Data Science from Scratch Implemented in Swift

Data Science from Scratch
First Principles with Python Swift
Joel Grus

Book: https://www.amazon.com/Data-Science-Scratch-Principles-Python/dp/1492041130/

Book Python Source: https://github.com/joelgrus/data-science-from-scratch

--

• Chapter 4: Linear Algebra
• Chapter 5: Statistics
• Chapter 6: Probability
• Chapter 7: Hypothesis and Inference
• Chapter 8: Gradient Descent
• Chapter 14: Simple Linear Regression

TODO

• Chapter 12: k-Nearest Neighbors
• Chapter 13: Naive Bayes
• Chapter 15: Multiple Regression
• Chapter 16: Logistic Regression

Overview

Reimplementing in Swift because it's a great modern language. Any language will do, of course. However, if I did it with Python, I'd probably cheat a little and I would definitely not think about what I'm doing as much.

Misc Thoughts and Issues

• Sticking with snake_case variable names for consistency
• Implemented some functions as Array extensions in Chapter 5.
  • Which is better variance(xs) vs xs.variance?
• Used FloatingPoint protocol so I had to force type to float by adding .0 in some instances.
• Type issues: Tried to use FloatingPoint protocol to make functions more generic. Had problems casting Element results to Float.
• Needed two linear algebra functions from Chapter 4: dot(), sum_of_squares(). I simply copied them into the statistics.swift file.
• Also, copied the Vector type.
• Need a plotting library.

Python's List Comprehensions

Pyhon's List Comprehension are a nice succinct feature. In Swift, you'll need to replace them with a combination of map, filter, forEach, for, etc I've rewritten some of the examples on page 30 in Swift:

Python

even_numbers = [x for x in range(5) if x % 2 == 0]  # [0, 2, 4]
squares      = [x * x for x in range(5)]            # [0, 1, 4, 9, 16]
even_squares = [x * x for x in even_numbers]        # [0, 4, 16]


assert even_numbers == [0, 2, 4]
assert squares == [0, 1, 4, 9, 16]
assert even_squares == [0, 4, 16]

square_dict = {x: x * x for x in range(5)}  # {0: 0, 1: 1, 2: 4, 3: 9, 4: 16}
square_set  = {x * x for x in [1, -1]}      # {1}


assert square_dict == {0: 0, 1: 1, 2: 4, 3: 9, 4: 16}
assert square_set == {1}

zeros = [0 for _ in even_numbers]      # has the same length as even_numbers


assert zeros == [0, 0, 0]

Swift

let even_numbers = (0..<5).filter {$0 % 2 == 0}
let squares = (0..<5).map {$0 * $0}
let even_squares = (0..<5).map {$0 * $0}.filter {$0 % 2 == 0}

var square_dict:[Int:Int] = [:]
(0..<5).forEach {square_dict[$0] = $0 * $0}

let zeros = even_numbers.map {_ in 0}

Simple implementation of Python's Counter (p26,66) class in Swift.

class Counter<T:Hashable> {
   public var dict:[T:Int] = [:]
   public let maxCount:Int
   
   init(_ xs:[T]) {
       var max = 0
       
       for elem in xs {
           if let n = dict[elem] {
               dict[elem] = n + 1
               if n + 1 > max {
                   max = n + 1
               }
           } else {
               dict[elem] = 1
           }
       }
       
       maxCount = max
   }
   
   public func mode_values() -> [T] {
       var xs:[T] = []
       for (k,v) in dict {
           if v == maxCount {
               xs.append(k)
           }
       }
       return xs
   }
}

Next I'll continue to work through the book. Not sure if I'll do it in order.