Two sides of this video

Introducing the F# language

  • Some practical information
  • Enough so that you can use it!
  • Some important things omitted

Introducing ML-style languages

  • Background for our TinyML
  • Basic features & principles
  • Some weird corner cases!

Getting started with F#

F# and .NET runtime

F# editors and tools

  • Microsoft Visual Studio (Win only)
  • JetBrains Rider (Win, Linux, Mac)
  • VS Code with Ionide (Win, Linux, Mac)
  • https://ionide.io

F# project types

Script-based development

  • Write code in .fsx file
  • Run using F# Interactive REPL
  • Can select & run out-of-order!

Project-based development

  • Project .fsproj with .fs sources
  • Standard build and run workflow
  • Live reload with Fable and JavaScript

F# project structure

Declaration order matters!

  • Helper function and types
  • Types defining domain model
  • Operations for working with it
  • User interface

How to organize F# projects

  • Namespaces or modules
  • Type declarations
  • Functions (inside modules)

Data type declarations in F#

  • Tuples and records
    Store multiple values of different types
  • Discriminated unions
    Represent one of multiple possible options
  • Collections, lists and maps
    Multiple values of the same type
  • Recursive declarations
    Type that can include values of itself
  • Type aliases
    Shorthand for a type with a long name

Demo

Simple expression evaluator

Selected advanced features

Lists and maps

  • Immutable collections
  • Linked (cons) lists with head/tail
  • Key-value maps with lookup

Recursion and laziness

  • Recursive functions using let rec
  • Works also for values, but beware!
  • Lazy<T> to represent lazy values

Demo

Maps, lazy values, recursion

(* Functions *)
let f = (fun x -> 10 + x)
f 32

(* Tuples *)
let t = (1, "hi")
fst t
snd t

(* Unions *)
let c1 = Case1(10)
let c2 = Case2(32)
match c1 with
| Case1 n -> n + 32
| Case2 n -> n + 10

Language features of TinyML (1/2)

Functions but only with single argument

Tuples of two element with getters

Unions without tag name with two cases

(* Let bindings *)
let x = 10 in x * 32

(* Let desugaring *)
(fun x -> x * 32) 10

(* Conditionals *)
if e then 10 else 32

(* Both are expressions *)
1 + (if e then 41 else 1)
1 + (let x = 1 in x + x)

(* Currying *)
let add = fun a -> fun b -> a + b
in (add 10) 32

Language features of TinyML (2/2)

let is a syntactic sugar

Everything (if and let too) is an expression

Functions that return functions (currying) if you need multiple parameters

Variable scoping

Lexical

Based on static block structure in code

Function value needs to capture variables (closure)

Dynamic

Based on dynamic evaluation structure

Operational semantics

Formally specify how expression evaluate

Substitution-based

We do not need variable context!

Call-by-name vs. call-by-value

Call-by-value (strict)

Evaluates function arguments first (ML)

Call-by-name (lazy)

Evaluates arguments when needed (Haskell)

type Expression =
  | Constant of int
  | Binary of
      string *
      Expression *
      Expression

val evaluate :
  Expression -> int

Basic interpreter structure (0/2)

Expression is the source
code that user writes

evaluate takes expression
and returns the result

type Value =
  | Number of int

type Expression =
  | Constant of int
  | Binary of
      string *
      Expression *
      Expression

val evaluate :
  Expression -> Value

Basic interpreter structure (1/2)

Adding values as the
result of evaluation

Value is what we
get as the result

evaluate takes expression
and returns value

type Value =
  | Number of int

type Expression =
  | Constant of int
  | Binary of   
      string *
      Expression *
      Expression
  | Variable of string

type VariableContext =
  Map<string, Value>

val evaluate :
  Expression -> VariableContext -> Value

Basic interpreter structure (2/2)

Adding variables and variable context

Variable can store only values (call-by-value)

evaluate takes context

Demo

Adding values and variables

Lab overview

TinyML interpreter step-by-step

TinyML - Basic tasks

  1. Simple numerical evaluator as the starting point
    This has already been done for you :-)

  2. Add unary operators (-) and conditional
    We only have numbers, so treat 1 as true

  3. Functions and application
    Tricky! Closure needs to capture variables!

  4. Let binding as syntactic sugar
    Evaluate let by treating it as apply/lambda

  5. Add a simple data type - tuples
    New value, constructor and destructor

TinyML - Bonus & super tasks

  1. Add more data types - unions
    New value, constructor and destructor (match)

  2. Add support for recursion
    Needs Lazy<Value> in variable context to work

  3. Add unit and create a list value
    Case1(Const(1), Case1(Const(2), Case2(Unit)))

  4. Implement call-by-name semantics
    Change variable context to store expressions

  5. Implement evaluation by substitution
    Toy approach, but you learn the semantics

Lessons learned

Functional language interpreter

  • Distinguishing Value and Expression
  • Recursive function with variable scope
  • Call-by-value and lexical variable scoping!
  • Nice constructor and destructor symmetry