Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00.05 --> 00:00:03.04
- [Instructor] When creating a new generative design study,
2
00:00:03.04 --> 00:00:05.08
we need to specify a method type
3
00:00:05.08 --> 00:00:08.03
which determines how the study will run.
4
00:00:08.03 --> 00:00:12.03
This can be either optimize, randomize, cross-product
5
00:00:12.03 --> 00:00:13.06
or like this.
6
00:00:13.06 --> 00:00:15.04
So in this lesson, let's take a look
7
00:00:15.04 --> 00:00:17.00
at what each of these mean,
8
00:00:17.00 --> 00:00:20.09
and how they can influence the generative design process.
9
00:00:20.09 --> 00:00:23.00
Let's start with optimize.
10
00:00:23.00 --> 00:00:25.05
The optimize method allows us to generate
11
00:00:25.05 --> 00:00:28.06
and evolve a multitude of design options
12
00:00:28.06 --> 00:00:31.02
to improve a solution based on a single
13
00:00:31.02 --> 00:00:34.01
or multiple objective design goals.
14
00:00:34.01 --> 00:00:36.09
In doing so, the optimization method will aim
15
00:00:36.09 --> 00:00:41.01
to either minimize or maximize some design criteria
16
00:00:41.01 --> 00:00:44.09
by tweaking the parameters that drive the design.
17
00:00:44.09 --> 00:00:48.01
For example, say that we wanted to optimize the amount
18
00:00:48.01 --> 00:00:50.02
of twist in this tower,
19
00:00:50.02 --> 00:00:54.01
to maximize the amount of sunlight entering the North face.
20
00:00:54.01 --> 00:00:55.09
There are three things to consider
21
00:00:55.09 --> 00:00:58.01
with an optimization study.
22
00:00:58.01 --> 00:01:00.08
Firstly, what are the variables that can be adjusted
23
00:01:00.08 --> 00:01:02.01
by the algorithm?
24
00:01:02.01 --> 00:01:05.00
In this example, the variables are the rotation
25
00:01:05.00 --> 00:01:06.01
of the tower.
26
00:01:06.01 --> 00:01:08.00
However, this could be many variables
27
00:01:08.00 --> 00:01:12.00
that impact the design, such as height, width, or length.
28
00:01:12.00 --> 00:01:14.08
It really depends on what we're trying to achieve.
29
00:01:14.08 --> 00:01:17.05
Secondly, we need to set some constraints
30
00:01:17.05 --> 00:01:20.04
so that the algorithm does not test scenarios
31
00:01:20.04 --> 00:01:23.05
that we know are not achievable or desirable.
32
00:01:23.05 --> 00:01:25.03
This reduces the search space
33
00:01:25.03 --> 00:01:27.08
for the algorithm giving it a better chance
34
00:01:27.08 --> 00:01:30.00
at finding an optimal solution.
35
00:01:30.00 --> 00:01:33.02
For example, with our tower, we may constrain the amount
36
00:01:33.02 --> 00:01:36.07
of twist to between 0 and 90 degrees,
37
00:01:36.07 --> 00:01:40.01
as any more than this would result in an undesirable design.
38
00:01:40.01 --> 00:01:43.02
Lastly, we need to define the objectives
39
00:01:43.02 --> 00:01:46.06
or goals of the optimization study.
40
00:01:46.06 --> 00:01:49.08
This allows the algorithm to score
41
00:01:49.08 --> 00:01:53.04
or rate the quality of life each generated solution.
42
00:01:53.04 --> 00:01:57.00
In our example, it is to maximize the amount of sunlight.
43
00:01:57.00 --> 00:01:59.04
However, this could be multiple objectives,
44
00:01:59.04 --> 00:02:01.08
such as maximizing the amount of sunlight
45
00:02:01.08 --> 00:02:05.06
while also trying to minimize the amount of cost.
46
00:02:05.06 --> 00:02:09.02
Once the optimization study has scored each design
47
00:02:09.02 --> 00:02:12.00
in the initial generation, or run,
48
00:02:12.00 --> 00:02:14.05
it will select the best performing solutions
49
00:02:14.05 --> 00:02:16.01
and then use their parameters
50
00:02:16.01 --> 00:02:19.01
to generate the next generation of solutions.
51
00:02:19.01 --> 00:02:22.02
Over time these optimizations studies improve
52
00:02:22.02 --> 00:02:25.08
by getting closer and closer to an optimal goal.
53
00:02:25.08 --> 00:02:28.04
So this method is useful when we want the study
54
00:02:28.04 --> 00:02:32.05
to improve a solution based on some scoring metric.
55
00:02:32.05 --> 00:02:34.05
Next is the randomized method.
56
00:02:34.05 --> 00:02:36.09
This study method is a little bit simpler
57
00:02:36.09 --> 00:02:38.09
as the goals don't need to be set.
58
00:02:38.09 --> 00:02:42.06
We simply set the variables that are constrained
59
00:02:42.06 --> 00:02:45.00
and then the study method will generate a number
60
00:02:45.00 --> 00:02:48.05
of random solutions within those bounds.
61
00:02:48.05 --> 00:02:50.08
The results can then be searched through.
62
00:02:50.08 --> 00:02:54.06
This method is useful if we don't have a specific goal
63
00:02:54.06 --> 00:02:56.01
that we want to optimize,
64
00:02:56.01 --> 00:02:58.03
and simply want to search a wide variety
65
00:02:58.03 --> 00:03:00.04
of potential solutions.
66
00:03:00.04 --> 00:03:02.08
Cross product is similar to randomize
67
00:03:02.08 --> 00:03:04.06
in that we don't set the goal,
68
00:03:04.06 --> 00:03:07.00
we just set the constrained variables
69
00:03:07.00 --> 00:03:10.09
for the algorithm to then generate every solution possible
70
00:03:10.09 --> 00:03:14.04
within those bounds at specific intervals.
71
00:03:14.04 --> 00:03:17.07
So for example, say we wanted to test the twist
72
00:03:17.07 --> 00:03:21.01
in our tower, but also controlling the width
73
00:03:21.01 --> 00:03:25.00
with the bounds of say 0 to 90 degrees for the twist
74
00:03:25.00 --> 00:03:27.07
and 10 to 20 meters for the width.
75
00:03:27.07 --> 00:03:30.07
With cross product, we specify the number
76
00:03:30.07 --> 00:03:35.01
of values to test with each variable within those bounds.
77
00:03:35.01 --> 00:03:38.04
In this example, it's five values for the twist
78
00:03:38.04 --> 00:03:40.07
and five values for the width.
79
00:03:40.07 --> 00:03:42.07
The algorithm will take the values
80
00:03:42.07 --> 00:03:46.06
of each variable evenly spaced by the number that we specify
81
00:03:46.06 --> 00:03:49.07
and test it against every other value specified
82
00:03:49.07 --> 00:03:54.00
for the other variables, creating a matrix of solutions.
83
00:03:54.00 --> 00:03:57.02
This is also useful like the randomized method
84
00:03:57.02 --> 00:04:00.02
to search through a wide variety of solutions
85
00:04:00.02 --> 00:04:02.08
with no particular goal to start with.
86
00:04:02.08 --> 00:04:05.03
Lastly, the like this method allows us
87
00:04:05.03 --> 00:04:07.08
to finally tune a design option
88
00:04:07.08 --> 00:04:10.00
that we've already arrived at.
89
00:04:10.00 --> 00:04:12.09
We do this by first specifying the variables
90
00:04:12.09 --> 00:04:14.06
that the algorithm can change,
91
00:04:14.06 --> 00:04:17.06
such as the amount of twist and width.
92
00:04:17.06 --> 00:04:20.01
The like this method will then generate a number
93
00:04:20.01 --> 00:04:23.07
of random solutions by adjusting these variables
94
00:04:23.07 --> 00:04:27.05
within 20% of the values that we set.
95
00:04:27.05 --> 00:04:29.04
So the end result will be a number
96
00:04:29.04 --> 00:04:33.01
of potential design solutions close to the variables
97
00:04:33.01 --> 00:04:35.01
that we have already determined.
98
00:04:35.01 --> 00:04:38.07
This is useful for fine tuning our design solutions,
99
00:04:38.07 --> 00:04:41.00
so best used when we're already close
100
00:04:41.00 --> 00:04:43.05
to the variables that we know that we want.
101
00:04:43.05 --> 00:04:46.04
Now, that we're up to speed with the different method types,
102
00:04:46.04 --> 00:04:49.05
let's go ahead and optimize the desk study
103
00:04:49.05 --> 00:04:50.07
from the previous lesson.
8489
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.