KUnet->Knet

LICENSE.md

@@ -1,4 +1,4 @@
-The KUnet.jl package is licensed under the MIT "Expat" License:
+The Knet.jl package is licensed under the MIT "Expat" License:
 
 > Copyright (c) 2015: Deniz Yuret.
 >

examples/adding.jl

@@ -4,7 +4,7 @@
 # Usage: julia adding.jl [opts], use --help for a full list of opts.
 
 using ArgParse
-using KUnet
+using Knet
 import Base: start, next, done
 include("irnn.jl")
 include("lstm.jl")
@@ -244,8 +244,8 @@ main()
 # this way forw calculation by the last three layers are wasted
 # but this way we are testing the general input/output
 
-# import KUnet: forw, back, ninputs, param, similar!, gpu, initforw, initback, setparam!, update, loss, axpy! # push, pop, get1
-# import KUnet: backprop, train, predict, nz
+# import Knet: forw, back, ninputs, param, similar!, gpu, initforw, initback, setparam!, update, loss, axpy! # push, pop, get1
+# import Knet: backprop, train, predict, nz
 # include("../src/rnn.jl")
 
 # setparam!(net1; nesterov=0.01)

examples/linreg.jl

@@ -1,4 +1,4 @@
-using KUnet
+using Knet
 
 # Simple linear regression.
 

examples/lstm.jl

@@ -29,7 +29,7 @@ end
 
 ### DEAD CODE
 
-# using KUnet
+# using Knet
 
 # function LSTM(n; fbias=0, dropout=0)
 #     Net(

examples/mnist2d.jl

@@ -1,7 +1,7 @@
 # Handwritten digit recognition problem from http://yann.lecun.com/exdb/mnist.
 
 using Base.Test
-using KUnet
+using Knet
 isdefined(:MNIST) || include("mnist.jl")
 include("mlp.jl")
 setseed(42)

examples/mnist4d.jl

@@ -2,8 +2,8 @@
 # 4-D convolution test
 
 using Base.Test
-using KUnet
-using KUnet: params, isapprox2
+using Knet
+using Knet: params, isapprox2
 isdefined(:MNIST) || include("mnist.jl")
 setseed(42)
 nbatch=100

examples/mnistpixels.jl

@@ -2,8 +2,8 @@
 # N., & Hinton, G. E. (2015). A Simple Way to Initialize Recurrent
 # Networks of Rectified Linear Units. arXiv preprint arXiv:1504.00941.
 
-using KUnet
-using KUnet: nextidx
+using Knet
+using Knet: nextidx
 import Base: start, next, done
 using ArgParse
 include("irnn.jl")

examples/mnistsparse.jl

@@ -2,7 +2,7 @@
 # Testing sparse arrays.
 
 using Base.Test
-using KUnet
+using Knet
 include("mlp.jl")
 
 isdefined(:MNIST) || include("mnist.jl")

examples/predict.jl

@@ -1,6 +1,6 @@
 using HDF5
 using ArgParse
-using KUnet
+using Knet
 
 function main()
     s = ArgParseSettings()
@@ -23,7 +23,7 @@ function main()
         default = 128
     end
     args = parse_args(s)
-    KUnet.gpu(!args["nogpu"])
+    Knet.gpu(!args["nogpu"])
     args["nogpu"] && blas_set_num_threads(20)
     x = h5read(args["x"], "/data")
     net = map(l->Op(l), split(args["net"],','))

examples/rnnlm.jl

@@ -5,7 +5,7 @@
 # Usage: julia rnnlm.jl ptb.train.txt ptb.valid.txt ptb.test.txt
 # Type julia rnnlm.jl --help for more options
 
-using KUnet, ArgParse
+using Knet, ArgParse
 import Base: start, next, done
 include("lstm.jl")
 

examples/s2c.jl

@@ -1,7 +1,7 @@
 # S2C: sequence to class model
 
-import KUnet: params, forw, back
-using KUnet: initback
+import Knet: params, forw, back
+using Knet: initback
 
 immutable S2C <: Model; net1; net2; params;
     S2C(a,b)=new(a,b,vcat(params(a),params(b)))

examples/train.jl

@@ -4,7 +4,7 @@
 
 using HDF5
 using ArgParse
-using KUnet
+using Knet
 
 function parse_commandline()
     s = ArgParseSettings()
@@ -79,7 +79,7 @@ end
 
 function main()
     args = parse_commandline()
-    KUnet.gpu(!args["nogpu"])
+    Knet.gpu(!args["nogpu"])
     args["nogpu"] && blas_set_num_threads(20)
     x = h5read(args["x"], "/data"); 
     y = h5read(args["y"], "/data"); 

examples/tutorial.jl

@@ -1,5 +1,5 @@
-require(Pkg.dir("KUnet/test/mnist.jl"))
-using KUnet
+require(Pkg.dir("Knet/test/mnist.jl"))
+using Knet
 using MNIST: xtrn, ytrn, xtst, ytst
 
 net = [Mmul(64), Bias(), Relu(),

src/KUnet.jl → src/Knet.jl

@@ -1,4 +1,4 @@
-module KUnet
+module Knet
 using Compat
 
 # Print date, expression and elapsed time after execution

src/nettest.jl

@@ -1,6 +1,6 @@
 using CUDArt, CUSPARSE, CUDNN, CUBLAS
-#using KUnet, 
-#import KUnet: forw, back, loss, ninputs, overwrites, back_reads_x, back_reads_y, gpu, axpb!, @gpu, issimilar, mul2!
+#using Knet, 
+#import Knet: forw, back, loss, ninputs, overwrites, back_reads_x, back_reads_y, gpu, axpb!, @gpu, issimilar, mul2!
 #using Base.LinAlg: axpy!, scale!
 
 include("util/gpu.jl")
@@ -155,35 +155,35 @@ end
 # # (6,Par)               8a00    1e00	(10,)	
 # # (7,Add,3,5)           8800*	1c00+	(10,5)	
 # # (8,Add,6,7)           8800*	1c00+	(10,5)	
-# # (9,KUnet.Relu,8)	8c00    1c00+	(10,5)	tosave
+# # (9,Knet.Relu,8)	8c00    1c00+	(10,5)	tosave
 # # (10,Par)              8e00    1a00	(10,3)	
 # # (11,Dot,10,1)         8800*	1200*	(10,5)	
 # # (12,Par)              9000    1600	(10,10)	
 # # (13,Dot,12,38)        8200+	1200*	(10,5)	
 # # (14,Par)              9400    1400	(10,)	
 # # (15,Add,11,13)        8200+	1200*	(10,5)	
 # # (16,Add,14,15)        8200+	1200*	(10,5)	
-# # (17,KUnet.Relu,16)	9600    1200*	(10,5)	tosave
+# # (17,Knet.Relu,16)	9600    1200*	(10,5)	tosave
 # # (18,Par)              9800    1000	(10,3)	
 # # (19,Dot,18,1)         8200+	0800!	(10,5)	
 # # (20,Par)              9a00    0c00	(10,10)	
 # # (21,Dot,20,38)        8800*	0800!	(10,5)	
 # # (22,Par)              9e00    0a00	(10,)	
 # # (23,Add,19,21)        8800*	0800!	(10,5)	
 # # (24,Add,22,23)        8800*	0800!	(10,5)	
-# # (25,KUnet.Relu,24)	a000    0800!	(10,5)	tosave
+# # (25,Knet.Relu,24)	a000    0800!	(10,5)	tosave
 # # (26,Par)	        a200    0600	(10,3)	
 # # (27,Dot,26,1)         8800*	fe00@	(10,5)	
 # # (28,Par)              a400    0200	(10,10)	
 # # (29,Dot,28,38)        8200+	fe00@	(10,5)	
 # # (30,Par)              a800    0000	(10,)	
 # # (31,Add,27,29)        8200+	fe00@	(10,5)	
 # # (32,Add,30,31)        8200+	fe00@	(10,5)	
-# # (33,KUnet.Relu,32)	aa00    fe00@	(10,5)	tosave
+# # (33,Knet.Relu,32)	aa00    fe00@	(10,5)	tosave
 # # (34,Mul,9,33)         8200+	fc00	(10,5)	
 # # (35,Mul,17,36)        8800*	fa00	(10,5)	
 # # (36,Add,34,35)	ac00    f800	(10,5)	tosave,toincr,tozero,tmp=f400
-# # (37,KUnet.Tanh,36)	ae00    f600	(10,5)	tosave
+# # (37,Knet.Tanh,36)	ae00    f600	(10,5)	tosave
 # # (38,Mul,37,25)	b000	f200	(10,5)	tosave,toincr,tozero,tmp=f400
 
 # # ops: 38

src/util/gpu.jl

@@ -2,7 +2,7 @@
 # loaded, not whether it is used.  The user can control gpu use by
 # using the gpu() function.
 GPU = true
-lpath = [Pkg.dir("KUnet/src")]
+lpath = [Pkg.dir("Knet/src")]
 for l in ("libkunet", "libcuda", "libcudart", "libcublas", "libcudnn")
     isempty(Libdl.find_library([l], lpath)) && (warn("Cannot find $l");GPU=false)
 end
@@ -35,7 +35,7 @@ macro gpu(_ex); if GPU; esc(_ex); end; end
 GPU && CUDArt.init!([CUDArt.CuModule(),], [CUDArt.device(),])
 
 # Additional cuda code
-const libkunet = Libdl.find_library(["libkunet"], [Pkg.dir("KUnet/src")])
+const libkunet = Libdl.find_library(["libkunet"], [Pkg.dir("Knet/src")])
 
 # For debugging
 function gpumem()

src/util/linalg.jl

@@ -12,7 +12,7 @@ scale!{T}(a,x::CudaArray{T})=(scal!(length(x),convert(T,a),x,1); x)
 vecnorm(x::CudaArray)=nrm2(x)
 
 ### MMUL
-# This is not a complete implementation.  The goal is to support KUnet
+# This is not a complete implementation.  The goal is to support Knet
 # operations for sparse/dense matrices on cpu/gpu.  The operations needed:
 #
 # mmul forw: A_mul_B!(y, w, x)		A_mul_Bs!(y, w, x): cpu/gpu: kudense, array, sparse

test/cusparse.jl

@@ -1,4 +1,4 @@
-using CUDArt,CUBLAS,CUSPARSE,KUnet,Base.Test
+using CUDArt,CUBLAS,CUSPARSE,Knet,Base.Test
 include("isapprox.jl")
 
 csc2csr{T}(x::SparseMatrixCSC{T})=CudaSparseMatrixCSR{T}(CudaArray(convert(Vector{Cint},x.colptr)), CudaArray(convert(Vector{Cint},x.rowval)), CudaArray(x.nzval), (x.n,x.m), convert(Cint,length(x.nzval)), device())

test/isapprox.jl

@@ -1,4 +1,4 @@
-using CUDArt, KUnet
+using CUDArt, Knet
 import Base: isapprox
 
 function isapprox(x, y; 

test/rnntest.jl

@@ -1,8 +1,8 @@
 using Base.Test
 using CUDArt
-using KUnet
+using Knet
 
-# import KUnet: forw, back, ninputs, param, similar!, gpu, initforw, initback, push, pop, get1
+# import Knet: forw, back, ninputs, param, similar!, gpu, initforw, initback, push, pop, get1
 # include("../src/net.jl")
 
 include("isapprox.jl")
@@ -39,7 +39,7 @@ r = rnn
 inputs = Any[y]
 trn = true
 seq = false
-using KUnet: push, pop, dbg, forw, back, get1, initbatch
+using Knet: push, pop, dbg, forw, back, get1, initbatch
 
     initbatch(r, inputs...; trn=trn, seq=seq, a...)
     for i = 1:ninputs(r)

test/runtests.jl

@@ -1,5 +1,5 @@
 using Compat
-using KUnet
+using Knet
 @date include("testdense.jl")
 @date include("testsparse.jl")
 @date include("testconvert.jl")

test/testcolops.jl

@@ -1,4 +1,4 @@
-using KUnet, CUDArt, Base.Test
+using Knet, CUDArt, Base.Test
 
 iseq02(a,b)=(convert(Array,a)==convert(Array,b))
 

test/testconvert.jl

@@ -1,4 +1,4 @@
-using KUnet, CUDArt, Base.Test
+using Knet, CUDArt, Base.Test
 
 a=rand(3,5)
 

test/testdense.jl

@@ -1,6 +1,6 @@
 using Base.Test
 using CUDArt
-using KUnet
+using Knet
 
 for N in 1:5
     for A in (CudaArray, Array)

test/testkperceptron.jl

@@ -1,7 +1,7 @@
 using Base.Test
 using CUDArt
-using KUnet
-include(Pkg.dir("KUnet/test/mnist.jl"))
+using Knet
+include(Pkg.dir("Knet/test/mnist.jl"))
 sparse32{T}(a::Array{T})=convert(SparseMatrixCSC{T,Int32}, a)
 
 xtrn = MNIST.xtrn
@@ -39,7 +39,7 @@ for ker in (
                         )
                 # loc == :gpu && ker[1] == :perceptron && continue
                 println("\n$ker, $prc, $fmt, $loc")
-                KUnet.gpu(loc == :gpu)
+                Knet.gpu(loc == :gpu)
                 for p in (:xtrn, :xtst, :ytrn, :ytst); @eval $p=copy(MNIST.$p); end
                 prc == :double && (for p in (:xtrn, :xtst, :ytrn, :ytst); @eval $p=convert(Array{Float64},$p); end)
                 fmt == :sparse && (for p in (:xtrn, :xtst); @eval $p=sparse32($p); end)
@@ -48,7 +48,7 @@ for ker in (
 
                 net = (ker[1] == :perceptron ? 
                        Op[Mmul(nc;average=true,init=initzero), PercLoss()] :
-                       Op[KPerceptron(nc, KUnet.(ker[1]), ker[2])])
+                       Op[KPerceptron(nc, Knet.(ker[1]), ker[2])])
                 gc(); @date train(net, xtrn, ytrn; iters=niter,batch=nbatch)
                 @date a = accuracy(ytst, predict(net, xtst))
                 s = isdefined(net[1],:s) ? size(net[1].s) : 0
@@ -91,7 +91,7 @@ end
 # for i=1:2
 # ftrn = sparse(MNIST.xtrn)
 # ftst = sparse(MNIST.xtst)
-# @show fnet = Op[KPerceptron(10, KUnet.klinear0, [0f0])]
+# @show fnet = Op[KPerceptron(10, Knet.klinear0, [0f0])]
 #     gc()
 #     @date train(fnet, ftrn, ytrn; iters=niter,batch=nbatch)
 #     gc()
@@ -105,7 +105,7 @@ end
 # for i=1:2
 # ltrn = sparse(MNIST.xtrn)
 # ltst = sparse(MNIST.xtst)
-# @show lnet = Op[KPerceptron(10, KUnet.klinear, [0f0])]
+# @show lnet = Op[KPerceptron(10, Knet.klinear, [0f0])]
 # gc()
 #     @date train(lnet, ltrn, ytrn; iters=niter,batch=nbatch)
 # gc()
@@ -119,7 +119,7 @@ end
 # for i=1:2
 # qtrn = sparse(MNIST.xtrn)
 # qtst = sparse(MNIST.xtst)
-# @show qnet = Op[KPerceptron(10, KUnet.kgauss, [g0])]
+# @show qnet = Op[KPerceptron(10, Knet.kgauss, [g0])]
 #     @date train(qnet, qtrn, ytrn; iters=niter,batch=nbatch)
 #     @time println((i, size(qnet[1].s), 
 #                    accuracy(ytst, predict(qnet, qtst)),
@@ -131,7 +131,7 @@ end
 # for i=1:2
 # qtrn = sparse(MNIST.xtrn)
 # qtst = sparse(MNIST.xtst)
-# @show qnet = Op[KPerceptron(10, KUnet.kgauss0, [g0])]
+# @show qnet = Op[KPerceptron(10, Knet.kgauss0, [g0])]
 #     @date train(qnet, qtrn, ytrn; iters=niter,batch=nbatch)
 #     @time println((i, size(qnet[1].s), 
 #                    accuracy(ytst, predict(qnet, qtst)),
@@ -143,7 +143,7 @@ end
 # for i=1:2
 # ktrn = sparse(MNIST.xtrn)
 # ktst = sparse(MNIST.xtst)
-# @show knet = Op[KPerceptron(10, KUnet.kpoly, [c0,d0])]
+# @show knet = Op[KPerceptron(10, Knet.kpoly, [c0,d0])]
 #     @date train(knet, ktrn, ytrn; iters=niter,batch=nbatch)
 #     @time println((i, size(knet[1].s), 
 #                    accuracy(ytst, predict(knet, ktst)),
@@ -155,7 +155,7 @@ end
 # for i=1:2
 # ktrn = sparse(MNIST.xtrn)
 # ktst = sparse(MNIST.xtst)
-# @show knet = Op[KPerceptron(10, KUnet.kpoly, [c0,d0])]
+# @show knet = Op[KPerceptron(10, Knet.kpoly, [c0,d0])]
 #     @date train(knet, ktrn, ytrn; iters=niter,batch=nbatch)
 #     @time println((i, size(knet[1].s), 
 #                    accuracy(ytst, predict(knet, ktst)),
@@ -220,7 +220,7 @@ end
 # if false; info("KPerceptron+kgauss")
 # qtrn = sparse(MNIST.xtrn)
 # qtst = sparse(MNIST.xtst)
-# @show qnet = Op[KPerceptron(10, KUnet.kgauss, [g0])]
+# @show qnet = Op[KPerceptron(10, Knet.kgauss, [g0])]
 # for i=1:1
 #     @date train(qnet, qtrn, ytrn; iters=niter,batch=nbatch)
 #     @time println((i, size(qnet[1].s), 
@@ -245,7 +245,7 @@ end
 # end; end
 
 # if false; info("Poly+PercLoss+GPU+dense") # mmul, hcat, ctranspose do not work
-# KUnet.gpu(true)
+# Knet.gpu(true)
 # cnet = Op[Poly(c=c0,d=d0,w=CudaArray(w0)), PercLoss()]
 # for i=1:1
 #     @time train(cnet, xtrn, ytrn; iters=niter,batch=nbatch)