SwiftTFRecords

TFRecords (.tfrecord) file format builder and reader for Swift

The TFRecords format is briefly documented here, and described as the recommended format for feeding data into TenosorFlow here and here.

This library facilitates producing and importing data in the TFRecords format directly in Swift.

The library is not "official" - it is not part of TensorFlow, and it is not maintained by the TensorFlow team.

Usage - Produce a TFRecords file

The example below covers recommended API usage for generating a TFRecords file.

import SwiftTFRecords

var record = Record()

record["Int"] = 1
record["Float"] = 2.3
record["String"] = "Jacopo 😃"
record["FloatArray"] = [2.1, 2.2, 2.3]
record["Bytes"] = Feature.Bytes(Data([1, 2, 3, 4])
record["IntArray"] = Feature.IntArray([1, 2, 3, 4])

let tfRecord = TFRecords(withRecords: [record])

tfRecord.data.write(to: URL(fileURLWithPath: "file.tfrecord"))

Usage - Import a TFRecords file

The example below covers recommended API usage for read a TFRecords file.

import SwiftTFRecords

let data = try Data(contentsOf: URL(fileURLWithPath: "file.tfrecord"))

let tfRecord = TFRecords(withData: data)

for record in tfRecord.records {
    print("---")
    print(record["Int"]?.toInt())
    print(record["Float"]?.toFloat())
    print(record["String"]?.toString())
    print(record["FloatArray"]?.toFloatArray())
    print(record["Bytes"]?.toBytes())
    print(record["IntArray"]?.toIntArray())
}