Installation

• Required apache-arrow-glib 3.0.0+
• On Ubuntu 18.04, install Apache Arrow GLib, instructions from here:
  • sudo apt update
  • sudo apt install -y -V ca-certificates lsb-release wget -apache-arrow-glib wget https://apache.bintray.com/arrow/$(lsb_release --id --short | tr 'A-Z' 'a-z')/apache-arrow-archive-keyring-latest-$(lsb_release --codename --short).deb
  • sudo apt install -y -V ./apache-arrow-archive-keyring-latest-$(lsb_release --codename --short).deb
  • sudo apt update
  • sudo apt install -y -V libarrow-glib-dev
  • To see which version is installed: sudo apt list libarrow-glib-dev
• On macOS:
  • brew install apache-arrow-glib
  • To see which version is installed: brew list --versions apache-arrow-glib
• Generate and build the Swift files:
  • swift build
• Test:
  • swift test

Development

• The pkg-config file is at /usr/lib/x86_64-linux-gnu/pkgconfig/arrow.pc
• pkg-config --cflags arrow-glib
• Apache Arrow GLib interface documentation here
• Apache Arrow GLib example code here
• Note that the tests pass on Swift 5.3-RELEASE but do not build on Swift for Tensorflow 0.10. The library itself does successfully build on Swift for Tensorflow 0.10

Usage

Swift Arrays to/from .feather:

• Data types supported for array saving to and from .feather: Date, String, Double, Float, Int, Int64, and Bool.

let array: [Float] = (0..<10).map { _ in Float.random(in: 0.0...1.0) }
try array.toFeather(filePath: "./data/array.feather")
let arrayFromFeather = try [Float](fromFeather: "./data/array.feather")

PenguinTables to/from .feather:

• Data types supported as values inside PTables: Double, String, Int64, Bool

let row1: [Any] = ["e12fe9879b95b35479a1195bd2190b10", 2137.8 as Double, false]
let row2: [Any] = ["02528b1bca6c637a9d725488efa1de80", 2137.4 as Double, true]
let row3: [Any] = ["1de820d72a41bf02fdc55a8991797991", 879.5 as Double, true]
let rows = [row1, row2, row3]
let columnNames = ["id", "ask", "prohibited"]
let columns = rows.transposed()
let table = try PTable([columnNames[0]: PColumn(columns[0] as! [String]),
                        columnNames[1]: PColumn(columns[1] as! [Double]),
                        columnNames[2]: PColumn(columns[2] as! [Bool])])
assert(table.columnNames == columnNames.sorted())
let filePath = "tableMultiple.feather"
try table.toFeather(filePath: filePath)

let columnsDecoded = try PTable(fromFeather: filePath)
let doubleColumnDecoded: PTypedColumn<Double> = try columnsDecoded[columnNames[1]]!.asDType()
assert(doubleColumnDecoded[2] ==  879.5)
assert(columnsDecoded.columnNames == columnNames.sorted())

Further data types can be supported, the code just needs to be extended for them.