what is monad

functor

functor can be interpreted as A container that is mappable between categories.
endofunctor is a functor that mappable inside single category.
First of all, by saying container, it would definitely hold something inside, whether it is a single value or a set of value or nothing at all.
Another important property is mappable, which means the data inside the container can be transformed into another by a map function.

This is how we do the work:

1. grab the data from inside the container
2. use a mapping logic to transform all the data
3. put the transformed them into a new container

// we define a simplest class here
class Wrapper {
  // it does nothing but just hold the value and no else
  constructor (value) {
    this.value = value
  }
  // it provisions a function named "map", where the parameter f is a function
  map (f) {
    // f could transform the value to another one, before putting into another wrapper
    return new Wrapper(f(this.value))
  }
}

The usage can be as follow:

const a = Wrapper(1);
const f = (x) -> x + 3;
const b = a.map(f)

So we call Wrapper is a functor, we can place data into it and use map function to do data transformation.

function

function is basically a map between 2 objects

Mathematically speaking \(f(x) = y\)

where $f, x, y$ is called symbol $f$ is the mapping logic between input and output. $x$ is the input and $y$ is the output, we do not care about the type of these 2 symbols, they can be concrete value and can also be mapping logic(function)

This formation can be written into another one

Because $x, y$ can be function themselves, then it is called high-order function

curry

For function with multiple input, can use a technique named partial apply to reduct it into single input function, this is called curry, for instance:
Say we have an expression (+ a b) where both a and b are input, we can transform it into 2 partial function, and can be executed one by one in order to align with the original expression.

In clojure we can use partial function to currify the multi-input function

; use partial function separately
(def p (partial + a)) ;only use 1 input here to form a new function
(p b)                 ;provide the rest input to complete the invocation

; or combine into single expression
((partial + a) b)

curry is essentially the transformation of below:

\((a, b) \rightarrow c \equiv a \rightarrow (b \rightarrow c)\) This transform multi-input function into a single input function, which returns another function that is single input as well.

By mathematical convention in lambda calculus, the right arrow would associate first, so the formation above would become:

If we rewrite the formation above in clojure, in order to guarantee associativity:

(defn compose
  [a, b]
  (fn [p] (a (b p))))

; the expression below has associativity
(=
  (compose (compose a b) c)
  (compose a (compose b c)))

Let’s see a concrete example:

; define basic function
(defn add [x] (+ x 1))
(defn minus [x] (- x 2))
(defn multiply [x] (* x 3))

; define different association
(def p (compose (compose add minus) multiply))
(def q (compose add (compose minus multiply)))

; these 2 expressions should be exactly same
(=
  (p 1)
  (q 1))

In Clojure or Javascript, p and q are different functions, but they mathematically are exactly the same.

example

We will provide 2 examples to illustrate how and where monad comes from:

future

Given 2 functions as below that could do single work:

(defn get-user-by-id [id] (future ...id...))             ; str -> future<User>
(defn get-department-by-user [user] (future ...user...)) ; User -> future<Department>

These 2 functions are in this form:

It is the container Future that hold the data.

(defn compose [f g]
  (fn [x] (future (f @(g x)))))

(def get-department-by-user-id
  (compose get-department-by-user get-user-by-id))

(get-department-by-user-id 123)

Now the compose function is of type:

(User->Department)->(str->User)->(str->Department)

List

Given 2 functions like below:

(defn duplicate [x] [x x])
(defn positive [x] (if (pos? x) [x] []))

These 2 functions are in this form, $a \rightarrow [b]$.
If we rewrite the list symbol as another type, we have:

where List is the container and b is the value inside.

(defn compose [f g]
  ; notice we use mapcat function to flat & map
  (fn [x] (->> x (g) (mapcat f)))) ; execute from right to left

(def p (compose duplicate positive))

(->> [-1 1 3] (mapcat p))

monad

We observe that not only asynchronous future but also list follow the same pattern:

Where C is the mappable container, this is the functor that we mentioned above.
We can also try to use function composition, such as:

When we replace C with future we get asynchronous invocation function; when replace with list, we get list transformation function. If we use identity as C, then it is the regular function.

Then we want to define identity unit, because of associativity, the identity is as below:

(=
  (compose f identity)
  (compose identity f)
  (f))
; compose(f, unit) == f

We can see compose function is of type:

(a -> C b) -> (a -> C a) -> (a -> C b)

We can determine the type of identity: (a -> C a)

For future the identity is:

(defn identity [x] (future x))

For list the identity is:

(defn identity [x] [x])

We observe that both future and list has identity unit. Is this always the case?
The answer is true if we go with monad, as the definition of it is:

monad is a monoid in the category of endofunctors

monoid would of course has identity according to its definition, and it maps objects in one category.
Now let’s define function bind:

; bind :: m a -> (a -> m b) -> m b
(defn bind [ma f]
  (compose f (fn [_] ma)))