hypothesis function space in decision tree - machine-learning

Question

I am reading the book "Artificial Intelligence" by Stuart Russell and Peter Norvig (Chapter 18). The following paragraph is from the decision trees context.
For a wide variety of problems, the decision tree format yields a
nice, concise result. But some functions cannot be represented
concisely. For example, the majority function, which returns true if
and only if more than half of the inputs are true, requires an
exponentially large decision tree.
In other words, decision trees are good for some kinds of functions
and bad for others. Is there any kind of representation that is
efficient for all kinds of functions? Unfortunately, the answer is no.
We can show this in a general way. Consider the set of all Boolean
functions on "n" attributes. How many different functions are in this
set? This is just the number of different truth tables that we can
write down, because the function is defined by its truth table.
A truth table over "n" attributes has 2^n rows, one for each
combination of values of the attributes.
We can consider the “answer” column of the table as a 2^n-bit number
that defines the function. That means there are (2^(2^n)) different
functions (and there will be more than that number of trees, since
more than one tree can compute the same function). This is a scary
number. For example, with just the ten Boolean attributes of our
restaurant problem there are 2^1024 or about 10^308 different
functions to choose from.
What does author mean by "answer" column of the table as a 2^n-bit number that defines the function?
How did author derive (2^(2^n)) different functions?
Please elaborate on above question, preferably with simple example, such as n = 3.

Answer 1

Consider a general truth table for a 3-input function, where the result for each triple is also a Boolean (1 or 0), represented by variables i through 'p':
A B C f(a,b,c)
0 0 0 i
0 0 1 j
0 1 0 k
0 1 1 l
1 0 0 m
1 0 1 n
1 1 0 o
1 1 1 p
We can now represent any function on three variables as an 8-bit number, ijklmnop. For instance, and is 00000001; or is 01111111; one_hot (exactly one input True) is 01101000.
For 3 variables, you have 2^3 bits in the "answer", the complete function definition. Since there are 8 bits in the "answer", there are 2^8 possible functions we can define.
Does that outline the field of comprehension for you?
More detail on an example function
You simply (once you see the pattern) make the eight bits correspond to the entires in the table. For instance, the table for one-hot looks like this:
A B C f(a,b,c)
0 0 0 0
0 0 1 1
0 1 0 1
0 1 1 0
1 0 0 1
1 0 1 0
1 1 0 0
1 1 1 0
Reading down the "answer" column, labeled f(a,b,c), you get the 8-bit sequence 01101000. That 8-bit number is sufficient to completely define the function: the rows listing all the combinations of a, b, c are in a fixed (numerical) sequence.
You can write any such function in a template format:
def and(a, b, c):
and_def = '00000001'
index = 4*a + 2*b + 1*c
return and_def[index]
Now, if we generalize this to any 3-input binary function:
def_bin_func(a, b, c, func_def)
return func_def[4*a + 2*b + 1*c]
If you wish, you can further generalize the template for a list of inputs: concatenate the bits and use that integer as the index into the func_def string.
Does that clear it up?