canUseGibbs(unsigned r1, unsigned c1, unsigned r2, unsigned c2) const;
149 155 
     void updateAPMatrix(unsigned row, unsigned col, float delta);
150 156
151 
-public:
152 
-
153 
-    PatternGibbsSampler(const RowMatrix &D, unsigned nFactor,
154 
-        float alpha=0.f, float maxGibbsMass=0.f, bool singleCell=false);
155 
-
156 
-    PatternGibbsSampler(const std::string &pathToData, unsigned nFactor,
157 
-        float alpha=0.f, float maxGibbsMass=0.f, bool singleCell=false);
158 
-
159 
-    void sync(AmplitudeGibbsSampler &sampler);
160 
-
161 157 
     AlphaParameters alphaParameters(unsigned row, unsigned col);
162 158 
     AlphaParameters alphaParameters(unsigned r1, unsigned c1, unsigned r2,
163 159 
         unsigned c2);
... ... 
@@ -165,62 +161,93 @@ public:
165 161 
     float computeDeltaLL(unsigned row, unsigned col, float mass);
166 162 
     float computeDeltaLL(unsigned r1, unsigned c1, float m1, unsigned r2,
167 163 
         unsigned c2, float m2);
164 
+
165 
+public:
166 
+
167 
+    template <class DataType>
168 
+    PatternGibbsSampler(const DataType &data, unsigned nPatterns);
169 
+
170 
+    void sync(AmplitudeGibbsSampler &sampler);
171 
+    void recalculateAPMatrix();
168 172 
 };
169 173
170 
-/******************* IMPLEMENTATION OF TEMPLATED FUNCTIONS *******************/
174 
+/******************** IMPLEMENTATION OF TEMPLATED FUNCTIONS *******************/
171 175
172 
-template <class T, class MatA, class MatB>
173 
-GibbsSampler<T, MatA, MatB>::GibbsSampler(const RowMatrix &D,
174 
-unsigned nrow, unsigned ncol, unsigned nPatterns, float alpha,
175 
-float maxGibbsMass, bool singleCell)
176 
+template <class DataType>
177 
+AmplitudeGibbsSampler::AmplitudeGibbsSampler(const DataType &data, unsigned nPatterns)
176 178 
     :
177 
-mMatrix(nrow, ncol), mOtherMatrix(NULL), mDMatrix(D),
178 
-mSMatrix(mDMatrix.pmax(0.1f)), mAPMatrix(D.nRow(), D.nCol()),
179 
-mQueue(nrow * ncol, alpha), mLambda(0.f), mMaxGibbsMass(maxGibbsMass),
180 
-mAnnealingTemp(0.f), mNumRows(nrow), mNumCols(ncol), mAvgQueue(0.f),
181 
-mNumQueues(0.f)
179 
+GibbsSampler(data, true, nPatterns)
182 180 
 {
183 
-    mBinSize = std::numeric_limits<uint64_t>::max()
184 
-        / static_cast<uint64_t>(mNumRows * mNumCols);
185 
-    uint64_t remain = std::numeric_limits<uint64_t>::max()
186 
-        % static_cast<uint64_t>(mNumRows * mNumCols);
187 
-    mQueue.setDomainSize(std::numeric_limits<uint64_t>::max() - remain);
188 
-    mDomain.setDomainSize(std::numeric_limits<uint64_t>::max() - remain);
181 
+    mQueue.setDimensionSize(mBinSize, mNumCols);
182 
+}
189 183
190 
-    float meanD = singleCell ? gaps::algo::nonZeroMean(mDMatrix) :
191 
-        gaps::algo::mean(mDMatrix);
192 
-    mLambda = alpha * std::sqrt(nPatterns / meanD);
193 
-    mMaxGibbsMass = maxGibbsMass / mLambda;
184 
+template <class DataType>
185 
+PatternGibbsSampler::PatternGibbsSampler(const DataType &data, unsigned nPatterns)
186 
+    :
187 
+GibbsSampler(data, false, nPatterns)
188 
+{
189 
+    mQueue.setDimensionSize(mBinSize, mNumRows);
194 190 
 }
195 191
196 192 
 template <class T, class MatA, class MatB>
197 
-GibbsSampler<T, MatA, MatB>::GibbsSampler(const std::string &pathToData,
198 
-unsigned nrow, unsigned ncol, unsigned nPatterns, float alpha,
199 
-float maxGibbsMass, bool singleCell)
193 
+template <class DataType>
194 
+GibbsSampler<T, MatA, MatB>::GibbsSampler(const DataType &data,
195 
+bool amp, unsigned nPatterns)
200 196 
     :
201 
-mMatrix(nrow, ncol), mOtherMatrix(NULL), mDMatrix(pathToData),
202 
-mSMatrix(mDMatrix.pmax(0.1f)), mAPMatrix(mDMatrix.nRow(), mDMatrix.nCol()),
203 
-mQueue(nrow * ncol, alpha), mLambda(0.f), mMaxGibbsMass(maxGibbsMass),
204 
-mAnnealingTemp(0.f), mNumRows(nrow), mNumCols(ncol), mAvgQueue(0.f),
205 
-mNumQueues(0.f)
197 
+mDMatrix(data), mSMatrix(mDMatrix.pmax(0.1f, 0.1f)),
198 
+mAPMatrix(mDMatrix.nRow(), mDMatrix.nCol()),
199 
+mMatrix(amp ? mDMatrix.nRow() : nPatterns, amp ? nPatterns : mDMatrix.nCol()),
200 
+mOtherMatrix(NULL), mQueue(mMatrix.nRow() * mMatrix.nCol()), mLambda(0.f),
201 
+mMaxGibbsMass(100.f), mAnnealingTemp(1.f), mNumRows(mMatrix.nRow()),
202 
+mNumCols(mMatrix.nCol()), mAvgQueue(0.f), mNumQueues(0.f)
206 203 
 {
204 
+    // calculate atomic domain size
207 205 
     mBinSize = std::numeric_limits<uint64_t>::max()
208 206 
         / static_cast<uint64_t>(mNumRows * mNumCols);
209 
-    uint64_t remain = std::numeric_limits<uint64_t>::max()
210 
-        % static_cast<uint64_t>(mNumRows * mNumCols);
211 
-    mQueue.setDomainSize(std::numeric_limits<uint64_t>::max() - remain);
212 
-    mDomain.setDomainSize(std::numeric_limits<uint64_t>::max() - remain);
207 
+    mQueue.setDomainSize(mBinSize * mNumRows * mNumCols);
208 
+    mDomain.setDomainSize(mBinSize * mNumRows * mNumCols);
209 
+
210 
+    // default sparsity parameters
211 
+    setSparsity(0.01, false);
212 
+}
213 
+
214 
+template <class T, class MatA, class MatB>
215 
+template <class DataType>
216 
+void GibbsSampler<T, MatA, MatB>::setUncertainty(const DataType &unc)
217 
+{
218 
+    mSMatrix = MatB(unc);
219 
+}
220 
+
221 
+template <class T, class MatA, class MatB>
222 
+void GibbsSampler<T, MatA, MatB>::setSparsity(float alpha, bool singleCell)
223 
+{
224 
+    mQueue.setAlpha(alpha);
213 225
214 226 
     float meanD = singleCell ? gaps::algo::nonZeroMean(mDMatrix) :
215 227 
         gaps::algo::mean(mDMatrix);
228 
+
229 
+    unsigned nPatterns = mDMatrix.nRow() == mMatrix.nRow() ? mMatrix.nCol() :
230 
+        mMatrix.nRow();
231 
+
216 232 
     mLambda = alpha * std::sqrt(nPatterns / meanD);
217 
-    mMaxGibbsMass = maxGibbsMass / mLambda;
218 233 
 }
219 234
220 235 
 template <class T, class MatA, class MatB>
221 
-T* GibbsSampler<T, MatA, MatB>::impl()
236 
+void GibbsSampler<T, MatA, MatB>::setMaxGibbsMass(float max)
222 237 
 {
223 
-    return static_cast<T*>(this);
238 
+    mMaxGibbsMass = max;
239 
+}
240 
+
241 
+template <class T, class MatA, class MatB>
242 
+void GibbsSampler<T, MatA, MatB>::setAnnealingTemp(float temp)
243 
+{
244 
+    mAnnealingTemp = temp;
245 
+}
246 
+
247 
+template <class T, class MatA, class MatB>
248 
+void GibbsSampler<T, MatA, MatB>::setMatrix(const MatA &mat)
249 
+{
250 
+    mMatrix = mat;
224 251 
 }
225 252
226 253 
 template <class T, class MatA, class MatB>
... ... 
@@ -229,13 +256,19 @@ void GibbsSampler<T, MatA, MatB>::update(unsigned nSteps, unsigned nCores)
229 256 
     unsigned n = 0;
230 257 
     while (n < nSteps)
231 258 
     {
259 
+        // populate queue, prepare domain for this queue
232 260 
         mQueue.populate(mDomain, nSteps - n);
233