Add Mnist loader

src/deep_learning.cr

@@ -1,6 +1,13 @@
-# TODO: Write documentation for `DeepLearning`
+require "./mnist/**"
+
 module DeepLearning
   VERSION = "0.1.0"
 
   # TODO: Put your code here
 end
+
+images = File.open("./data/train-images.idx3-ubyte")
+labels = File.open("./data/train-labels.idx1-ubyte")
+
+loader = Mnist::Loader.new(images, labels)
+data = loader.call

src/mnist/loader.cr

@@ -0,0 +1,70 @@
+module Mnist
+  IMAGE_IDENTIFIER = 2051
+  LABEL_IDENTIFIER = 2049
+  ENDIAN           = IO::ByteFormat::BigEndian
+
+  alias Image = Array(UInt8)
+  alias Label = UInt8
+
+  struct Data
+    getter image, label
+
+    def initialize(@image : Image, @label : Label)
+    end
+  end
+
+  class LoaderException < Exception
+    def initialize(message = "The file doesn't have correct identifier")
+      super(message)
+    end
+  end
+
+  class ImageLoaderException < LoaderException
+  end
+
+  class LabelLoaderException < LoaderException
+  end
+
+  class MatchingSizeException < LoaderException
+    def initialize(message = "Image and Label count doesn't match")
+      super(message)
+    end
+  end
+
+  class Loader
+    @max_size : UInt32
+
+    def initialize(@images_io : IO, @labels_io : IO)
+      image_id = @images_io.read_bytes(UInt32, ENDIAN)
+      raise ImageLoaderException.new unless image_id == IMAGE_IDENTIFIER
+
+      label_id = @labels_io.read_bytes(UInt32, ENDIAN)
+      raise LabelLoaderException.new unless label_id == LABEL_IDENTIFIER
+
+      image_count = @images_io.read_bytes(UInt32, ENDIAN)
+      label_count = @labels_io.read_bytes(UInt32, ENDIAN)
+      raise MatchingSizeException.new unless image_count == label_count
+
+      @max_size = image_count
+    end
+
+    def call(wanted_size : UInt32 = @max_size) : Array(Data)
+      size = [wanted_size, @max_size].min
+      @images_io.seek(8)
+      @labels_io.seek(8)
+
+      row_size = @images_io.read_bytes(UInt32, ENDIAN)
+      col_size = @images_io.read_bytes(UInt32, ENDIAN)
+
+      image_slice = Bytes.new(row_size * col_size)
+
+      Array.new(size) do |i|
+        @images_io.read(image_slice)
+        image = image_slice.to_a
+        label = @labels_io.read_bytes(UInt8, ENDIAN)
+
+        Data.new(image, label)
+      end
+    end
+  end
+end

src/neural/network.cr

@@ -0,0 +1,4 @@
+module Neural
+  class Network
+  end
+end