https://github.com/vkuznet/swiftmlexample

https://github.com/vkuznet/swiftmlexample

Last synced: 30 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/vkuznet/swiftmlexample
• Owner: vkuznet
• License: mit
• Created: 2020-08-26T17:56:47.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-08-31T19:38:39.000Z (over 4 years ago)
• Last Synced: 2024-10-30T06:27:33.364Z (3 months ago)
• Language: Swift
• Size: 20.5 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
## SwiftMLExample

Repository with end-to-end ML example based on

[TF Swift tutorial](https://www.tensorflow.org/swift/tutorials/model_training_walkthrough).

We extend the tutorial to provide example how to wrap up everything

into single package, organize codebase into separate modules,

add external arguments, etc.

### Build notes

Swift provides [package manager](https://swift.org/getting-started/#using-the-package-manager)

which helps to setup initial project area. To do that just create a new

directory and run within it the following command:

```

# initalize the new package (it is already done here)

swift package init --type executable

```

This will initialize the project and create Package.swift file which

you can later customize to include your set of dependencies, etc.

This is already done for this package.

To maintain the codebase you need to use the following commands (within

this project area):

```

# build/compile our codebase (the entire build will be located in .build area)

swift build

# run our executable

swift run swift-ml --help

# run our training for 300 epochs using 64 batch size and save model to model.tf file

swift run swift-ml train -e 300 --batch-size 64 --model-filename model.tf

# run test ML action, i.e. load the model and make predictions

swift run swift-ml test --model-filename model.tf

# run MNIST training

swift run swift-ml mnist

# clean-up our build

swift package clean

# create full release

swift build --configuration release

# grab new executable from release area and put it into provide path

cp .build/release/swift-ml /path

```

Further customization can be done using swift

[Mint](https://github.com/yonaskolb/Mint) package.