Merge pull request #418 from c-p-murphy/add-fashion-mnist

Add FashionMNIST

Author: MikeInnes

src/data/Data.jl

@@ -13,6 +13,9 @@ end
 include("mnist.jl")
 export MNIST
 
+include("fashion-mnist.jl")
+export FashionMNIST
+
 include("cmudict.jl")
 using .CMUDict
 

src/data/fashion-mnist.jl

@@ -0,0 +1,64 @@
+module FashionMNIST
+
+using ..MNIST: gzopen, imageheader, rawimage, labelheader, rawlabel
+
+const dir = joinpath(@__DIR__, "../../deps/fashion-mnist")
+
+function load()
+  mkpath(dir)
+  cd(dir) do
+    for file in ["train-images-idx3-ubyte",
+                 "train-labels-idx1-ubyte",
+                 "t10k-images-idx3-ubyte",
+                 "t10k-labels-idx1-ubyte"]
+      isfile(file) && continue
+      @info "Downloading Fashion-MNIST dataset"
+      download("http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/$file.gz", "$file.gz")
+      open(file, "w") do io
+        write(io, gzopen(read, "$file.gz"))
+      end
+    end
+  end
+end
+
+const TRAINIMAGES = joinpath(dir, "train-images-idx3-ubyte")
+const TRAINLABELS = joinpath(dir, "train-labels-idx1-ubyte")
+const TESTIMAGES = joinpath(dir, "t10k-images-idx3-ubyte")
+const TESTLABELS = joinpath(dir, "t10k-labels-idx1-ubyte")
+
+"""
+    images()
+    images(:test)
+
+Load the Fashion-MNIST images.
+
+Each image is a 28×28 array of `Gray` colour values (see Colors.jl).
+
+Returns the 60,000 training images by default; pass `:test` to retreive the
+10,000 test images.
+"""
+function images(set = :train)
+  load()
+  io = IOBuffer(read(set == :train ? TRAINIMAGES : TESTIMAGES))
+  _, N, nrows, ncols = imageheader(io)
+  [rawimage(io) for _ in 1:N]
+end
+
+"""
+    labels()
+    labels(:test)
+
+Load the labels corresponding to each of the images returned from `images()`.
+Each label is a number from 0-9.
+
+Returns the 60,000 training labels by default; pass `:test` to retreive the
+10,000 test labels.
+"""
+function labels(set = :train)
+  load()
+  io = IOBuffer(read(set == :train ? TRAINLABELS : TESTLABELS))
+  _, N = labelheader(io)
+  [rawlabel(io) for _ = 1:N]
+end
+
+end

test/data.jl

@@ -10,4 +10,7 @@ using Test
 @test MNIST.images()[1] isa Matrix
 @test MNIST.labels() isa Vector{Int64}
 
+@test FashionMNIST.images()[1] isa Matrix
+@test FashionMNIST.labels() isa Vector{Int64}
+
 @test Data.Sentiment.train() isa Vector{Data.Tree{Any}}