Power-Programmierung/CD1.mdf/j/os2/what_is.j

Text File | 1992-03-15 | 14KB | 424 lines

               What is J?

            An introduction

                Part 1

                 by

            Leroy J. Dickey

            October, 1991


J is a high powered general purpose programming language.  This dialect
of APL uses the ASCII character set, has boxed arrays, complex numbers,
and phrasal forms.  Any time one wants to do calculations with more
than one number at a time, or more than one name in a list, a
collective (an array) is the right structure to use, and J is designed
to make it easy to do collective calculations.

People who write programs for themselves and who want their answers
quickly will be happy with the swift and concise way programs can be
written in J.

This article consists of several examples that illustrate some of the
power of the language J.  In each of the examples presented, output
from an interactive J session has been captured, and few lines of
explanation have been added.  Lines that are indented with four spaces
are those that were typed in to the J interpreter, and the lines that
are indented with only one space are the responses from J.  Other text
lines are comments that have been added later.  The topics that have
been chosen for inclusion do not come close to telling everything about
J, but some of them represent subjects that at one time or another the
author wished that he had known and understood.

Because my professional interests are mathematical, my examples are
primarily mathematical in nature.  But J is a general purpose computing
language, and is well suited to a broad spectrum of programming needs.

In this discussion about J, the words noun and verb are used to stand
for data and function, respectively.  Thus 'abc' is a noun,  1 2 3 4
is a noun, and + is a verb.  One may assign data to a name, as in the
statement    alpha5=.'abcde', and the name, (such as "alpha5"), used to
refer the data, is said to be a pronoun.  Similarly, one may associate
a name with a verb, and such a name is said to be a proverb,
(pronounced "pro-verb", not "prah-verb").

Thus, in the assignment
    plus =. "+"
the name "plus" is a proverb used to refer to the verb "+".  That is,
"plus" stands for "+".  There are other language elements called
conjunctions, which join two verbs or proverbs, and there are elements
called adverbs, which modify the meaning one verb or proverb.


Example 1:  Functional notation

This example shows how calculations are done with a list of numbers and
how the notation is functional.  That is, each verb (function) acts on
the data to its right.   It is easy to create a collective.  Here, for
instance, are 9 numbers:

    i. 9
 0 1 2 3 4 5 6 7 8

    a =. i. 9
    a
 0 1 2 3 4 5 6 7 8

The factorials of these numbers are:
    ! a
 1 1 2 6 24 120 720 5040 40320

The reciprocals of the above:
    % ! a
 1 1 0.5 0.166667 0.0416667 0.00833333 0.00138889 0.000198413 2.48016e_5

And the sum of reciprocals of the factorials of a is:
    +/ % ! a
 2.71828


Those who know some mathematics may recognize an approximation to the
number ``e'', the base for the natural logarithms.  Of course J has
other ways to calculate this number; the point here is the use of the
natural left to right reading of the meaning of the sequence symbols
"+/%!a" as "the sum of the reciprocals of the factorials of a".

Those who have done some programming in almost any other language, will
see that the expression "+/%!i.9" is a remarkably short one to produce
an approximation to ``e''.  Moreover, it is possible to pronounce this
program in English in a way that precisely conveys its meaning, and
those who know the meaning of the words in the sentence will be able to
understand it.  And anybody who knows the mathematics and who knows J
will recognize that the program does exactly what the words say and
that it will produce an approximation to ``e''.

The author finds this expression in J much easier to think about and
to understand than a corresponding program in a ``traditional'' language
intended to compute the same number.  Here, for example, is what one
might write in Fortran, to accomplish the same result:


    REAL SUM, FACTRL
    SUM = 1.0
    FACTRL = 1.0
    DO 10 N = 1, 8
        FACTRL = FACTRL * N
        SUM = SUM + 1.0 / FACTRL
     10 CONTINUE
    PRINT, SUM
    STOP
    END


Compare this Fortran program with the J program that uses only a few
key strokes: +/ % ! i. 9 .  Not only is the J program shorter, but it
is easier to understand, even for the person who has just learned the
meaning of the symbols.  Some have estimated that for typical
applications, the ratio of the number of lines of Fortran or C code to
the number of lines of J code is about 20 to 1.



Example 2:

In this example, two verbs (functions) are defined.  As before, lines
indented with four spaces were given as input during a J session, and
the ones immediately after, with one leading space, were produced by
the J interpreter.  Other lines are comments, and were added later.


    Sum =. +/
    Average =. Sum % #
    c =. 1 2 3 4
    Average c
 2.5

The first two lines create proverbs, the third creates a pro-noun,
and the fourth invokes the proverb (function) Average with the
pronoun (data) c.  The meaning of   Average c  is this:

        (+/ % #) 1 2 3 4

and this may be thought of as:  find Sum c, (add up the entries
in c), find # c (the count of c), and then find the quotient of those
two results.

Example 3:  Continued fractions

In this example, the verb ``pr'' (plus reciprocal), is considered.

    pr =. +%

The next two input lines show a use of this hook.  From them you
can see that the meaning of "a pr b" is "a plus the reciprocal of b":

    5 pr 4
 5.25

    4 pr 5
 4.2

The function "pr" is defined above as  +% , and this diagram
for    4 +% 5   may help you to understand its meaning.

                         +
                        /  \
                       4    %
                            |
                            5

Because of the shape of the diagram, the combination of two verbs
in this way is called a hook.  To continue with our example,

    1 pr 1 pr 1
 1.5

The above input line may be read as ``1 plus the reciprocal
of (1 plus the reciprocal of 1)''.  And here below is an analogous
line with four ones:

    1 pr 1 pr 1 pr 1
 1.66667

    1 pr (1 pr (1 pr 1))
 1.66667

    pr / 1 1 1 1
 1.66667

    pr / 4 # 1
 1.66667

The above input line has exactly the same meaning as the three input
lines that come before it.  Notice that (of course) the value
of the result is the same.  The "/" is an adverb called "over"
and the result of ``pr /'' is a new verb, whose meaning is
derived from the verb ``pr''.

This expression could be written another way, using a traditional
mathematical notation the way that mathematicians have been using
to write continued fractions for years:

        1
   1 + ------------------------
        1
    1 + ------------------
            1
        1 + ------------
            1

This particular continued fraction is a famous one because it is
known to converge to the golden ratio.  That means that if you
just keep on going, with more and more ones, the value gets closer and
closer to the golden ratio.  One can see the values of the continued
fraction expansion "converging" by using "\", the scan adverb.  It
creates a sequence of initial sequences.  Here is a nice, compact
expression that shows the values for the first 15 partial continued
fractions.

    pr /\ 15$ 1
 1 2 1.5 1.66667 1.6 1.625 1.61538 1.61905 1.61765 1.61818 1.61798
      1.61806 1.61803 1.61804 1.61803

In this, you can see that the numbers seem to be getting closer and
closer to some limit, probably between 1.61803 and 1.61804.  If you
concentrate only on the odd numbered terms, (the first, the third, the
fifth, and so on), you will see a sequence of numbers that is
monotonically increasing.  On the other hand, the even numbered terms,
( 2, 1.66667, 1.625, and so on), form a sequence of numbers that is
monotonicly decreasing.  Of course these observations do not constitute
a proof of convergence, but they might convince you that a proof can
be found, and in the light of these observations, the proof probably
follows along these lines.

Can you think of a simpler way to represent and evaluate continued
fractions?  I can not!

The golden ratio has been known for thousands of years, and is the
number, bigger than one, such that itself minus 1 is its own reciprocal.
The golden ratio is usually denoted by the lower case greek letter tau.
It is said that pictures whose sides are in the ratio 1 to tau are
the most pleasing to the eye.

Example 4.  Using rank operator and fork.

In this example, a second phrasal form, the fork, is discussed.
First some data are built.

    a =. 100 200
    b =. 10 20 30
    c =. 1 2 3 4

This is the beginning of the data construction.  We use the verb
Sum, a modification of the verb plus, and it is defined as before:
    Sum =. +/
But in this example, the usage is different. To get an idea about
the dyadic action of the verb Sum, notice what Sum does with
just b and c:

    b Sum c
 11 12 13 14
 21 22 23 24
 31 32 33 34

Once you have seen this example, you are ready to see and understand
the next step.  Here the collective ``data'' is built, using a, b,
and c.

    data =. a Sum b Sum c
    data
 111 112 113 114
 121 122 123 124
 131 132 133 134

 211 212 213 214
 221 222 223 224
 231 232 233 234

This rank three array may be thought of as having has two planes,
each having three rows and four columns.

    $ data
 2 3 4

Now examine what "Sum" does with the data:

    Sum data
 322 324 326 328
 342 344 346 348
 362 364 366 368

Can you see what is happening?  It is adding up corresponding
elements in the two planes.  Twelve different sums are performed,
and the result is an array that is 3 by 4.  The action happens
over the first dimension of the array.

But we might wish to add up numbers in rows.  We can do it this way:

    Sum "1 data
 450 490 530
 850 890 930

And to add up numbers in each column:

    Sum "2 data
 363 366 369 372
 663 666 669 672

The expressions "1 and "2 are read as 'rank 1' and 'rank 2'.  The rank
1 objects of 'data' are the rows, the rank 2 objects are the planes,
and the rank 3 object is the whole object itself.  So, when one asks
for  Sum "2 data ,  one asks for the action to happen over the first
dimension of the rank two arrays.

Now, recall the definition of Average.

    Average =. Sum % #

Apply this proverb "Average" to the pronoun "data", to get:

    Average data
 161 162 163 164
 171 172 173 174
 181 182 183 184

The result is the average of corresponding numbers in the two planes.
That is, the average is calculated over the objects along the first
dimension of data.  If we would like to know the average of the
numbers in the rows, we type:

    Average "1 data
 112.5 122.5 132.5
 212.5 222.5 232.5

and finally, averaging the columns of the two planes:

    Average "2 data
 121 122 123 124
 221 222 223 224


Again, compare the sizes of these results with the sizes of
the sizes of the results above where we were asking simply about
sums.

    Sum "1 data
 450 490 530
 850 890 930

    Sum "2 data
 363 366 369 372
 663 666 669 672

What is exciting about this example is not only how easily the nouns
and verbs were built, but how every verb acts in a uniform and
predictable way on the data.  Rank one action is the same kind of
action whether one is summing, or averaging, or whatever.  It has been
said that the concept of rank is one of the top ten ideas to come along
in computing in the last decade.


Conclusion:

What has been written in this article only begins to scratch the
surface of the power of J, but it does show some of the flavor.  J is
powerful because one get results quickly.  It is a programming language
for people who need to write programs for themselves, but who don't
want to spend inordinately large amounts of time getting their
results.  If you have to write programs for others, this could still be
a language for you, if you get paid by the job, and not by time.


Availability:

J is available by "ftp" from watserv1.waterloo.edu in the directory
"languages/apl/j" and is also available from certain other servers.
At the Waterloo site, versions of J are available that run on several
kinds of computers.


Learning J:

Because J is new, there are not many materials about J yet.
The novice should get a copy of J and let J be the guide and
the interpreter (pun intended).  There is a tutorial that has
tons of material in a small space.  This tutorial comes with
the interpreter.  There is also a "status.doc" file that comes
with J.  This tell what features have been implemented and
which have not.


The author of J:

The inventor-developer of J is Kenneth E. Iverson, the same man who
invented APL, a real pioneer of computing.  Some years ago, he retired
from IBM and went to work for I.P.Sharp Associates, a Toronto company
that developed the world's first private packet switching network,
the finest APL available, and was at one time the worlds largest
time sharing service, all APL.  The descendant of this company,
Reuter:file, still has offices in many of the world's major cities.
Now Iverson is retired (again) and lives in Toronto, where he spends
his time developing and promoting J.

The programmer for J is Roger Hui.  Roger hails from Edmonton, and has
lived in Toronto since he joined I.P.Sharp Associates in 1975.  The
source code for J, written in C, is worthy of some comment, because it
makes extensive use of macro expansion.  Its style, influenced by the
work of Arthur Whitney of Morgan Stanley in New York, reflects Roger's
strong background in APL.  Today, Roger spends much of his time
developing J and does occasional contract work in APL.


Copyright (c) 1991 by Leroy J. Dickey
Permission is granted to use this for
non-profit and educational purposes.
It may be given away, but it may not be sold.


Revised 1991-12-04