HandyFunctions.markdown

Handy Functions

Predictions

When using Vision you will receive the model's predictions as an array of VNClassificationObservation or VNCoreMLFeatureValueObservation objects.

To get the top-5 classification predictions, do:

if let observations = request.results as? [VNClassificationObservation] {
  let top5 = top(5, observations)
  print(top5)
}

If you're using Core ML directly, then classifications are given as a [String: Double] dictionary. To get the top 5:

if let prediction = try? model.prediction(data: pixelBuffer) {
  let top5 = top(5, prediction.classLabelProbs)
  print(top5)
}

where classLabelProbs is the name of the model's output with the dictionary (this may be different depending on the model).

MLMultiArray reshaping and transposing

Reshaping means you change the MLMultiArray's shape so that the dimensions get different sizes. It also lets you add or remove dimensions. However, the actual data stays the same; reshaping is just another way of looking at it.

For example, assume that multiArray has a shape of [1, 3, 200, 150, 1]. Then you can get rid of these dimensions of size 1 as follows:

withExtendedLifetime(multiArray) {
  if let reshapedArray = try? multiArray.reshaped(to: [3, 200, 150]) {
    /* do something with the reshaped array */
  }
}

Important: The reshaped() method returns a new MLMultiArray object that uses the data pointer and data type of the original array. The caller is responsible for keeping the original object alive, for example using withExtendedLifetime.

Transposing or permuting also changes the dimensions, but unlike reshaping actually moves the data around in memory.

For example, given a multi-array of shape [3, 200, 150] you could turn it into a [200, 150, 3] array by writing:

if let transposedArray = try? multiArray.transposed(to: [1, 2, 0]) {
  /* do something with the transposed array */
}

Note that transposing makes a copy of the data and so you can let the original array be deallocated without problems.

Other features

New Array functions:

• argmax(): returns the largest element in the array, as well as its index
• argsort(): returns the indices of the array's elements in sorted order
• gather(): creates a new array containing the elements at the specified indices

Machine learning math functions:

• sigmoid(): logistic sigmoid on scalars or vectors
• softmax()