《Inferring Decision Trees Using the Minimum Description Length Principle*》
Information And Computation 80, 227-248(1989)
My difficulty is: how to get 18.170 bits when computing Coding Decision Tree Costs?
here are some relevant part in this article.
---------------the decision tree is:--------------
3rd page of article( page 229 on the top-right corner)
-----------------------------formula for computing cost-------------
10th page of article ( page 236 on the top-left corner)
The total cost for this procedure is thus:
L
(
n
,
k
,
b
)
=
l
o
g
2
(
b
+
1
)
+
l
o
g
2
[
C
n
k
]
L(n,k,b)=log_{2}(b+1)+log_{2}[C_n^{k}]
L(n,k,b)=log2(b+1)+log2[Cnk]
-----------------------the sequence relevant to the above decision tree-------------------------------
13th page of article ( page 239 on the top-right corner)
All above are trying to encode the whole decision tree,
and then send the decision tree from sender to receiver.
My difficulty is: how to get 18.170 bits mentioned above?
My understanding is:
Outlook: 2 bits
Humidity lg(3)bits
Windy lg(2)bit (not mentioned in article ,I just guess)
the whole sequence in this paper is :
1 Outlook 1 Humidity 0 N 0 P 0 P 1 Windy 0 N 0 P
then,
there are 8digits above, the following I guess may be wrong:
3 decicion nodes cost:3 bits
5 leaves cost: 5 bits
2+lg(3)+log(2)+3+5+?=18.17
?=5.585,
but how to get 5.585?
Although the following picture(encoding exception) has the number 5.585,
I guess it cannot be used as the explanation of above 5.585.
Could you tell me how to get “18.170 bits” mentioned in this article ?
Thanks very much!