Skip to content
This repository was archived by the owner on Apr 23, 2025. It is now read-only.

update mnist for new layers api #109

Merged
merged 3 commits into from
Apr 19, 2019
Merged

update mnist for new layers api #109

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9c2bf93
update mnist for new layers api
brettkoonce Apr 18, 2019
63ab5f4
remove public keyword
brettkoonce Apr 18, 2019
549a214
indentation
brettkoonce Apr 18, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 6 additions & 3 deletions MNIST/main.swift
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,9 @@ func readMNIST(imagesFile: String, labelsFile: String) -> (images: Tensor<Float>

/// A classifier.
struct Classifier: Layer {
typealias Input = Tensor<Float>
typealias Output = Tensor<Float>

var conv1a = Conv2D<Float>(filterShape: (3, 3, 1, 32), activation: relu)
var conv1b = Conv2D<Float>(filterShape: (3, 3, 32, 64), activation: relu)
var pool1 = MaxPool2D<Float>(poolSize: (2, 2), strides: (2, 2))
Expand All @@ -64,7 +67,7 @@ struct Classifier: Layer {
var layer1b = Dense<Float>(inputSize: 128, outputSize: 10, activation: softmax)

@differentiable
func applied(to input: Tensor<Float>) -> Tensor<Float> {
func call(_ input: Input) -> Output {
let convolved = input.sequenced(through: conv1a, conv1b, pool1)
return convolved.sequenced(through: dropout1a, flatten, layer1a, dropout1b, layer1b)
}
Expand All @@ -83,7 +86,7 @@ let (images, numericLabels) = readMNIST(imagesFile: "train-images-idx3-ubyte",
let labels = Tensor<Float>(oneHotAtIndices: numericLabels, depth: 10)

var classifier = Classifier()
let optimizer = RMSProp<Classifier, Float>()
let optimizer = RMSProp(for: classifier)

// The training loop.
for epoch in 1...epochCount {
Expand All @@ -95,7 +98,7 @@ for epoch in 1...epochCount {
let y = minibatch(in: numericLabels, at: i)
// Compute the gradient with respect to the model.
let 𝛁model = classifier.gradient { classifier -> Tensor<Float> in
let ŷ = classifier.applied(to: x)
let ŷ = classifier(x)
let correctPredictions = ŷ.argmax(squeezingAxis: 1) .== y
correctGuessCount += Int(Tensor<Int32>(correctPredictions).sum().scalarized())
totalGuessCount += batchSize
Expand Down