tcovert

using DataFrames

df1 = DataFrame(v1 = rand(Int16, 10000),
                v2 = rand(Int8, 10000),
                v3 = randn(10000))
df2 = DataFrame(v1 = rand(Int16, 10000),
                v2 = rand(Int8, 10000),
                v4 = randn(10000))

# this works

tcovert

using DataFrames

df = DataFrame(v1 = rand(Int16, 500),
               v2 = rand(Int16, 500),
               v3 = rand(Int16, 500),
               v4 = rand(Int16, 500))

unique0 = size(unique(df[:, [:v1, :v2, :v3]]), 1)
unique1 = size(unique(df[:, [:v1, :v2, :v3, :v4]]), 1)

tcovert

using DataFrames

df1 = DataFrame(v1 = rand(Int16, 500))
df2 = DataFrame(v2 = rand(Int16, 500))

df = join(df1, df2, kind = :cross)
# this works
groupby(df, [:v1, :v2])

# this doesn't

tcovert

addprocs(4)
blas_set_num_threads(1)

function f(K, k, y, masks, ranges)
    function dowork(i)
        z = K[masks[:, i], masks[:, i]] \ y[ranges[i, 1]:ranges[i, 2]]
        z = k[:, ranges[i, 1]:ranges[i, 2]] * z
        return z
    end
    r = pmap(dowork, collect(1:size(ranges, 1)))

tcovert

using DataFrames

ids = rand(15000:25000, 1600)
ids = unique(ids)
ids = String[string("f",x) for x in ids]
keys = 1:200
keys = String[string("k",x) for x in keys]

temp = join(DataFrame(id = ids), DataFrame(key = keys), kind = :cross)
keepflag = rand(size(temp,1))