~1
                          LOGIC PROGRAMMING LESSON 5


This lesson is about errors, numbers, and comparisons.

~2
Please start by switching your logic programming tutor into Prolog mode,
and then loading facts from the knowledge base in "mars". The "load"
command was introduced in Supplement 1, and is like "restore" in that it
expects to be given the name of a file. However, it only loads facts
into the database, and so you can't "show" them, though you can ask
questions about them. Files that you load are also inaccessible to
"restore" (and vice versa): if you try to restore mars, the Tutor will
tell you that it can't find the file.

So, give the two commands:
      prolog.
      load mars.

~3
You now have a number of facts defining the predicate "costs", which
gives the prices of sweets. They all follow the pattern

      costs( mars, 15 ).

where the first argument is the name of a sweet, and the second is a
NUMBER representing its price in p.

So far, we have used ATOMS and VARIABLES in facts. We haven't said much
about what you can do with numbers, though they did crop up in various
exercises.

Please note that these numbers should not include units: Prolog will not
accept facts like

      costs( mars, 15 pence ).

So, for the moment, you must just remember (or make a note) of what the
numbers in each fact refer to: pence, dollars, yen, volts, shocks on the
Richter scale, population density in heads per hectare...

~4
Now try asking Prolog questions which will answer the enquiries below.

1) What is the price of a Nutty? (I've used the atom "nutty" for this
    sweet).

2) What are the sweets that cost 15p?

3) Now, by asking ONE question with an "and" (i.e. a comma) in it, find
   those things which cost the same as a Marathon (I've used the name
   "marathon" for that sweet).

~5
By now you should be able to do the first two with ease. To do the
third, translate this into Prolog:

    Find a P and S such that a Marathon costs P,
    and there exists a sweet S such that S costs P.

~6
Here are the answers. Of course, you can use whatever names you want for
the variables, as long as they start with a capital letter.

1) costs(nutty,What)?

2) costs(What,15)?

3) costs(marathon,P), costs(S,P)?

The third question can be understood as "Find all S costing P, where a
Marathon costs P". This way of using the variable P is a frequent trick
of logic programmers: one part of the question gives the variable a
value, and a later part of the question finds facts which mention that
value.

~7
A note on idioms: Each programming language has its own idioms:
frequently occurring combinations of commands which experienced
programmers recognise without detailed analysis. Basic programmers, for
example, recognise the commands
      FOR i = 1 TO 10 : LET a(i) = 0 : NEXT i
as an idiom meaning
      Set each element of array a to zero

The Marathon example demonstrates a Prolog idiom worth recognising:
schematically, the question
      Thing has property P and Object has property P ?
finds Objects with the same property P as Thing.

Thus,
      costs(marathon,P), costs(S,P)?
finds sweets with the same cost as a Marathon has.

(POLITE NOTICE: Of course, some statements look like common idioms but
aren't - care is essential).

~8
I shall now introduce comparisons. In programming, it's useful to be
able to compare numbers. You can get a long way without them, but they
do come in useful eventually.

Prolog provides several predicates for such comparisons. Two of them are
> and <, meaning greater-than and less-than. Comparisons occurred in the
skiing knowledge base, when you were using the Tutor in Logic mode, but
I didn't say anything in particular about them.

To see how they work in Prolog, try a few questions like
      >( 1, 2)?
      <(45,74)?

~9
That you got sensible answers to these shows that you can use > and <
just like any other predicate. However, it's inconvenient for us to have
to put brackets around the arguments to these symbols, because we're so
used to seeing expressions like 1>2 and 45<74 in maths. Prolog therefore
allows us to write these comparisons more naturally as

      1 > 2?
      45 < 74?

Try this. Note that it is grammatically correct even when you are as now
in Prolog mode and not Logic. We say that the comparison predicates are
OPERATORS - this is Prolog jargon for predicates that can be written
between their arguments, a la maths.

~10
You can use these predicates in questions. Can you formulate one
question that ask which sweets cost more than 15p?

~11
Some people find this tricky, and are tempted to ask something like
      What > 15?
which is incorrect. This temptation is probably because English allows
"What costs more than 15p?". However, in Prolog, you can't abbreviate in
this way, and you must painstakingly build up the question as a sequence
of elementary questions separated by comma.

If you have no idea how to attack this, try translating the question
below into Prolog:

      Find an X such that X costs P and P > 15.

~12
You should have asked

      costs( X, P ), P > 15?

though perhaps with different variable names. If you're not convinced,
ask the question above, and then use the "analyse" command as
demonstrated in Lesson 4. This will give you a translation back into
English.

~13
You also have some facts defining the "likes" predicate, where the first
argument is someone's name, and the second names a sweet. Try
translating

      Who likes sweets costing more than 15p?

into one Prolog question.

~14
If you're having trouble, try translating this form:

      Find an S such that W likes S and S costs P and P > 15?

~15
You should have got

      likes(W,S), costs(S,P), P > 15?

(but perhaps with different variable names).

~16
Now try

      What sweets cost more than 13p and less than 20p?

~17
If you're having trouble, think of it as

      Find an S such that S costs P and P > 13 and P < 20?

~18
You should have got

      costs(S,P), P > 13, P < 20?

As before, if you can't see why, try translating back into English,
either on your own, or with the help of analyse.

Next, try

      Who likes sweets that cost more than a Mars Bar?

~19
Think of it as

      Find an X such that
          X likes S and
          S costs P and
          mars costs M and
          P > M.

~20
The answer should have been something like

      likes(X,S), costs(S,P), costs(mars,M), P > M?

~21
Finally, try

      Who likes sweets that cost more than the one that Adam likes?

~22
Hint: think of it as

      Find an X such that
          X likes XS and
          XS costs XP and
          adam likes AS and
          AS costs AP and
          AP < XP?

~23
Now for a change of viewpoint.

Some people always ask how Prolog knows that 2 > 1 and that 32 > 2. Does
it have a gigantic set of facts somewhere which say things like

      2>1.  3>1.  4>1.  ...
      3>2.  4>2.  5>2.  ...
      4>3.  5>3.  6>3.  ...

and so on for all the numbers which you could ever ask about? The answer
is no. All computers have special instructions for comparing numbers.
When answering most predicates, Prolog looks up the answers in the
database. However, when it encounters > and <, it treats them as
special, and runs the number-comparison instructions to get an answer.

For this reason, we say that > and < are BUILT-IN. You don't have to
define them: the implementor has done so already.

~24
You can confirm this by trying to ask a question like

      mars > twix ?

then Prolog will give you an ERROR-MESSAGE , saying that it can't answer
the question (you may have seen all too many error messages already, if
you have mis-typed things). The message will say that > needs numbers.
This is entirely fair. Mars Bars and Twixes are not numbers, so you
can't compare them (to be exact, you can't compare their names).

~25
Perhaps we could fool Prolog into believing that mars > twix, by
asserting the fact

      mars > twix.

Go on, try it. Prolog will give another message, saying that you are not
allowed to update built-in facts. The reason for this is efficiency. It
is much faster for a computer to use its number-comparing instructions
than for it to search the knowledge base for answers. Most programmers
want numerical operations like > to be as fast as possible, and very few
want to write facts like "mars > twix".

The optimum engineering solution is to make Prolog prohibit such extra
facts, so that it can always use the number-comparing instructions, and
run comparisons as fast as possible.

~26
Still concerning error messages and speed, try asking

      X > 1 ?

Logically, this means 'which X are greater than 1?' or 'find all numbers
bigger than 1'. Prolog will not give you the answers; instead, you will
get another error. This is because there is an infinity of numbers
greater than 1, and to generate them all would take longer than I have
for the course, especially with the VAX as slow as it is today. Instead,
Prolog rejects the query, and insists that when you use > , each side
has a (numeric) value.

~27
You have now seen two kinds of error.

The first is where Prolog rejects input which ought to be reasonable, in
order to run more quickly (in "X > 1?"). This is one of many instances
where Prolog does not do everything that logic does. The cave-world
exercise in Supplement 2 was another instance of this, where you may
have written predicates that were logically correct, but that got Prolog
into an infinite loop. Yet another instance is that Prolog won't let you
assert two facts in one go by using "and" in the head, as in
      loves(john,mary), likes(mary,wine).

The second kind of error is where Prolog rejects rubbish ("mars>twix").

Much of the business of programming is learning the details of a
language (to avoid the first kind of error), and learning how to make
programs do only sensible things (avoiding the second kind).

~28
Now let's look at the order in which Prolog evaluates things. Consider
the question
      costs(S,P), P > 15?

Now, try re-ordering it as
      P > 15, costs( S, P )?

You will get yet another error-message: once again, > needs numbers.
Why? Surely the erroneous question is logically equivalent to one which
works? And surely Prolog will accept anything which is logically valid?

The reason for the error is that Prolog answers questions from left to
right. It can't find a value for P until answering the second part of
the question - "S costs P"?. By then, it has already tried to compare P
with 15, despite P not yet having a value.

This is another way in which Prolog will reject some logically valid
commands. When using > , remember that questions are answered from left
to right, and that variables being compared must have received their
values from earlier parts of the question.

~29
Now for some more comparison predicates.

As well as the built-in predicates "<" and ">" , Prolog recognises
several others. The complete set is:

      >     <   =<  >=  \=  =

The final four symbols mean 'less than or equal to', 'greater than or
equal to', 'not the same as', and 'equal'. A handy mnemonic for which
way round to write =< and >= is that you place the equals in such a
position that the signs don't look like arrows. So <= and => are WRONG.

These combinations of symbols are the nearest that your keyboard can
come to the mathematical equivalents - even modern computer keyboards do
not contain more than about 100 printable symbols.

~30
Now that you have the \= operator, can you ask a question to find all
sweets that don't cost 15p?

~31
Think of it as

      Find an X such that
          X costs P and
          P is not equal to 15?

~32
The = and \= operators can be used to compare atoms as well as numbers.
So whereas
      mars > twix?
will give an error,
      mars = twix?
and
      mars \= twix?
will not.

Can you now ask one question that finds the price of all sweets that are
neither Yorkies nor Mars bars? Use atoms "yorkie" and "mars".

~33
Think of it as

      Find an X such that
          X costs P and
          X isn't a yorkie and
          X isn't a mars bar.

~34
You should have typed something like

      costs( X, P ), X \= yorkie, X \= mars?

~35
Now, can you find all items which are NOT Marathons? Not their prices,
but just their names.

~36
That was a trick question.

Some people will have tried the answer

      costs( X, P ), X \= marathon?

on the assumption that the "costs" predicate mentions every sweet worth
talking about.

~37
Others will have tried this:

      X \= marathon ?

If you did that, Prolog would have replied "no", and not set X to
anything. Why?

Well, the nearest Prolog could get to an answer would be to generate an
infinite number of names, naming non-Marathons (perhaps: mars, galaxy,
nebula, black hole, whitehall, front hall...). This is both impossible
and not useful, rather like wanting a solution to X > 1.

So the implementors of Prolog have made things easy for themselves by
saying: we shall only make \= work sensibly if, when it's comparing
variables, both those variables have already been set. Otherwise, we'll
just make it give the (misleading) answer "No".

~38
In practice, when asking such a question, you will have some
classification in mind - in this case, into kinds of food. Perhaps
something like

      type_of(mars,sweet).
      type_of(tango,drink).
      type_of(peperami,savoury).

and so on. You can then ask questions like

      type_of(X,food), X \= marathon ?

That's the best way to deal with situations where you may need a list of
all the individuals that aren't something else.

~39
To summarise what has gone so far:

You can use NUMBERS as arguments to predicates.

Prolog provides BUILT-IN predicates for comparing numbers. These are
      >   >=   =<   <   \=  =

You can write these BETWEEN their arguments, without brackets. That
resembles maths notation: for this reason, they're called OPERATORS.

You will get an ERROR-MESSAGE if you try to compare atoms (i.e. names)
with these predicates.

You will also get one if you try to re-define them by adding facts like
      1 > 2.
      mars > twix.

~40
Although a question like
      X > 1?
makes logical sense, Prolog won't answer it. Instead, it will give you
an error. That's because the question has an infinite number of answers,
and the computer is only finite.

The questions
      costs(S,P), P > 15?
and
      P > 15, costs( S, P )?
are logically equivalent. However, Prolog can't answer the second one.
That's because it works from left to right inside questions, and the
variable P must have gained a value in order for the comparison to work.

~41
There are two operators for comparing non-numeric (as well as numeric)
items. These are
      \=  =

Like the numeric comparisons, \= can't be used to compare variables that
have no value. A question like
      X \= marathon
can't be answered, because it would have to generate an infinite list of
individuals, most of which would have nothing at all to do with your
program (although, for this lesson, you may care that a Mars isn't a
Marathon, you don't care that elephants, abstract justice, and
pandiculation aren't one either).

But in this case, Prolog doesn't give an error message. Instead, it
(misleadingly) says "No".

~42
To make all the comparison operators safe, so that they behave as
logic dictates they should, you should simply remember that Prolog
answers questions from left to right. Always put the comparison
operators in such a position that any variables they compare will
already have gained values.

~43
Finally, a word on Prolog execution in general. We have seen that Prolog
answers questions from left to right. Another aspect of execution is in
which order it visits facts. Given the facts
      loves(john,mary).
      loves(hendrik,elaine).
which order do you get the solutions in when you ask
      loves(X,Y)?

The answer is: from top to bottom. When trying to answer a simple goal,
Prolog starts at the top of the database and works downwards.

~44
There is more to Prolog execution than this. Suppose you ask
      loves(X,Y), Y \= mary?
with the above facts stored. What Prolog does is to start at the left of
the question with the first goal, for loves.

It searches down the database until it can get an answer. This gives it
X as john and Y as mary.

Having got an answer for the first part of the question, it now resumes
its left-to-right progress, and tries to answer the comparison. Here, it
finds that Y\=mary is false. Now, some systems might give up at that
point. But Prolog has been designed to try all possible combinations.

It returns, or BACKTRACKS, to the loves(X,Y) and tries RE-DOING it,
looking for the next fact down. This gives it another answer, X as
hendrik and Y as elaine.

With this new answer, it resumes its rightward movement, and again tries
Y\=mary. This time, it is true, and therefore the whole question
succeeds.

~45
I shall not say more about the order of execution here: it's best
explained individually, and with plenty of diagrams, by your teacher.
However, if you want to watch Prolog at work, you can use the "trace"
command. This is a bit like "why". It takes the previous question you
typed, and re-runs it, writing out details about everything Prolog does
on the way.

So if you have just asked
      loves(X,Y), Y \= mary?
and you then say
      trace.
you will be able to observe Prolog visiting each fact and each part of
the question. Remember that we also call facts, CLAUSES: this is also
how trace refers to them.

The output from trace can be quite voluminous. As with "why", you can
send it to a printer, so
      trace ctc.
will send the trace to the CTC printer.

~46
You might now try tracing some questions. Start with really simple ones,
otherwise you will drown in the output. For instance, you might delete
all facts, enter ONE fact about loves, and ask a question about it.
Choose one question that will succeed, and another that will fail, and
see what trace shows in each case.

Then work up to two loves facts, like those I showed above, and ask a
question which depends on the bottom one, such as
      loves(hendrik,X)?
      loves(X,Y)?

Then try a compound question such as
      loves(X,Y), X \= mary?

You can try the same kind of thing with the "mars" knowledge base, using
the comparison operators. Doing this should give you a feel for the way
that Prolog orders its inferences, and how this way of doing things
enables it to give logically correct answers.

~47
You have now reached the end of Lesson 5. Before going on to Lesson 6,
please do any exercises your teacher has set for this lesson.
