Initial commit.

monoid-design-in-class.code-workspace

+{
+	"folders": [
+		{
+			"path": "."
+		}
+	],
+	"settings": {
+		"files.eol": "\n",
+		"files.exclude": {
+			"**/node_modules": true
+		},
+		"files.trimFinalNewlines": true,
+		"files.trimTrailingWhitespace": true
+	}
+}

monoid-design.md

+# Monoids
+
+*   Monoid:
+    *   `M = (S, ∙, e)`
+
+*   Parts:
+    *   Set of elements: `S` (actions that can always be taken)
+    *   Binary operator: `∙` (do one action followed by the other)
+    *   Identity element: `e` (the do-nothing action)
+
+*   Restrictions:
+    *   `e ∈ S` (one of the actions is the do-nothing action)
+    *   `∙: S × S → S` (for every sequence of actions, there is an equivalent single action)
+
+*   Monoid Laws:
+    *   Associativity: `∀a.∀b.∀c. (a∙b)∙c = a∙(b∙c)` (grouping does not matter)
+    *   Two-sided identity: `∀a. e∙a = a∙e = a` (the do-nothing action does nothing)
+
+--------------------------------------------------------------------------------
+
+# Example: The Lightswitch Monoid
+
+*   Monoid:
+    *   `L = ({HOLD, OFF, ON, FLIP}, ∙, HOLD)`
+
+*   Parts:
+    *   Set of elements: `{HOLD, OFF, ON, FLIP}`
+    *   Binary operator: `∙`
+    *   Identify element: `HOLD`
+
+*   Combination Table:
+
+    ```
+        ∙ HOLD  OFF   ON    FLIP
+         \----------------------
+    HOLD |HOLD  OFF   ON    FLIP
+    OFF  |OFF   OFF   ON    ON
+    ON   |ON    OFF   ON    OFF
+    FLIP |FLIP  OFF   ON    HOLD
+    ```
+
+*   Monoid Laws:
+    *   Associativity: `∀a.∀b.∀c. (a∙b)∙c = a∙(b∙c)`
+    *   Two-sided identity: `∀a. HOLD∙a = a∙HOLD = a`
+
+--------------------------------------------------------------------------------
+
+# Reducers
+
+*   Example Reducer:
+    *   `f(A) = fold(∙, HOLD, A)`
+
+*   Examples Reducer Evaluation using a Left Fold:
+    *   `f([OFF, ON, FLIP, HOLD])    = HOLD ∙ OFF ∙ ON ∙ FLIP ∙ HOLD    = OFF`
+    *   `f([FLIP, HOLD, FLIP, FLIP]) = HOLD ∙ FLIP ∙ HOLD ∙ FLIP ∙ FLIP = FLIP`
+    *   `f([])                       = HOLD                             = HOLD`
+
+## Iterative Reducers
+
+```
+function iterativeReduce(monoid, sequence) {
+  let result = monoid.identity;
+  for (const element of sequence) {
+    result = monoid.combine(result, element);
+  }
+  return result;
+}
+```
+
+*   Example Evaluation: `((((HOLD ∙ FLIP) ∙ HOLD) ∙ FLIP) ∙ FLIP) = FLIP`
+
+## Recursive Reducers
+
+```
+function recursiveReduce(monoid, sequence) {
+  if (sequence.length === 0) {
+    return monoid.identity;
+  }
+  if (sequence.length === 1) {
+    return sequence[0];
+  }
+  const middle = Math.floor(sequence.length / 2);
+  const left = recursiveReduce(monoid, new View(sequence, 0, middle));
+  const right = recursiveReduce(monoid, new View(sequence, middle, sequence.length));
+  return monoid.combine(left, right);
+}
+```
+
+*   Example Recursion:
+
+    ```
+                 In: [OFF ∙ ON ∙ FLIP ∙ HOLD]
+                 Out: OFF
+                    /               \
+                   /                 \
+        In:  [OFF, ON]              In: [FLIP, HOLD]
+        Out: ON                     Out: FLIP
+           /       \                   /       \
+          /         \                 /         \
+    In:  [OFF]     In: [ON]     In:  [FLIP]    In: [HOLD]
+    Out: OFF       Out: ON      Out: FLIP      Out: HOLD
+    ```
+
+*   Example Evaluation: `(FLIP ∙ HOLD) ∙ (FLIP ∙ FLIP) = FLIP`
+
+## Parallel Reducers (Hybrid Idea)
+
+*   Pictorially:
+
+    ```
+                                Thread 0
+                                In: Entire List
+                               //\\
+
+                                ……
+
+           /              /            \                    \
+          /              /              \                    \
+    Thread 1        Thread 2        Thread 3        …    Thread n
+    In: 1st Part    In: 2nd Part    In: 3rd Part    …    In: nth Part
+    ```
+
+*   In Code:
+
+    ```
+    async function parallelReduce(monoid, sequence, threadCount) {
+      const jobs = [];
+      for (let threadIndex = 0; threadIndex < threadCount; ++threadIndex) {
+        const low = Math.floor(threadIndex * sequence.length / threadCount);
+        const high = Math.floor((threadIndex + 1) * sequence.length / threadCount);
+        const job = dispatch(monoid, sequence.slice(low, high));
+        jobs.push(job);
+      }
+      return iterativeReduce(monoid, await Promise.all(jobs));
+    }
+    ```
+
+--------------------------------------------------------------------------------
+
+# Parallel List Summation
+
+Problem: Given a list of numbers, compute their total.
+
+## Monoid Identification
+
+*   Interpretation of input elements as actions:
+    *   …
+*   Combination of two actions:
+    *   …
+*   Representation of actions:
+    *   …
+*   Identity element:
+    *   …
+
+*   Monoid:
+    *   `A = (…, …, …)`
+
+--------------------------------------------------------------------------------
+
+# Parallel List Multiplication
+
+Problem: Given a list of numbers, compute their product.
+
+## Monoid Identification
+
+*   Interpretation of input elements as actions:
+    *   …
+*   Combination of two actions:
+    *   …
+*   Representation of actions:
+    *   …
+*   Identity element:
+    *   …
+
+*   Monoid:
+    *   `A = (…, …, …)`
+
+--------------------------------------------------------------------------------
+
+# Parallel Minimization
+
+Problem: Given a list of numbers, compute the minimum.
+
+## Monoid Identification
+
+*   Interpretation of input elements as actions:
+    *   …
+*   Combination of two actions:
+    *   …
+*   Representation of actions:
+    *   …
+*   Identity element:
+    *   …
+
+*   Monoid:
+    *   `A = (…, …, …)`
+
+--------------------------------------------------------------------------------
+
+# Parallel Averaging
+
+Problem: Given a list of numbers, compute the average.
+
+## Obstacle
+
+It is not clear how to interpret the input elements as actions.
+
+As another way to look at the problem, if we know the average of a left-hand side and the average of a right-hand side, there is no reliable way to combine those averages.  For example:
+
+*   The average of `[1, 5]` is `3`, and the average of `[6]` is `6`, so `3 ∙ 6` should be the average of `[1, 5, 6]`, or `4`.
+*   But the average of `[3]` is `3`, and the average of `[4, 8]` is `6`, so `3 ∙ 6` should also be the average of `[3, 4, 8]`, or `5`.
+
+This is the same issue we saw in divide-and-conquer design: we must be asking the wrong question.
+
+If we ask for …, deferring …, then we can find a monoid.
+
+## Monoid Identification
+
+*   Interpretation of input elements as actions:
+    *   …
+*   Combination of two actions:
+    *   …
+    *   …
+*   Representation of actions:
+    *   …
+*   Identity element:
+    *   …
+
+*   Monoid:
+    *   `A = (…, ∙, …)` where `(a, b) ∙ (c, d) = (…, …)`
+
+--------------------------------------------------------------------------------
+
+# Parallel Parentheses Balancing
+
+Problem: Given a string of parentheses, determine if the parentheses are balanced.
+
+*   Example of balanced parentheses: `()(())`
+*   Example of unbalanced parentheses: `…`
+*   Example of unbalanced parentheses with same number of each type: `…`
+
+## Clues about the Monoid from an Iterative Design
+
+```
+function isBalanced(parentheses) {
+  let nesting = 0;
+  for (const character of parentheses) {
+    if (character === '(') {
+      ++nesting;
+    } else if (character === ')') {
+      --nesting;
+      if (nesting < 0) {
+        return false;
+      }
+    }
+  }
+  return nesting === 0;
+}
+```
+
+## Clues about the Monoid from Generated Submonoids
+
+*   Taking the submonoid generated by `(`:
+    *   The identity is different than `(`, which is different than `((`, which is different than `(((`, etc.
+    *   So it seems like our monoid will need ….
+
+*   Taking the submonoid generated by `)`:
+    *   The identity is different than `)`, which is different than `))`, which is different than `)))`, etc.
+    *   So it seems like our monoid will need ….
+
+## Quotienting a Free Monoid by Relations
+
+*   The sequence `…` cancels to the identity.
+*   If we simplify any input by repeatedly removing occurrences of `()`, we eventually end up with a string of the form ….
+
+## Parentheses Monoid
+
+*   Monoid:
+    *   `P = (…, ∙, …)` where `(a, b) ∙ (c, d) = (…, …)`
+
+## Divide-and-Conquer Design
+
+```
+                         In:  '()(())'
+                         Out:
+                       /            \
+                      /              \
+                     /                \
+           In:  '()('                  In:  '())'
+           Out:                        Out:
+             /  \                        /  \
+            /    \                      /    \
+    In: '('       In:  ')('     In: '('       In: '))'
+    Out:          Out:          Out:          Out:
+                    /  \                        /  \
+                   /    \                      /    \
+           In: ')'       In:  '('      In: ')'       In: ')'
+           Out:          Out:          Out:          Out:
+```
+
+## New Iterative Solution
+
+```
+function encode(character) {
+  if (character === '(') {
+    return […, …];
+  }
+  if (character === ')') {
+    return […, …];
+  }
+  return [0, 0];
+}
+
+function combine([leftCloses, leftOpens], [rightCloses, rightOpens]) {
+  const cancellations = …;
+  return […, …];
+}
+
+function iterativeReduce(parentheses) {
+  let result = …;
+  for (const element of …) {
+    result = combine(result, element);
+  }
+  return result;
+}
+```
+
+## Parallel Solution
+
+```
+async function parallelReduce(parentheses, threadCount) {
+  const jobs = [];
+  for (let threadIndex = 0; threadIndex < threadCount; ++threadIndex) {
+    const low = Math.floor(threadIndex * parentheses.length / threadCount);
+    const high = Math.floor((threadIndex + 1) * parentheses.length / threadCount);
+    const job = dispatch(parentheses.slice(low, high));
+    jobs.push(job);
+  }
+  const [closes, opens] = iterativeReduce(await Promise.all(jobs));
+  return …;
+}
+```
+
+*   This type of parallel algorithm is called a *map-reduce* algorithm.
+*   (Modern map-reduce frameworks also support a *shuffle step* for rearranging
+    data between the map and reduce steps.)