I am learning to do differentially expressed gene analysis on microarray data(typically Affaymetrix GeneChip Human Genome U133 Plus 2.0 Array) of tumor vs normal. I've read some tutorials and know the basic steps. But I never saw people talk about when exactly to annotate the probes to gene symbols.
I now do DEG analysis on probe level and map probes to gene symbols after I get a list of differentially expressed probes. However, many differentially expressed probes map to the same gene, which resluts in much less DEGs than differentially expressed probes. I've also seen poeple map probes to gene symbols before DEG analysis, exactly after they get eset( eset <- exprs(rma(affyBatch)) ) . I am really confused. 
I give the briefing of my code below:

library(GEOquery)
library(affycoretools)
library(limma)
library(dplyr)

gse54129 <- getGEO('GSE54129', destdir = '../GSE/series_matrix', getGPL = FALSE)
pdata <- pData(phenoData(gse54129[[1]]))
group <- ifelse(pdata$characteristics_ch1 == 'tissue: normal', 'Normal', 'Tumor')
group <- factor(group, levels=c('Normal', 'Tumor'))
table(group)
# Normal  Tumor 
#     21    111
###### load CEL files #####
library(affy)
files <- paste('../GSE/CEL/GSE54129/', list.files('../GSE/CEL/GSE54129/', pattern = 'CEL.gz'), sep = '')
affyBatch <- ReadAffy(filenames=files)
sampleNames(affyBatch)
sampleNames(affyBatch) <- substr(sampleNames(affyBatch), 1, 10)
sampleNames(affyBatch)

data.rma <- rma(affyBatch)

eset <- exprs(data.rma)

plotPCA(eset, addtext=colnames(eset), groups=factor(pdata$characteristics_ch1), legend=FALSE)

##### DEG #######
mode(eset)
# [1] "numeric"

design <- model.matrix(~group)

fit <- lmFit(eset, design)

fit2 <- eBayes(fit)
topTable(fit2, coef = 2, adjust.method = 'BH')
res <- decideTests(fit2, lfc = 2)
summary(res)
#    (Intercept) groupTumor
# -1           0        554
# 0            0      53609
# 1        54675        512

deg.54129 <- topTable(fit2, coef = 2, number = Inf, adjust.method = 'BH')
deg.54129$gene <- rownames(deg.54129)

deg.54129.sig2 <- filter(deg.54129, adj.P.Val < 0.05 & abs(logFC) >= 2)
dim(deg.54129.sig2)
# [1] 1066    7

library(hgu133plus2.db)
probes <- deg.54129.sig2$gene
gene.symbol <- as.character(as.list(hgu133plus2SYMBOL[probes]))
gene.entrez <- as.character(as.list(hgu133plus2ENTREZID[probes]))

deg.54129.sig2 <- cbind(PROBE=deg.54129.sig2$gene, SYMBOL=gene.symbol, ENTREZID=gene.entrez,
                        deg.54129.sig2[, -ncol(deg.54129.sig2)])
deg.54129.sig2 <- arrange(deg.54129.sig2, SYMBOL, desc(abs(logFC)))
deg.54129.sig2 <- deg.54129.sig2[!duplicated(deg.54129.sig2$SYMBOL), ]
dim(deg.54129.sig2)
# [1] 729   9
save(deg.54129.sig2, file = 'gse54129.deg.probe_level.rda')

It doesn't make any difference whether you map to probes before the DE analysis or after, although I myself would always do it before.

You could follow the example case study in Section 17.2 of the limma User's Guide. The example is for hgu95 affy arrays instead of u133plus2 but the code should be pretty straightforward to adapt.