Merge pull request dmlc#25 from slundberg/pull-request/0287171a

Add support for transposed sparse matrices.

Author: tqchen

src/xgboost_lib.jl

@@ -31,12 +31,12 @@ type DMatrix
         finalizer(sp, JLFree)
         sp
     end
-    function DMatrix{K<:Real, V<:Integer}(data::SparseMatrixCSC{K, V}; kwargs...)
-        handle = XGDMatrixCreateFromCSC(convert(SparseMatrixCSC{Float32, Int64}, data))
-        sp = new(handle,  _setinfo)
+    function DMatrix{K<:Real, V<:Integer}(data::SparseMatrixCSC{K, V}, transposed::Bool=false; kwargs...)
+        handle = (transposed ? XGDMatrixCreateFromCSCT(data) : XGDMatrixCreateFromCSC(data))
         for itm in kwargs
             _setinfo(handle, string(itm[1]), itm[2])
         end
+        sp = new(handle, _setinfo)
         finalizer(sp, JLFree)
         sp
     end

src/xgboost_wrapper_h.jl

@@ -24,13 +24,27 @@ function XGDMatrixCreateFromFile(fname::ASCIIString, slient::Int32)
     return handle[]
 end
 
-function XGDMatrixCreateFromCSC(data::SparseMatrixCSC{Float32, Int64})
+function XGDMatrixCreateFromCSC(data::SparseMatrixCSC)
     handle = Ref{Ptr{Void}}()
     @xgboost_ccall(
         :XGDMatrixCreateFromCSC,
         (Ptr{UInt64}, Ptr{UInt32}, Ptr{Float32}, UInt64, UInt64, Ref{Ptr{Void}}),
         convert(Array{UInt64, 1}, data.colptr - 1),
-        convert(Array{UInt32, 1}, data.rowval - 1), data.nzval,
+        convert(Array{UInt32, 1}, data.rowval - 1), convert(Array{Float32, 1}, data.nzval),
+        convert(UInt64, size(data.colptr)[1]),
+        convert(UInt64, nnz(data)),
+        handle
+    )
+    return handle[]
+end
+
+function XGDMatrixCreateFromCSCT(data::SparseMatrixCSC)
+    handle = Ref{Ptr{Void}}()
+    @xgboost_ccall(
+        :XGDMatrixCreateFromCSR,
+        (Ptr{UInt64}, Ptr{UInt32}, Ptr{Float32}, UInt64, UInt64, Ref{Ptr{Void}}),
+        convert(Array{UInt64, 1}, data.colptr - 1),
+        convert(Array{UInt32, 1}, data.rowval - 1), convert(Array{Float32, 1}, data.nzval),
         convert(UInt64, size(data.colptr)[1]),
         convert(UInt64, nnz(data)),
         handle

test/runtests.jl

@@ -3,6 +3,20 @@ using FactCheck
 
 include("utils.jl")
 
+facts("Sparse matrices") do
+    X = sparse(randn(100,10) .* bitrand(100,10))
+    y = randn(100)
+    DMatrix(X, label=y)
+
+    X = sparse(convert(Array{Float32,2}, randn(10,100) .* bitrand(10,100)))
+    y = randn(100)
+    DMatrix(X, true)
+
+    X = sparse(randn(100,10) .* bitrand(100,10))
+    y = randn(100)
+    DMatrix(X)
+end
+
 facts("DMatrix loading") do
     dtrain = DMatrix("../data/agaricus.txt.train")
     train_X, train_Y = readlibsvm("../data/agaricus.txt.train", (6513, 126))