Made a few improvements to the 'noinit' LDA scripts.

scripts/run_lda.sbatch

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-#SBATCH --partition=mstephens
+#SBATCH --partition=broadwl
 #SBATCH --account=pi-mstephens
 #SBATCH --cpus-per-task=4
 #SBATCH --mem=32G

scripts/run_lda_noinit.R

@@ -28,7 +28,7 @@ countsfile <- out$counts
 k          <- out$k
 outfile    <- tail(unlist(strsplit(countsfile,"/")),n = 1)
 outfile    <- substr(outfile,1,nchar(outfile) - 6)
-outfile    <- paste0("lda-",outfile,"-k=",k,".rds")
+outfile    <- paste0("lda-",outfile,"-noinit-k=",k,".rds")
 rm(parser,out)
 
 # Initialize the sequence of pseudorandom numbers.
@@ -50,7 +50,7 @@ lda <- LDA(counts,k,
            control = list(alpha = 1,
                           estimate.alpha = FALSE,
                           verbose = 1,
-                          em = list(iter.max = 10,tol = 0),
+                          em = list(iter.max = 100,tol = 0),
                           var = list(iter.max = 20,tol = 0),
                           keep = 1))
 t1 <- proc.time()

scripts/run_lda_noinit.sbatch

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-#SBATCH --partition=mstephens
+#SBATCH --partition=broadwl
 #SBATCH --account=pi-mstephens
 #SBATCH --cpus-per-task=4
 #SBATCH --mem=8G

scripts/run_lda_noinit_all.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+# The shell commands below will submit Slurm jobs to run LDA for all
+# data sets, and for different choices of K.
+MAIN_SCRIPT=run_lda_noinit.sbatch
+
+# Run LDA on the newsgroups data.
+#
+#                     data        k
+sbatch ${MAIN_SCRIPT} newsgroups  2
+sbatch ${MAIN_SCRIPT} newsgroups  3
+sbatch ${MAIN_SCRIPT} newsgroups  4
+sbatch ${MAIN_SCRIPT} newsgroups  5
+sbatch ${MAIN_SCRIPT} newsgroups  5
+sbatch ${MAIN_SCRIPT} newsgroups  6
+sbatch ${MAIN_SCRIPT} newsgroups  7
+sbatch ${MAIN_SCRIPT} newsgroups  8
+sbatch ${MAIN_SCRIPT} newsgroups  9
+sbatch ${MAIN_SCRIPT} newsgroups 10
+sbatch ${MAIN_SCRIPT} newsgroups 11
+sbatch ${MAIN_SCRIPT} newsgroups 12