Lecture 59 - Coverage (Decision Tree Analysis Contd...)

IIT Kanpur July 201
20 followers
Jun 12, 2024
[Music]
uh welcome back my dear friends a very
good morning good afternoon and good
evening to all of you and as you know
this is the NP Muk uh lecture series
which is spread over 12 week 60 lectures
and the topic is multicriteria decision
making and we in the last set of
lectures and we are still two left to
crap up this course and as you know in
this course uh each week we had five
lectures each lecture being for half an
hour and after each five lecture we had
assignments of 10 questions each and
there would be a final examination
covering the whole set of topics for
multier decision making
uh my good name is ragunan s Gupta from
the IM department at atican food and if
you remember I'll just highlight the
important points we have covered in
depth multiobjective methodologies
different methods multi uh then
non-parametric methods different methods
univariate utility Theory multivariate
utility Theory but just very
fleetingly and under nonp us Electra EPS
Electra topsis Vore HP
da and in the last lecture we started
the decision tree now for the last
lecture and this lecture as the diagrams
cannot be made uh in Excel sheet hence
I'm using the word document so the
general caption for all the lectures
which you used to see uh that uh we are
in this lecture covering this like this
set of points that coverage would
definitely be there when the slides are
shared with you but for here to maintain
continuity and without erasing whatever
small calculation I'm doing please
excuse me I'm continuing in the same
format now if you remember in the
problem for the decision
tree we were
at the point D2 D2 the value was 194 I'm
not reading the decimal 194 minus 150
which is um 44 to be compared with zero
so 44 was positive we'll take that when
we want to find out the value of
C1 now for C1 remember this now the
logic would have become clear to you the
value for high
demand probability was 7 the value was
44.2 as as I said I won't be reading the
decimal but it's 2 so 7 into
44.2 and for the low demand where you
will stop the probability is3 so 3
multiplied by 0 so total value the
expected mean value at C1 comes out to
be 30.9 so it's still
positive so when I come to thir so which
is here which is C1 now there was an
investment of 20 million for the pilot
FR and test marketing so if I consider
that value 30.9 consider is 31 31 - 20
is 10.9 still positive hence the
decision Tre would be to set up the plan
and the path to be taken are which I
will highlight
here are as
follows which
means the company I'm putting a hash for
the other diagram to because as a
complicated diagram in the second
example it will be
colored so for
the Spectrum company it will build up
test
marketing then there is a high
demand obviously with
7 then it will build the actual plant at
150 and the expected value either is 30
million for one year for 20 years for a
high demand of 60% or low Demand on 40%
whichever it is the thing would be true
so it will take the path as
given so if you see we are solving from
the right but the overall diagram would
be given now the
problem
is and the question which I did pose to
you is that can we consider any other
set of criteria to evaluate yes we can
use the concept of variance and try to
basically find out whether the
variability can be utilized in order to
rank them or any other measures now
immediately when I say any other
measures even though not very evident
the idea should immediately ring some
bell in your mind in the DM method if
you remember what we did we were trying
to find out the
efficiency
and the word efficiency was the ratio of
the bundle of outputs divid by the
bundle of
inputs depending on which model you are
going to use if you remember we had the
decreasing return to scale increasing
return to scale and constant return to
scale and one another case we for the
efficiency we maximize the efficiency
and in the case when we want to take the
ratio of the input to the output we took
minimization of
that so whichever Direction you're
looking it's basically the efficiency we
are considering now obviously in the
concept of expected varant value and
variance no such term as efficiency can
be coined but we can also consider the
ratio of expected value to
variance and consider it as one of the
the
main way or criteria of trying to
basically rank the decisions or the
paths accordingly so that we'll take
maximum value of the expected value to
variance and rank them from the highest
to lowest or else we can take the ratio
of variance to expected value and rank
them from the lowest to the highest so
it's again I'm saying it's not
efficiency but it's the way of trying to
analyze the problem accordingly so the
problem which we did for expector Value
can also be looked from the point of
view of variance ratio of expected value
to variance ratio of I to expect a value
and all the ways can be done now comes
the second problem and here the second
problem is is a little bit more
intuitive much more
involved and here till now for the first
problem the probabilities are given very
straightforward no complication here the
pro probabilities will be given but with
the twist say that you will understand
that depending on one process how the
probability would be decided for the
next stage and an oil company while
evaluating the oil Basin is considering
three Alternatives which are drill
conduct systemic
test before at a cost of 20,000 and find
the nature of the underlying oil Basin
and finally do nothing so the company
can have three Alternatives of act
course of action which are it can drill
directly it can conduct sismic
test before which costs about
20,000 and another thing can be do
nothing now the question would
immediately be when it drills obviously
there's a cost so cannot this cost be
incorporated in both alternative one and
two yes it can be I have not considered
that in order to make things
simple if the company drills then is
likely to find the oil Basin of three
types it's dry wet or
soaking a dry wheel yields nothing
because totally dry which means that
probability or the chance or the
quantity of finding oil is almost
zero if it is wet it provides moderate
quantity of of oil and if it is soaking
it will generate substantial quantity of
oil if the oil company had conducted
cisic tests obviously syic test would
give you uh a lot of information about
the rock layers
how permeable they are how hard they are
what type of rocks they
are depending on that sedimentary
rocks uh perity is there whether they
have a long of History where oil can be
found out at greater depths so those
concepts of
tests results would be basically give a
lot of information for the
company so let me continue reading it if
the oil company sismic test then it can
learn about the underlying structure of
the oil Bine before deciding whether to
drill or not to drill the underlying oil
missing structure may be depending on
the test it can be found out there's no
structure there's an open structure and
there's closed structure if no structure
is found then the prospect of finding
oil is Bleak if an open structure is
discovered then the prospect of finding
oil is fair while finding finally if the
structure is closed then the process of
finding oil is bright
so we already know the structures are no
open closed and we already know the
wells are dry wet soaking so obviously
it me mean that this combinations of
doing some test and what is the actual
drill results would give you different
types of combination which is very
interesting to analyze the company also
knows the following and this is what the
information is given stage by stage
and please mark the values because when
we come to the solution all those values
will be important by the way when you
are um solving the problems trying to
read the slides would be all be all with
there with you so please look at the
slides accordingly the probabilities of
various oil bearing states are it is dry
probability is5 wet Point 2 soaking
point2 if you sum it up the total value
comes out to be one and as it should be
probabilities of various soil geological
states are as
follows no structure point 4 open
structure3 closed
structure3 again the probability comes
out to some comes to be one so these are
two different things probabilities of
various oil bearing States dry weight
soaking and what is the structure open
no structure open structure closed
structure so one is for the drilling
part one is for the test
part also we have the following joint
probability distribution based on this
information which would be interesting
to note also we have the following joint
prob distribution between underlying
geological structure and oil bearing
State now by the way if you remember I
said that all the values will be
important so please remember these
values again I'm repeating the p is
basically probability which you should
be aware by this time as we continue
doing the problem
so state of dry5 state of wet 025 state
of soaking 25 these are the
probabilities again the probabilities
for no structure point 4 open structure3
closed structure3 so note them down as
we continue discussing the joint
probability overall table and here is
the
table and this is interesting so if you
see I have made a different coloring
schemes the vertical ones which is
basically denotes the underlying
geological structure which are no
structure open structure closed
structure are blue in color the vertical
part and the oil bearing states which
are horizontal and they are dry weight
soaking are shown in pink in color now
what is interesting if you don't know if
you have noted down few minutes back I
said note down the values of
4.3.3
5.2.2 these are the margin values what
are the margin values if you see at the
right in the table the right column and
the bottom row they will give you the
marginal values at the margins so the
marginal probability of the states of
the oil bearing States if you see they
were given as .5.
25.25 and marginal probabilities for the
geological structure were given as
4.3.3 interestingly sum up of all these
three values.
52525 is 1 sum of
4.3.3 is 1 also interestingly the values
which are given here like say for
example 32 what it means it means the
oil Bing state is dry and there's no
structure and the probab is 32 if I
continue along the
horizontal row it's dry with open
structure 0.15 it is dry with closed
structure is 03 if you add up
32153 the total value comes out to be
0.5 as it is along the
margin similarly for the row wiet the
corresponding probabilities for no
structure open structure closed
structures are
041 0.11 and the total sum
is25 for the last row soaking oil
bearing State the corresponding
probabilities for no structure open
structure closed
structure the values are
0405
six the total value sum
is25 if I follow the same logic along
the vertical for no structure with oil
bearing State as dry weight soaking the
probabilities of 3204 04 the total sum
is4 similarly for open structure oil
bearing States being three as they are
written the probabilities are
0.15.10 05 sum is3 finally for close
structure the probabilities are
03111 16 the sum
is.3 so this would basically give you
all the information which is needed and
we will utilize that for the
problem finally the all oil company has
the following set of information apart
from the uh condition probability table
which is regarding the net present value
of the three states which is net present
value for the Dy state is given as minus
0 0.6 million which is
negative net present value of the weight
state is +8 million and for the net
present value for the soaking state is +
2.4 million because as you know for uh
dry State wet State and soaking State
the probabilities of finding oil was
least than medium and the
highest now let us go to the solution
and before I go to the solution even
though nothing is written here let us
see that diagram now this diagram I have
to scroll up and down so please bear
with me as I said the diagram which I
tried
drawing considering there are so many
states decision Forks chance Forks it
was difficult but I tried my level best
to give the idea again the the coloring
scheme Remains the Same blue means
positive input some value red means
something going out negative
again the same concept we start from the
left hand side go on to the right hand
side to draw the diagram whole set of
gamut of all the options the fork and
the chance and we calculate from the
right to the left and follow the part
which gives us the maximum value based
on the criteria which you are trying to
maximize or
minimize now diagram is like follows so
we start at D1 and then consider the
concept so d11 which is the arm which is
going up is basically you drill without
the test D12 you conduct the test and if
you remember there was a price of 20,000
which is given as red color
0.02 and the other AR which is basically
here shown here is D13 is that do
nothing so obviously that was also an
option now when you drill
it what are the so obviously you don't
have any idea about the structure you
immediately drill it and find out the
characteristics as dry wet soaking the
chance Fork is C11 as you are starting
from there for the first one which is
dry then C12 for wet and C13 for soaking
so the chance uh Fork which is at the
top where the the mouse is hovering is
given by the number
C1 and what are the Net Present Value if
you remember net present value for the
drive was
Min -.6 which is shown in red color +8
which is shown in blue color plus 2.4
which is shown in blue color what are
the probabilities dry has a probability
of of 1 by two which is 50% wet has a
probability of uh 1x4 which is 25% and
soaking has a probability 1x4 which is
25% so these values are the marginal
values now the question will come up why
did I take the marginal value because
here no uh test has been conducted so
these probabilities are unconditional
based on the drilling only so there's no
conditional fact added to this now if I
consider the again I'm saying the
diagram I'm I'm scrolling up and down so
when you see the video and also refer to
the diagram it will be much easier for
you when you conduct the test what three
things occur one is you find no
structure which I'm highlighting and
this is given by c21 why c21 because the
arm which you reach the chance con the
point which you reach after conduct test
is given by the point
C2 so c21 is basically no structure then
you have c22 which is given as open
structure which is given here I'm just
made it here so that there's no
confusion and the last point is
basically c23 is close structure now
very interestingly the probabilities are
given probabilities are again the
probabilities for the unconditional part
for conducting at the test because you
conduct the test and then only you drill
in this case so conducting the test
probabilities over 40% 30% 30% here they
are 4X 10 for no structure 3x 10 for
open structure again 3x10 for the close
structure which is probabilities of c21
4X 10 probability of c22 3x 10
probability of c23
3x10 now when when when the no structure
is found then you drill once you drill
obviously you will find out as dry wet
soaking similarly when you find out open
structure again you will drill you will
find at dry wet soaking and when finally
the third case of close structure again
you will do it as drill and then dry wet
soaking these have been
Incorporated where now if you see no
structure and drill
the decision fork or decision arm or
decision point is D2 and the
corresponding directions it can move is
basically moves to
C3 and the other arm goes to d22 that
means you do not drill so basically D21
is Drilling and the outcomes are dry
weight
soaking and the Net Present Value we
have kept it fixed whether you conduct
the test then drill or you only drill
The Net Present values are for dry
weight soaking are negative 0.6 positive
0.8 positive
2.4 and interestingly the corresponding
arms are mentioned as c31 C32 c33 which
is for dry weight soaking and the
probabilities are different now how they
are different if you
see these dry States have been
encountered after the drilling and when
was the drilling considered when there
was no structure that means if I go to
the no structure part no structure part
and if I consider the concept of
drilling so what it would be with no
structure if I consider this this
column the total probability is point4
for no structure dve being 32 so the
hence the conditional probability would
be32
by4 similarly when I go to no structure
wet the conditional probability will be
04 by4 and finally for no structure with
soaking oil varing
State the the case is 04 by4 so if I
consider the values it is 32
by4 04 by4 44 by4 and here the values
are 04
by4 comes out to 1 by 10 04 by4 for the
third which is soaking comes out to be
10 1 by 10 and for the case when need 32
by4 it comes out to be 4X
5 similarly there is a arm d22 which is
don't drill which you already mentioned
and following this logic if I go to open
structure which is
D3 and then you drill D31 other part is
d32 which you do not drill after you
drill again you count out the same thing
dry with soaking Net Present Value being
.6 postive point8 postive
2.4 and the corresponding probabilities
are now we go to the open
structure open structure here is the
second
column the corresponding probabilities
are are for drve weight soaking
condition probabilities are 0.15
by3 .1 0
by3 and the last one is 05
by3 so if I go to the open structure
drive at soaking values are
here
half 1/3 1
16 so if I
consider half is here
05.15 by3 is half .1 by3 is 1 by
3 so you see the value of 1 by 3
here and the last value 1X 6 so if I see
the value of 05 by3 is 1x 6 so obviously
for this case when there
is open structure d32 would be do not
drill obviously you have to have a do
not drill in order to basically find it
out and finally for the Clos
structure with D4 there's d42 do not
drill and with d41 where there's a drill
again you have dry weight soaking Net
Present Value again I'm I'm assuming
simplistically as minus 0.6 plus 0.8 +
2.4 hence - 0.6 is red in color plus 0.8
is in blue in color plus 2.4 is in blue
in color and the corresponding
probabilities are given which I'll go
back to verify is 1 by 10 11 by 30 16 by
30 from where do we get we get it
from the open
structure did I do anything wrong no we
get it from the closed structure sorry
we are in the last ter in the Clos
structure the probabilities are
03 by3 .11 by 316 by3 so if I go 03
by3 comes out to be 1 by 10 and the
corresponding values of 11 by 30 16 by
30 are for the last two
values so once you have that the total
diagram is done now I need to do the
calculations from the right to the left
and here is where we'll again try to
find out for the first case we will
basically for this
arm we'll multiply the probabilities
with the Net Present Value so
probability of 4X 5 into 6 and remember
I'm not I'm showing it so obviously 6
will be negative because a negative
Value Plus 1X 10 into 88 is positive
plus 1X 10 by 2.4 2.4 is positive So
based on this value you compare with d22
which is
zero then you calculate the values of
this
D3 1 root which is 1x 2 into -.6 + 1X 3
into into plus uh 0.8 + 1X 6 into + 2.4
you compare with
d32 and finally for the case where you
have the close structure when you want
to find out d41 you find out 1X 10 into
-
0.6 + 11 by3 into + 0.8 + 16 by 30 into
+ 2.4 you compare with
d42 so once you have compared then when
you reach D2 D3 and D4 the values will
be found out so if they are
negative then obviously you will take
the negative value or you will take the
value zero y z if you remember in Theory
I did discuss we can consider the values
if they are negative as zero so once you
reach D2 D3 D4 the corresponding values
which you will multiply for D2 would be
4x10 because there the no structure
probability for D3 it will be 3x10 and
D4 it will be 3x 10 and correspondingly
once the overall value is found out for
C2 remember that 20,000 value is there
that has to be incorporated and the
problem solved accordingly similarly
when fin U the variance can also be
brought in the picture the ratio of
variance to expector Value can also be
brought into the picture the ratio of
say for example um expected value to
variance can also be brought into the
picture and the problem solved
accordingly so with this I'll end this
class we have discussed in detail two
problems in decision trees Ive solved in
detail about expected value it can be
extended for other criterias also the
slides would be shared with you I won't
say the slides they are the word
document will be shared to with you when
you see the lectures please refer to the
diagram they can be variety of problems
but once you get the hang over of this
uh this this flow of how the problem can
be solved you can solve it accordingly
but remember from left to right the
diagrams are drawn the chance folks C
and D the C and D are not the
concordance and discordance
probabilities are to be considered
conditional probabilities are to be
considered Net Present values have to be
considered provided is expected value in
other case you can consider the variance
also so with the again I'm saying I'm
ending this class in decision trees and
discuss a simple
problem uh for the last set of lecture
for this whole course thank you very
much and have a nice
[Music]
day
[Music]
[Music]
