Skip to content

Commit

Permalink
more thorough testing of datatable.optimize levels, part of #2572
Browse files Browse the repository at this point in the history
  • Loading branch information
@MichaelChirico
MichaelChirico committed Feb 16, 2019
1 parent b116f95 commit 5dddf66
Showing 1 changed file with 69 additions and 60 deletions.
129 changes: 69 additions & 60 deletions inst/tests/tests.Rraw
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1865,10 +1865,18 @@ setnames(ans2,"x","V1")
setnames(ans2,"y","V2")
test(654, ans1, ans2)

options(datatable.optimize = Inf)
test(656, DT[,mean(x),by=grp1,verbose=TRUE], output="GForce optimized j to.*gmean")
test(657, DT[,list(mean(x)),by=grp1,verbose=TRUE], output="GForce optimized j to.*gmean")
test(658, DT[,list(mean(x),mean(y)),by=grp1,verbose=TRUE], output="GForce optimized j to.*gmean")
options(datatable.optimize = 0L)
test(656.1, DT[ , mean(x), by=grp1, verbose=TRUE], output='(GForce FALSE)')
test(657.1, DT[ , list(mean(x)), by=grp1, verbose=TRUE], output="(GForce FALSE)")
test(658.3, DT[ , list(mean(x), mean(y)), by=grp1, verbose=TRUE], output="(GForce FALSE)")
options(datatable.optimize = 1L)
test(656.2, DT[ , mean(x), by=grp1, verbose=TRUE], output='(GForce FALSE)')
test(657.2, DT[ , list(mean(x)), by=grp1, verbose=TRUE], output="(GForce FALSE)")
test(658.2, DT[ , list(mean(x), mean(y)), by=grp1, verbose=TRUE], output="(GForce FALSE)")
options(datatable.optimize = 2L)
test(656.3, DT[ , mean(x), by=grp1, verbose=TRUE], output="GForce optimized j to.*gmean")
test(657.3, DT[ , list(mean(x)), by=grp1, verbose=TRUE], output="GForce optimized j to.*gmean")
test(658.3, DT[ , list(mean(x), mean(y)), by=grp1, verbose=TRUE], output="GForce optimized j to.*gmean")
tt = capture.output(DT[,list(mean(x),mean(y)),by=list(grp1,grp2),verbose=TRUE])
test(659, !length(grep("Wrote less rows", tt))) # first group is one row with this seed. Ensure we treat this as aggregate case rather than allocate too many rows.

Expand Down Expand Up @@ -2390,8 +2398,12 @@ test(864.3, rbindlist(list(data.table(logical(0),logical(0)), DT<-data.table(baz

# Steve's find that setnames failed for numeric 'old' when pointing to duplicated names
DT = data.table(a=1:3,b=1:3,v=1:6,w=1:6)
options(datatable.optimize = 0L)
test(865.1, ans1<-DT[,{list(name1=sum(v),name2=sum(w))},by="a,b",verbose=TRUE], output="(GForce FALSE)")
options(datatable.optimize = 1L)
test(865.2, ans1<-DT[,{list(name1=sum(v),name2=sum(w))},by="a,b",verbose=TRUE], output="(GForce FALSE)")
options(datatable.optimize = 2L)
test(865, ans1<-DT[,{list(name1=sum(v),name2=sum(w))},by="a,b",verbose=TRUE],
test(865.3, ans1<-DT[,{list(name1=sum(v),name2=sum(w))},by="a,b",verbose=TRUE],
output="GForce optimized.*gsum[(]v[)], gsum[(]w[)]") # v1.9.7 treats wrapped {} better, so this is now optimized
options(datatable.optimize = Inf)
test(866, names(ans1), c("a","b","name1","name2"))
Expand Down Expand Up @@ -4453,57 +4465,50 @@ unlink(f)
#################################
# FR #2722 optimise j=c(lapply(.SD,sum, ...)) - here any amount of such lapply(.SD, ...) can occur and in any order
set.seed(45L)
dt <- data.table(a=sample(2,10,TRUE), b=sample(3,10,TRUE), c=sample(4,10,TRUE), d=sample(5,10,TRUE))
dt = data.table(a=sample(2,10,TRUE), b=sample(3,10,TRUE), c=sample(4,10,TRUE), d=sample(5,10,TRUE))
dt2 = data.table(x=c(1,1,1,2,2,2), y=1:6)

options(datatable.optimize=1L)
ans2 <- dt[, c(lapply(.SD, mean), lapply(.SD, sum)), by=a]
options(datatable.optimize=Inf)
test(1268.1, dt[, c(lapply(.SD, mean), lapply(.SD, sum)), by=a, verbose=TRUE], ans2,
output="GForce optimized j to 'list(gmean(b), gmean(c), gmean(d), gsum(b), gsum(c), gsum(d))'")
options(datatable.optimize=0L)
# auto-naming behavior is different for no-optimization case; just check optimization is off
test(1268.1, dt[, c(lapply(.SD, mean), lapply(.SD, sum)), by=a, verbose=TRUE], output = 'All optimizations are turned off')
test(1268.2, dt[, c(lapply(.SD, mean), .N), by=a, verbose=TRUE], output = 'All optimizations are turned off')
test(1268.3, dt[, c(list(c), lapply(.SD, mean)), by=a, verbose=TRUE], output="All optimizations are turned off")
test(1268.4, dt[, c(sum(d), lapply(.SD, mean)), by=a, verbose=TRUE], output="All optimizations are turned off")
test(1268.5, dt[, c(list(sum(d)), lapply(.SD, mean)), by=a, verbose=TRUE], output="All optimizations are turned off")
# newly added tests for #861 -- optimise, but no GForce
test(1268.6, dt[, c(list(sum(d), .I), lapply(.SD, mean)), by=a, verbose=TRUE], output="All optimizations are turned off")
# don't optimise .I in c(...)
test(1268.7, dt2[, c(.I, lapply(.SD, mean)), by=x, verbose=TRUE], output="All optimizations are turned off")

options(datatable.optimize=1L)
ans2 <- dt[, c(lapply(.SD, mean), .N), by=a]
options(datatable.optimize=Inf)
test(1268.2, dt[, c(lapply(.SD, mean), .N), by=a, verbose=TRUE], ans2,
output = "lapply optimization changed j from 'c(lapply(.SD, mean), .N)' to 'list(mean(b), mean(c), mean(d), .N)'")
test(1268.8, ans1 <- dt[ , c(lapply(.SD, mean), lapply(.SD, sum)), by=a, verbose=TRUE], output="Old mean optimization.*(GForce FALSE)")
test(1268.9, ans2 <- dt[, c(lapply(.SD, mean), .N), by=a, verbose = TRUE], output="Old mean optimization.*GForce FALSE")
test(1268.10, ans3 <- dt[, c(list(c), lapply(.SD, mean)), by=a, verbose=TRUE], output = 'Old mean optimization.*GForce FALSE')
test(1268.11, ans4 <- dt[, c(sum(d), lapply(.SD, mean)), by=a, verbose = TRUE], output="Old mean optimization.*GForce FALSE")
test(1268.12, ans5 <- dt[, c(list(sum(d)), lapply(.SD, mean)), by=a, verbose=TRUE], output="Old mean optimization.*GForce FALSE")
test(1268.13, ans6 <- dt[, c(list(sum(d), .I), lapply(.SD, mean)), by=a, verbose=TRUE], output="Old mean optimization.*GForce FALSE")
test(1268.14, ans7 <- dt2[, c(.I, lapply(.SD, mean)), by=x, verbose=TRUE], output="Old mean optimization.*GForce FALSE")

options(datatable.optimize=1L)
ans2 <- dt[, c(list(c), lapply(.SD, mean)), by=a]
options(datatable.optimize=Inf)
test(1268.3, dt[, c(list(c), lapply(.SD, mean)), by=a, verbose=TRUE], ans2,
test(1268.15, dt[, c(lapply(.SD, mean), lapply(.SD, sum)), by=a, verbose=TRUE], ans1,
output="GForce optimized j to 'list(gmean(b), gmean(c), gmean(d), gsum(b), gsum(c), gsum(d))'")
test(1268.16, dt[, c(lapply(.SD, mean), .N), by=a, verbose=TRUE], ans2,
output = "lapply optimization changed j from 'c(lapply(.SD, mean), .N)' to 'list(mean(b), mean(c), mean(d), .N)'")
test(1268.17, dt[, c(list(c), lapply(.SD, mean)), by=a, verbose=TRUE], ans3,
output = "lapply optimization changed j from 'c(list(c), lapply(.SD, mean))' to 'list(c, mean(b), mean(c), mean(d))")

test(1268.4, dt[, c(as.list(c), lapply(.SD, mean)), by=a],
error = "j doesn't evaluate to the same number of columns for each group")

options(datatable.optimize=1L)
ans2 <- dt[, c(sum(d), lapply(.SD, mean)), by=a]
options(datatable.optimize=Inf)
test(1268.5, dt[, c(sum(d), lapply(.SD, mean)), by=a, verbose=TRUE], ans2,
test(1268.18, dt[, c(sum(d), lapply(.SD, mean)), by=a, verbose=TRUE], ans4,
output = "GForce optimized j to 'list(gsum(d), gmean(b), gmean(c), gmean(d))'")

options(datatable.optimize=1L)
ans2 <- dt[, c(list(sum(d)), lapply(.SD, mean)), by=a]
options(datatable.optimize=Inf)
test(1268.6, dt[, c(list(sum(d)), lapply(.SD, mean)), by=a, verbose=TRUE], ans2,
test(1268.19, dt[, c(list(sum(d)), lapply(.SD, mean)), by=a, verbose=TRUE], ans5,
output = "GForce optimized j to 'list(gsum(d), gmean(b), gmean(c), gmean(d))'")

# newly added tests for #861
# optimise, but no GForce
options(datatable.optimize=1L)
ans2 <- dt[, c(list(sum(d), .I), lapply(.SD, mean)), by=a]
options(datatable.optimize=Inf)
test(1268.7, dt[, c(list(sum(d), .I), lapply(.SD, mean)), by=a, verbose=TRUE], ans2,
test(1268.20, dt[, c(list(sum(d), .I), lapply(.SD, mean)), by=a, verbose=TRUE], ans6,
output = "lapply optimization changed j from 'c(list(sum(d), .I), lapply(.SD, mean))' to 'list(sum(d), .I, mean(b), mean(c), mean(d))'")

# don't optimise .I in c(...)
options(datatable.optimize=1L)
dt = data.table(x=c(1,1,1,2,2,2), y=1:6)
ans2 <- dt[, c(.I, lapply(.SD, mean)), by=x]
options(datatable.optimize=Inf)
test(1268.8, dt[, c(.I, lapply(.SD, mean)), by=x, verbose=TRUE], ans2,
test(1268.21, dt2[, c(.I, lapply(.SD, mean)), by=x, verbose=TRUE], ans7,
output = "lapply optimization is on, j unchanged as 'c(.I, lapply(.SD, mean))'")

test(1268.22, dt[, c(as.list(c), lapply(.SD, mean)), by=a],
error = "j doesn't evaluate to the same number of columns for each group")


### FR #2722 tests end here ###

# Wide range numeric and integer64, to test all bits
Expand Down Expand Up @@ -5949,7 +5954,7 @@ test(1437.17, DT[!a %chin% c("A", "B") & x == 2], DT[c(4, 5, 6)])
## queries with j are optimized (Correct results are tested extensively below)
test(1437.18, DT[x == 2, .(test = x+y), verbose = TRUE], output = "Optimized subsetting")
test(1437.19, DT[x == 2, test := x+y, verbose = TRUE], output = "Optimized subsetting")
## optimize option level 3 is required
## optimize option level 3 is required to get optimized subsetting
options(datatable.optimize = 2L)
test(1437.21, DT[x == 2, verbose = TRUE], output = "^ x y")
options(datatable.optimize = Inf)
Expand Down Expand Up @@ -7505,12 +7510,12 @@ test(1564.3, dt[a==5, .SD][, b := 1L], data.table(a=5L, b=1L))

# Fix for #1251, DT[, .N, by=a] and DT[, .(.N), by=a] uses GForce now
dt = data.table(a=sample(3,20,TRUE), b=1:10)
options(datatable.optimize = 0L)
test(1565.1, ans <- dt[, .N, by=a, verbose=TRUE], output="All optimizations are turned off")
options(datatable.optimize = 1L)
test(1565.2, dt[ , .N, by=a, verbose=TRUE], ans, output="lapply optimization is on, j unchanged")
options(datatable.optimize = Inf)
ans1 = dt[, .N, by=a]
ans2 = capture.output(dt[, .N, by=a, verbose=TRUE])
test(1565.1, length(grep("GForce optimized j to", ans2))>0L, TRUE) # make sure GForce optimisation works
options(datatable.optimize = 1L) # make sure result is right
test(1565.2, ans1, dt[, .N, by=a])
test(1565.3, dt[ , .N, by=a, verbose=TRUE], ans, output = "GForce optimized j to")

# Fix for #1212
set.seed(123)
Expand Down Expand Up @@ -13129,11 +13134,15 @@ suppressWarnings(rm(`___data.table_internal_test_1967.68___`))
test(1967.68, setDT(`___data.table_internal_test_1967.68___`), error = 'Cannot find symbol')

### [.data.table verbosity & non-equi-join tests
options('datatable.optimize' = 2)
test(1967.69, x[order(a), .N, verbose = TRUE], 5L,
options(datatable.optimize = 0L)
verbose_output = capture.output(x[order(a), .N, verbose = TRUE])
test(1967.69, !any(grepl('order optimization', verbose_output, fixed = TRUE)))
test(1967.70, any(grepl('[1] 5', verbose_output, fixed = TRUE)))
options('datatable.optimize' = 1L)
test(1967.71, x[order(a), .N, verbose = TRUE], 5L,
output = "i changed from 'order(...)' to 'forder(")
setkey(x)
test(1967.70, x[x, .N, on = 'a', verbose = TRUE], 5L,
test(1967.72, x[x, .N, on = 'a', verbose = TRUE], 5L,
output = "on= matches existing key")
options(datatable.optimize = Inf)

Expand All @@ -13146,19 +13155,19 @@ y = data.table(
i4 = c(-26L, 6L, -30L, -26L, -23L, 38L, -40L, -26L, -23L, 24L)
)
x[ , '_nqgrp_' := 5]
test(1967.71, x[y, on = .(i1 <= i1, i4 >= i4)], error = "'_nqgrp_' is reserved")
test(1967.73, x[y, on = .(i1 <= i1, i4 >= i4)], error = "'_nqgrp_' is reserved")
x[ , '_nqgrp_' := NULL]
test(1967.72, x[y, max(i4), on = .(i1 <= i1, i4 >= i4), verbose = TRUE], 38L,
test(1967.74, x[y, max(i4), on = .(i1 <= i1, i4 >= i4), verbose = TRUE], 38L,
output = 'Recomputing forder with non-equi.*done')
test(1967.73, x[!y, sum(i4), on = 'i1', by = .EACHI, verbose = TRUE],
test(1967.75, x[!y, sum(i4), on = 'i1', by = .EACHI, verbose = TRUE],
data.table(i1 = c(169L, 369L), V1 = c(270L, 179L)),
output = "not-join called with 'by=.EACHI'.*done")
test(1967.74, x[!y, sum(i4), on = 'i1', verbose = TRUE], 510L,
test(1967.76, x[!y, sum(i4), on = 'i1', verbose = TRUE], 510L,
output = 'Inverting irows for notjoin.*sec')
x[ , v := 0]
### hitting by = A:B branch
test(1967.75, x[ , .(v = sum(v)), by = i1:i4], x[-10L])
test(1967.76, x[1:5, sum(v), by = list(i5 = 1:5 %% 2L), verbose = TRUE],
test(1967.77, x[ , .(v = sum(v)), by = i1:i4], x[-10L])
test(1967.78, x[1:5, sum(v), by = list(i5 = 1:5 %% 2L), verbose = TRUE],
data.table(i5 = 1:0, V1 = c(0, 0)), output = 'i clause present but columns used in by not detected')

# gforce integer overflow coerce to double
Expand Down

0 comments on commit 5dddf66

Please sign in to comment.