Tags SICP

Tree-recursion, yay !

Exercise 2.24

The interpretation is (1 (2 (3 4))). I’ll skip the drawings because they’re too time consuming for such a simple exercise.

Exercise 2.25

[22]> (setf lst '(1 3 (5 7) 9))
(1 3 (5 7) 9)
[23]> (car (cdr (car (cdr (cdr lst)))))
7

[24]> (setf lst '((7)))
((7))
[25]> (car (car lst))
7

[37]> (setf lst '(1 (2 (3 (4 (5 (6 7)))))))
(1 (2 (3 (4 (5 (6 7))))))
[38]> (cadr (cadr (cadr (cadr (cadr (cadr lst))))))
7

The last one is tricky, because cdr returns a list, so I use cadr (car of cdr) instead to get inside the lists.

Exercise 2.26

[39]> (setf x '(1 2 3))
(1 2 3)
[40]> (setf y '(4 5 6))
(4 5 6)
[41]> (append x y)
(1 2 3 4 5 6)
[42]> (cons x y)
((1 2 3) 4 5 6)
[43]> (list x y)
((1 2 3) (4 5 6))

Exercise 2.27

Similarly to count-leaves, we must differentiate between three cases:

1. nil 2. pair 3. not pair (atom)

The CL function for checking if something is a pair is consp[1]:

(defun deep-reverse (lst)
  (cond ((null lst) nil)
        ((consp (car lst))
          (append
            (deep-reverse (cdr lst))
            (list (deep-reverse (car lst))))) ; [A]
        (t 
          (append 
            (deep-reverse (cdr lst))
            (list (car lst))))))

Take a look at the line marked with [A]. The call to list is very important here, because otherwise append just stitches the contents of the list deep-reverse returns. list makes it append them as a single list, which is what we need.

Exercise 2.28

(defun fringe (lst)
  (cond ((null lst) nil)
        ((not (consp lst)) (list lst))
        (t
          (append 
            (fringe (car lst))
            (fringe (cdr lst))))))

Exercise 2.29

a.

(defun make-mobile (left right)
  (list left right))

(defun left-branch (mobile)
  (first mobile))

(defun right-branch (mobile)
  (second mobile))

(defun make-branch (len structure)
  (list len structure))

(defun branch-len (branch)
  (first branch))

(defun branch-structure (branch)
  (second branch))

I’m using CL’s convenience accessors for lists. first is equivalend to car, second to cadr (the car of the cdr). CL supports such accessors up to tenth, together with the generic accessor nth.

b.

I’m adding another abstraction – the predicate structure-is-weight?.

(defun structure-is-weight? (structure)
  (atom structure))

(defun weight-of-branch (branch)
  (let ((struct (branch-structure branch)))
    (if (structure-is-weight? struct)
      struct
      (weight-of-mobile struct))))

(defun weight-of-mobile (mobile)
  (+  (weight-of-branch (left-branch mobile))
      (weight-of-branch (right-branch mobile))))

Note how weight-of-mobile and weight-of-branch are defined. These are mutually recursive functions. Defining them this way makes for a very natural code, IMO.

c.

I’m going to use the same technique (mutually recursive functions) to figure out whether a given mobile is balanced:

(defun torque-of-branch (branch)
  (*  (branch-len branch)
      (weight-of-branch branch)))

(defun branch-balanced? (branch)
  "A branch is balanced either when it has a structure
  that's a simple weight, or when the structure is 
  a balanced mobile" 
  (let ((struct (branch-structure branch)))
    (or
      (structure-is-weight? struct)
      (mobile-balanced? struct))))

(defun mobile-balanced? (mobile)
  (let ((lb (left-branch mobile))
        (rb (right-branch mobile)))
    (and
      (= 
        (torque-of-branch lb)
        (torque-of-branch rb))
      (branch-balanced? lb)
      (branch-balanced? rb))))

Note the documentation string added to branch-balanced?. It is a standartized feature of Common Lisp, allowing me to find out what a function does from the REPL:

[5]> (documentation  'branch-balanced? 'function)
"A branch is balanced either when it has a structure
  that's a simple weight, or when the structure is
  a balanced mobile" 
[6]>

d.

I only need to change the accessors left-branch, right-branch, branch-len, branch-structure and the predicate structure-is-weight?. The rest of the code builds on top of the abstraction created by these functions and doesn’t need to be changed.

Exercise 2.30

(defun square-tree-direct (tree)
  (cond ((null tree) nil)
        ((not (consp tree)) (square tree))
        (t
          (cons (square-tree-direct (car tree))
                (square-tree-direct (cdr tree))))))

(defun square-tree-map (tree)
  (mapcar 
    (lambda (subtree)
      (if (consp subtree)
        (square-tree-map subtree)
        (square subtree)))
    tree))

Exercise 2.31

(defun tree-map (func tree)
  (mapcar 
    (lambda (subtree)
      (if (consp subtree)
        (tree-map func subtree)
        (funcall func subtree)))
    tree))

Exercise 2.32

In set theory, the set of all subsets of some set S is called the powerset of S. So I’m naming the function accordingly. This is a piece of completely futile mathematical pedantry :-)

(defun powerset (s)
  (if (null s)
    (list nil)
    (let ((rest (powerset (cdr s))))
      (append 
        rest 
        (mapcar (lambda (r) 
                  (cons (car s) r)) 
                rest)))))

To understand why this works, let’s do the most natural thing and “walk over” the code mentally. So, the powerset of S is nil if S itself is nil. This is trivial. But what happens when S is a list ?

Well, it is somewhat similar to change counting from the previous sections. We “pick up” the first element of the list and concatenate two sets:

1. The powerset of all the elements without the first. 2. The powerset of all the elements without the first, with the first element prepended to each subset.

Consider the set (1). Its powerset is (nil (1)). Now, consider the larger set (2 1). According to the procedure above, its powerset is the concatenation of:

1. The powerset of all the elements without 2, that is (1). Which we already saw is (nil (1)). 2. The element 2 prepended to each subset of the powerset of (1), that is, 2 prepended to nil and 2 prepended to (1), in total: ((2) (2 1))

Concatenating the above two gives: (nil (1) (2) (1 2)), which is indeed the powerset of (2 1). Using the same reasoning, we can compute the powerset of (3 2 1). I hope it is clear by now why the procedure is correct.





	

1 The convention of CL is to use the ending p for predicates, although it’s not too consistent – consider the predicates null and atom, for example. Personally I prefer Scheme’s idiom of ending a predicate with a question sign.