Dynamo from BIM Automation to Generative Design - Part 2 of 4

JACQUELINE ROHRMANN: So great to have everyone back. Now that we've learned a bit more about Dynamo, we're going to look more into what you came here for, which is the generative design part and the refinery. So let me explain a bit of background for what generative design is about.

So if you look at the traditional design process, it can sometimes feel a bit like a game of Battleship where you create one design and then you test it and see if it works. And then you notice something that doesn't work, so you try out something else. And you go back and forth like that. But at some point, you will run out of either time or money. So you have to pick a final design.

But through generative design, we can test a lot more designs at a time and, this way, get better results. And how this works is it actually mimics nature's process of evolution. So in nature, we have mutation that creates different versions of a species. And then through natural selection, those that aren't adapted as well, they go extinct. And the ones who are adapted better, they prosper and they create more offspring. And that way, through years and years, we end up with very well-adapted species.

And if we translate that into the computer world, we of course don't need millions of years but to only a few minutes or hours. The workflow basically is it generates a bunch of designs, evaluates them, and then evolves them with something that is very similar to reproduction and mutation. And it repeats this process for many generations.

And so now, if we want to create a generative design study, we need to do a bit more preparation than we would do in the traditional design process. So first, we need to develop a concept, basically. So we need to understand from the beginning what we want to achieve, what our design, what our building, what our infrastructure project is supposed to look like. What goals is it supposed to fulfill?

And then define a problem that we have and that we want to solve. And also, find measures on how to evaluate whether a possible design is good or bad. And how do we measure the quality of a design so that we can find better designs and see which one is bad and which one is good?

So in the center of every generative design study is a parametric model. And the parameters of that model are all the things we don't know. So say we're looking at the buildings. So that would be, for example, the length that we have to decide on, the number of levels, where to position it, the orientation, and all that. So those are the things that we want to define in the end.

And then on the other side, we have the design goals. So what I said, the measures that we need to define on how to evaluate the building. So in the end, what do we want to achieve with that project, with that design? So those can be very simple.

And for example, we want to create maximum space for people to move around in. Or we want to minimize the rent that the user has to pay. But they can also be very complex, like, how much daylight exposure will I have at a desk in an office building, and we want to have as much daylight exposure as possible for everyone? These are the three most important parts of a generative design study.

And then when we have that, so the parametric model, we do design in Dynamo. That's what we're going to use Dynamo for. And then we have an add-in, Refinery, which can take these parameters if we define them as inputs-- like you see here-- it will be able to twitch those parameters.

And then on the other side, take a look at the design goals and measure the performance of a building. So it would create different configurations of the parameters. And then on the other side, you have the design goals defined.

And then it sees the outcome, the results, and knows whether or not to go into that direction more or not. And it can learn and produce better designs over time. So the whole process goes from you start out with a concept and your constraints and your goals. I'm going to go into that a bit more later.

You start out with you, the people, and define what you want to achieve, what you're looking for. Then it goes into the computer part, the AI part, basically, with Refinery, that will create iterations and optimize your model. And then it goes back to you because you don't end up with one single solution that is the optimum.

But you actually end up with a range of different designs that are equally good. And you get to choose in which direction you want to go. So if you have multiple goals at the same time, you can decide if one thing is more important to you than another. Or also, maybe look at things like beauty and aesthetics that are hard to measure in numbers and pick the design that appeals most to you.

So constraints is the things that define, basically, what our design can look like in the end. So it's what limits the design space. Because these are things that you can't change. So you have to look at the site requirements and health and safety requirements or city regulations and all that stuff. And those are the limits of your design space.

And then within those limits, you can create a design concept. So as I said, you need to define what parameters you have. So what are the variables, the options, the things that you don't know yet, that you haven't decided on yet, that you want to optimize?

And this is a critical step. Because if you have too many things and you let it decide on all things and don't make any kind of constraints, you just end up with a very huge design space, a too-huge solution space, that you won't be able to explore fully. And you might not be able to find the best solution.

But if you have too few inputs, you might exclude a very good option. So you have to put a lot of thought into that in the beginning. Does the room have to be rectangular, or can it have any shape? But if it has any shape, I might get a really crazy result. So that might not be a sensible parameter to create.

And then the second half of the problem is the goals. So how to measure the design and how to measure the performance that-- so for example, a design goal would be to maximize comfort for the people living in that building. So then you have to think of, how can I measure the comfort? And you need to find indicators for your design goals that you can actually calculate and put into numbers so they get comparable. And that puts all of the designs on an objectively comparable level so that the algorithm can work with them.

So you start off with your design concept. And then what Refinery will do is it will create different variations of the design, so with different parameter configurations. And then it will evaluate those and give you back their scoring in the design goals. And that way, it can know which ones perform better and then let those evolve into the next generation.

But in the face of multiple design goals, how do you determine actually which one is better than another? Because some might be scoring very high, like this one objective, but very low and in another. So how can you say that? And this is where the concept of nondomination comes in.

And so let's say we have just two design goals that we both want to minimize. So for example, we want to minimize unused space and we want to minimize cost. So those would be the design goals.

And then each dot is one design solution that has some scoring on these two goals. And then how you say one is better than the other is by saying-- if it scores better in all of the objective, one solution dominates the other. So for example, C would be dominated by both A and B because they score lower, which in our case is better in both of the design measures.

Was that kind of clear? And this is why you don't end up with one solution but with a range of different solutions, which is called the Pareto front. So for example, one non-dominated solution would be this one. Because it has a very good scoring on this axis and there is no other solution that scores better in both of the objectives. But still, there is variation. And in the end, you can decide whether you want to go more in one direction or the other.

OK, yeah, and this process repeats itself over how many generations you tell it to. So it generates a population of designs. Then it evaluates the outcome and transfers those that perform best into the next generations where they create offspring. And then this process is repeated until you either tell it to stop or it reached the best possible solutions.

One thing that I forgot to mention is we also have the concept called utopia point, which would in this case be here. So minimum on both scales, which is not possible. Because usually you have some kind of trade-off situation, like you can't minimize one without going up in the other. So this is just one thing to keep in mind, to know in which direction you are trying to go. So what we would want to do in the optimization is to lower the Pareto front as far down to that utopia point as it would be possible.

So the population size is then how many design solutions you have in one generation. And then the number of generations is how many times you repeat that process over and over again. And then if there are no questions, then we would start to look into Refinery. OK--

So the first thing, the first study that we're going to do, is very simple. So we're just saying we have three boxes that could represent buildings. And one is fixed, but the other two can move around, and they can all change in their height. And what we want to do is we want to maximize the volume-- because that will be the space that we can rent-- while also minimizing the surface area, because the facade is the most expensive part of the building. So we want to find the position-- how do we have to move them into each other so that we still get the maximum volume while minimizing the facade area?

So let's open up Revit. And if yours is quicker than mine, just open an empty project. Just go to New and then create one from the template. And then open up Dynamo. And open from the second folder the Introduction to Generative Design. The Dynamo file is in [? three ?] [? box. ?] OK-- everybody there?

OK, are you seeing the same thing as I am? OK, cool. OK, so this is what a graph that we could use for Refinery looks like. So on one side, we have the inputs, which would be the parameters of our parametric model. And then we have a graph in the middle, which is just creating the geometry that we want to have. And then on the other side, we have the outputs.

So to Refinery, what happens between the inputs and the outputs doesn't really matter. But it is for us to create the geometry and then to calculate the outputs as well. So if we take a look at-- let's start with the inputs. So as I said, we have those three boxes. And then each box has a height that is parametric. And then the second and the third box are also parametric in their location.

So if you're in automatic mode and just start moving around those parameters a little bit, you will see [INAUDIBLE] you will see how the model in the background changes, and then location as well. And because we have Refinery installed, if you Right-Click on one of the nodes, you can see that they are defined as is input. And that will give Refinery the authority to-- what we just did manually with moving the slider it will do automatically and try out different values for those parameters.

And then on the other side, you can find the two outputs that are our design goals. So as I said, the surface area and then the combined volume of the [INAUDIBLE] And when you change the parameters, you can see that on the other side the values of those two outputs change as well. And so if you Right-Click them, you can see that they are defined as is output.

And so that way, Refinery knows that these are the ones that [INAUDIBLE] measures and the design goals and can read their value. And so what happens in between here is just it creates a [? queue ?] point from the point defined by the positions and then makes a solid bi-union out of those three and just returns the area and the volume, so essentially what we want to have in the end. So before we start Refinery, just quickly Save As just as a new shorter name and put it somewhere just to make sure.

OK, and then we can click here on the [INAUDIBLE] on Refinery. And the first thing we need to do is we need to export this script so that we can use it in Refinery. So I'll hit Export. And then it runs a validation process. And it will tell you, hopefully, here that the validation is complete. So you can click on Export. And then we're ready to launch Refinery.

So this is the Refinery UI that you should see. And we're going to hit New Study. And then the file that you just exported with the name that you gave it should come up here. Those are all the scripts that are available for the Refinery study. So select the one you just made. And now we get-- did you all come to this point now? OK, cool.

So what we can see here is the settings of the new study that we're going to create. So we can select what inputs we want Refinery to be able to change. So that's all of the inputs that we created.

And then here the outputs come up. And now here we have the option to either minimize or maximize. So as I said before, the surface area, this time we want to minimize. And the volume we want to maximize.

And then here at the bottom, you have the generation settings. So one is the population size, so how many solutions it will create for each generation. Let's maybe put a bit higher number, like 50. And then here you can choose how many generations it should run for us. So let's say 20. And now we should be all set to hit Generate. And we need to wait a moment.

But now you will see the first designs coming up. And you can see here all the variations that it creates. And it also gives you here in the UI this little graph to check the data.

So what will be maybe more sensible to do is to sort them by their surface area on the y-axis, and then on the x-axis as the volume, because that's what we're interested in. Did it work? So cool, so you should get something similar like me, which is a straight line. Can anybody tell where the utopia point would be in this example?

AUDIENCE: [INAUDIBLE] bottom right [INAUDIBLE]

JACQUELINE ROHRMANN: The bottom right? Well, it depends on-- so we have the surface area on the-- yeah, exactly. Because we want to minimize the surface area, so the lower the better. And then we want to maximize the volume, so the further right the better. Yeah, exactly. And what we see here now is that it comes up in a straight line. Because for every surface area, for every combination that there is, there is one position where the volume is at its best position.

And so to try out something different, if we create another new study. And now you see here in the top last time we chose Optimize. But there are actually other different options in Refinery.

So for example, Randomize would be just to get random solutions without the algorithm that improves the solution, but it just generates a bunch of random solutions. So if we try out that one, we can see here that we can just tell it to create a number of solutions. So let's again put in 48 and then click Generate.

And then if we sort by surface area and volume now, you can see the difference is that the solutions are much more all over the place and not as neatly in a straight line as before. Because those are the optimal solutions, and the other ones were just randomized. Any questions so far? Yeah--

AUDIENCE: What's the Randomize method [INAUDIBLE]

JACQUELINE ROHRMANN: It's quick, and you can just create more of a variety because it just gives you solutions that are all over the place. So if you want to look at what the entire solution space looks like, you could use the Randomize version. Or if we-- here you probably saw that there's also the cross-product.

So our inputs we defined with the sliders. We defined a range in which the inputs can vary, in which the parameters can vary, so here. And if we do a cross-product study, we can say, give me 10 steps inside this parameter combined with 10 different variations of the other parameter. And show me what those combinations would look like.

So yeah, that gives you a very structured way to explore the design space. So here cross-product, and now we can say, [INAUDIBLE] maybe. Let's put [INAUDIBLE]. You see that the number of total solution gets higher the higher the numbers you put, because it's multiplying them.

Maybe because we have so many inputs, let's just stick with two and hit Generate. And so it created 228 solutions. And if we look at this graph now, we can see that for each parameter, it created all the possible combinations. I gave it two parameters and here it evaluates them.

So for our optimization study, this graph looks very different. Because it only used those values of the parameters that were sensible. While here it used those parameters which I told it to use. Or in Random, you can see that it's very evenly distributed. Because it was just generating random values. And here it was only sticking to those that were proved to return good results. OK-- yeah-- sorry.

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: [INAUDIBLE]

AUDIENCE: Regarding Randomize process, if it's randomized, we should have a different solution every one. So maybe it's strange, but we have same solution as you when we do Randomize process. And [? how to ?] Randomize works at that [? moment ?]

PRESENTER 1: So when you go to the new study there, there's a value called seed, which is a way to initialize. And if you change to Randomize, I believe it has a seed, too. So it's kind of controlled randomization based on that seed. So if you run it again with the same seed value, it'll start from the same place and give you the same results.

JACQUELINE ROHRMANN: Does that answer your question? [INAUDIBLE] OK, or do you mean-- what are you talking about, the distribution of the results?

PRESENTER 1: I think that was his question.

JACQUELINE ROHRMANN: Yes-- wait, wait.

AUDIENCE: If you find a particular design that you like and you click on the button, does it show you that graphically? Does it show it to you? Because mine doesn't do that.

JACQUELINE ROHRMANN: So if you like this design, for example, it should highlight the curve of it. And if you look at the graph, then it will-- does that not work for you?

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: Yeah, well-- it doesn't actually do that, does it?

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: Yeah-- or do you mean with the filters?

PRESENTER 1: So we don't have a good way to find the page of this dot right now. But what you can do to find it is to use the filters here. And sorry if you were going to show this later.

So let's say we want to find this value right here. We can filter down by turning on the filters there to narrow down our selections. And then we can kind of find this one. So it's a little bit of a workaround right now.

JACQUELINE ROHRMANN: And then also, if you found a design that you like and you click on the 3D preview, it actually will change the values in Dynamo to the values of that model. So this is how you get it from Refinery back into Dynamo. Any other questions? OK, so you can just close that example. So Lily, were you going to--

PRESENTER 1: I just going to say one more thing. Could you show the range values and the effects of that? If you just open that-- if you zoom in to the inputs there. Just switch back to the graph, yeah. So if you zoom into the inputs there and you [? full ?] down-- yeah, so this maximum and minimum values here are critical to setting up your design space, and also the step value there.

So if you set those, the step we have right now, it's going from 19 to 200 by 1. So if you're considering the height of your box, maybe that's a little bit too granular, right? Maybe you want to, if you're thinking in feet here, maybe you want to test every 15 feet or something like that. Because that would be more of a floor-to-floor height.

And it would also make your design space smaller so that you were only getting values that were acceptable to you. So just to point out that Refinery will respect that range and step value that's set there. And it could be a really important thing in controlling your design space.

JACQUELINE ROHRMANN: Yeah, for sure. Where do I have to [INAUDIBLE] I don't know. OK--

OK, so next thing we want to try out is we're looking at an urban context. And we want to create a high-rise. And so there is this area like volume profile where we can build the high-rise in. And what we want to do is we want to minimize the unused area, so to use the full potential that we could use. And again, minimize facade area, maximize the total floor area. And if we have the time, we're going to look into solar exposure on each floors and on the roof, because maybe we have a solar roof as well.

OK, and then what we don't know about our tower yet is what shape it's going to have. And then so maybe we decided to have it split into three parts. So which height does each part have?

And yeah, so to do that, we are going to open the Revit file [INAUDIBLE] tower, and we're going to open the urban context with park. And open the Dynamo file for the face of the tower [INAUDIBLE] exercise.

PRESENTER 1: I think there was a question about the Revit folder.

JACQUELINE ROHRMANN: Yeah, one second. Do I need to show the Dynamo file as well again?

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: The [INAUDIBLE] the face of the tower and then exercise. OK, so if you have run, then this is what you should see. So what we did here is we have-- we import from our Revit model the context, like the surrounding buildings and our height profile. And then the first thing we want to do is we want to get the outline of the site that we are able to use. So we extract that from the model.

And then what our building is supposed to look like is we have these three towers. But we don't want them to have a boring shape. But we want to have a more interesting formed rectangle.

So our first input is going to be-- imagine if this would be, the big rectangle would be, the side outline. We have the corners of-- we made one corner parametric so that we can move that around in the rectangle to create this sort of twitched shape. So that is what you can see here. If you move that around-- maybe set it to automatic-- the shape of your building changes.

And then we have these three towers. And we also want to make the shape of each of them parametric. So then we are going to split the surface that we created before into three parts. So we have three parameters to decide where the splitting is going to happen. And those are these three.

So when you change them around, you see that one part gets bigger or smaller. And then the last thing is the height of each of these three tower parts. And because we don't have much time left, I'm not going to look too much into how we generate the design.

But this time we are going to create the evaluators ourselves, like the design goals, the outputs. So first one is the open area, so the unused area. So what we need for that is, first, I need the area of the entire site, and then the area, the footprint of our building. So what we need is a surface. I don't think my keyboard is working. And I can't Right-Click either. [INAUDIBLE]

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: This keyboard--

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: Ah, OK, perfect. OK, so what we want to do is we want to subtract the footprint of the building from the entire site area. So we need a surface subtract from. And then our-- ah, that's where it's hiding, here. Here is the site surface.

And then here you have the three surfaces that are the base of the tower. So but to make sure that they actually meet, we're going to thicken them. So here, Surface, Thicken.

And then-- and now from that surface that we gain, which is the unused area, we need to get the actual area. And so to make this an output, we can just use a Watch node. And by Double-Clicking-- oh, first we need to sum that up by using a math sum. OK--

PRESENTER 1: If you have lists that are feeding into outputs in Refinery, it won't work because it won't be able to give you a dot in the right place.

JACQUELINE ROHRMANN: Yes-- but first, we need to rename the note and give it a distinct name. So in this case, this would be the open area, which we want to minimize. You can rename it by Double-Clicking the Watch. And then when you Right-Click the note, you will see that here you can set it to an output. And this is how you created your first Dynamo output.

So next, we want to calculate the total facade area. Does anybody have maybe any ideas how we could do that? If, for example, like here we have the entire building as a solid. And then now we want to get all of the facade area. So one way to do it would be to explode it so we can get all the surface areas. And then we would need to take out all of the surfaces that are roofs or floors.

So there is actually a very handy node that we can get just to only get the facade. So that is, if you type in building and then deconstruct, and then you have here Deconstruct Facade Shell. And then you can just hand it the solid. And it will turn your dictionary with all the surfaces.

Now, again, now we have the surfaces. But in the end, what we want is the area so that we have a number. So let's let it calculate the actual area of those surfaces. And then, because we get a list, we need to sum that up.

And again, we can use a Watch node to create the Refinery output.

PRESENTER 1: The Building Deconstruct node is coming from a toolkit that we made called the Refinery Toolkit. And we have it installed on these machines. There's some instructions on how to get it in the Content folder. But we're experimenting with ways of just making it easier to do some of these building-type operations, making these conveniences.

AUDIENCE: [INAUDIBLE]

PRESENTER 1: We're planning to put it on the [? PACT ?] Manager soon. But right now, you can get it on Dynamo GitHub.

JACQUELINE ROHRMANN: And then by Right-Clicking it, we're making it an output.

And then our last evaluator, our last output, is the floor area. So what here in the generator it does is, once it has the solid, it will slice it up into levels. And we can now just take those. They are in a list here. And take out their surface area again. And that will be our floor area.

And because they are packed in three lists for the three different buildings, but we want to get the floor area, we just need to flatten the list. So you can see it has the same values as before. But it just doesn't have this list in a list structure anymore so that you get one list of all the floor areas.

And then we can create the sum from them and create another Watch node. That is the floor area. And set that to is output.

Any questions? Because then we would be ready to now save that again, Save As, and then use that in Refinery. So export and launch Refinery. And then if you create a new study, you should now have the option with the name you gave it.

Oh, but before we create, let's first-- because otherwise we're not going to see very much-- let's switch off all the surrounding buildings. Otherwise in the 3D preview, we're going to see only the other buildings. So here it's called color site context group. Just disable the preview. And then you only end up with your tower.

OK, save that again. And then launch Refinery. Do I need to export it again? OK, so make sure to export it again. And then create a new study.

So we said the open area we want to minimize. The facade area we want to minimize as well. And then the floor area we want to maximize.

And now let's see. If we put a very small population size like 20 and the generation of 10, let's compare those results to when we increase the population size or the number of generations. So let's start out with a small population size.

PRESENTER 1: So Refinery here is actually spinning up six different Dynamo cores on your machine and running them all at the same time in order to speed things up. It's running headless Dynamo. So maybe we should talk about at some point the Remember nodes in there. Yeah, so it does that to speed up the operation and give you results.

AUDIENCE: [INAUDIBLE]

PRESENTER 2: All of the above.

[LAUGHTER]

Typically I've found that CPU power has more value because less stuff is actually being stored in memory. It's basically just quick run it and spit it out. It's unlike standard Dynamo sandbox where every single node you have on canvas persists in memory forever. Every time it spins up a new instance, all that RAM is wiped out.

If you somehow had that much RAM need, I don't think you would be able to write the graph. That would tax the RAM. Whereas CPU compute time seems to matter a little bit more for getting the quick results. But that's entirely just based on my somewhat extensive use.

JACQUELINE ROHRMANN: So let's sort them now by facade area compared to floor area. And let's see if we can get nicer results if we put greater population sizes now. So let's just put a population size of 100 and let it run for 10 generations again. Because I see that we're running out of time. And you probably want to go to lunch.

PRESENTER 1: --till 12:30.

JACQUELINE ROHRMANN: Oh, we have until 12.30. Oh, then we can do more fun stuff that way.

OK, I broke-- maybe I was a bit ambitious with the with the 100. Huh, but it's not actually closing it.

PRESENTER 2: You may need to check Task Manager. Never mind-- it's running, it's running.

JACQUELINE ROHRMANN: I mean, it is still running. [INAUDIBLE]

PRESENTER 2: Just do a Right-Click on that and say in process [INAUDIBLE] because the core that crashed, it's not telling that Dynamo core to restart in Refinery.

JACQUELINE ROHRMANN: OK, so what do you want me to do?

PRESENTER 2: Right-Click on [INAUDIBLE]

JACQUELINE ROHRMANN: Oh, [INAUDIBLE]

PRESENTER 2: Refinery [INAUDIBLE] Right-Click on it. Go to Details. Right-Click on it.

JACQUELINE ROHRMANN: Right-Click on what, on Refinery again?

PRESENTER 2: Yeah, and say [INAUDIBLE] It should have stopped Refinery entirely. [INAUDIBLE]

JACQUELINE ROHRMANN: Do you think I will run into the error again? Do you think I will run into the error again?

PRESENTER 2: [INAUDIBLE]

JACQUELINE ROHRMANN: But can see you see that it's still running? Is that good, or should I stop it?

PRESENTER 2: [INAUDIBLE]

JACQUELINE ROHRMANN: I'm sorry?

PRESENTER 2: Take questions [INAUDIBLE]

JACQUELINE ROHRMANN: Yeah. Do you have any questions for me while we wait? Sorry, [INAUDIBLE] the mic.

PRESENTER 2: Hey, everybody.

AUDIENCE: When you set the minimum and maximum, can you set defaults to those so it goes to maximum when you need it to?

JACQUELINE ROHRMANN: Not at the moment. But I'm sure Lily can tell us if that's--

PRESENTER 1: Could you ask that one more time?

JACQUELINE ROHRMANN: If you can set a default to the minimum or maximum so that it remembers when you create a new study that you don't have to set it every time, like in the refinery UI.

PRESENTER 1: Well, in the Refinery, in Dynamo when you set the ranges for a minimum and maximum, those are set to certain values--

JACQUELINE ROHRMANN: [INAUDIBLE] whether we want to minimize or maximize a goal.

PRESENTER 1: Ah, we do not save that information right now. But we think that that is going to be really important for-- so as I was saying this morning, this class is aimed towards the script authors, people who are writing this stuff for maybe other people to use, right? And so being able to set up what a study does and what it maximizes and minimizes is critically important.

Because you might have a study that is-- well, you're going to get more results if you have outputs that are kind of fighting each other. Sometimes if you set both things to minimize, you'll just get one answer, which is not really what you're after with multivariant optimization. Does that answer your question?

PRESENTER 2: One thing which I've done in the past is to always minimize. So I'll just multiply by negative 1 to get that value so it's consistent. And then I don't have to worry.

JACQUELINE ROHRMANN: Yeah, what he's saying is the default as you saw it in the UI is always set to minimize. So if, for example, we would now have multiplied the total floor area by minus 1, we would also want to minimize it. Yeah, so this way, you don't have to set them. Because you will have all at minimum. But then it's more confusing to look at maybe, because you have negative numbers for the floor area. So it's kind of a question of what do you like more.

And what Lily was saying is that in the generative design study, it is always good to have opposing goals. So one thing that you would-- things that are fighting each other, as she put it. Because otherwise if everything goes into the same direction, then you will not end up with-- you need these trade-off situations where if I maximize one thing, the other thing is going to go down, and so that you can find interesting results.

Because if you just want to minimize something, it will just go to the smallest possible solution. But you could have figured it out yourself. So this is something where generative design comes in. It's particularly helpful if you have these kind of trade-off situations, right?

PRESENTER 1: Yeah. Do you want to explain a little bit about the Remember node and where that's placed in here?

JACQUELINE ROHRMANN: So if you install Refinery, for one thing, you get the Refinery button here. But you also get a package with custom nodes. But it's just one node, which is the Remember node. So what that does is-- we used it actually here in this graph as well.

When you import geometry from Revit, like we did here, then this node will be able to-- if you use the Select Model Element node and then import an element from Revit and then close the script and open it again, then you would need to re-import the geometry from Revit. It doesn't save it in the Dynamo file.

But if you use the data Remember node, it will actually store the geometry for you. Because Lily said the Refinery runs Dynamo on multiple [? threats. ?] It doesn't need to import the geometry from Revit every time. But it's actually stored within the Revit graph. And therefore, it can run much faster. Is that what you were getting at?

PRESENTER 1: Yeah, I mean, because it's writing these six headless Dynamos behind the scene, it's not starting up Revit every time it does that. Because that would take a long time.

And because most of the use cases that we were seeing from people, people said, I just want to use a little bit of Revit data as a starting point. So we developed this way to cache the data inside from Revit and be able to use it in these generative studies. Now, there may be more things that people want to do with the parametric elements that are inside Revit using some of these methods. We'd love to hear about your ideas for that. But this is kind of where we started as a way to do this and do it fast enough in running these things over and over again. Yeah--

[INAUDIBLE]

JACQUELINE ROHRMANN: One second, if you wait for the microphone.

AUDIENCE: Can you manipulate the number of Dynamos that are getting launched?

PRESENTER 1: We do not have control right now over the number of Dynamos that are being launched, although that is something that a lot of people have asked for. Because some people have very powerful machines that they want to run 20 Dynamos at once. They want to do a bigger design space, a bigger exploration. So we're looking into-- talk to Nate.

[SIDE CONVERSATION]

JACQUELINE ROHRMANN: OK, so I think, as Lily said in the beginning it's still in beta version. Sometimes you run into errors and just get stuck. So has it worked for somebody with a 100 population size? Are you still going? At six, oh, yeah.

PRESENTER 1: Do you want to show a little bit more where I have this up of the Explore interface and show the tabular data?

JACQUELINE ROHRMANN: So we already looked into this chart. And just like in this graph here, we can also here use filters. So for example, if we say, I'm only interested in buildings that would have more than 80,000 square feet of floor area, this would be a way to limit the results that we're looking at.

And then here down here, we have the 3D previews. But we could also have a list of just the pure data that we're looking at here. So each line would be one solution with all the different values. And then you can, of course, mix those two views up. Or you go back to the 3D.

PRESENTER 1: There's also a lot of power there in just the Sort By if you-- your discussion before about ranking.

JACQUELINE ROHRMANN: Yeah, so here we can change what the solutions that we're going to look at are sorted by. So if for me facade area is the most interesting part, now I'm having them sorted by facade area either from largest to smallest or the other way around. And then I could look back at the data and see, well, if I want to have lots of facade area, what options do I have for the floor area that comes with it? Or how am I using them?

AUDIENCE: Can you do multiple [INAUDIBLE]

JACQUELINE ROHRMANN: Not right now.

PRESENTER 1: But you can use the filtering and sorting at the same time. So you can get to some of maybe what you're after by combining those two.

JACQUELINE ROHRMANN: And then here, for example, you could also change the range of the filter and then sort, yeah.

PRESENTER 1: There's a question about whether you can export that data. So we do have-- when Refinery is running, it is creating a JSON file of all your results, which is stored in the document/refinery folder on your machine. There's also, for the optimization runs, there's Excel spreadsheets that are kept of all of the solutions that it's running. So you can dig into those.

We don't have good ways to view them yet. We want to do that. Because right now, we're only returning in the optimization runs what we call the hall of fame. That is the best designs that it came up with.

But a lot of times it can be helpful to learn from the designs that were rejected, right? Why are the good ones good? And so you can look in those Excel files and you can see kind of how it progressed in the different generations to come up with the solutions that it did.

AUDIENCE: [INAUDIBLE]

JACQUELINE ROHRMANN: Do you know it by heart?

PRESENTER 1: So if you go to percent app data percent slash refinery, which just is your user folder roaming, that's where the Excel files are. I don't know what they're called, what the user is called on these frame machines.

JACQUELINE ROHRMANN: What did you say was the file path again, [INAUDIBLE]

PRESENTER 1: OK, you need to go under user, the user name, app data.

JACQUELINE ROHRMANN: Yeah, I can't see app data.

PRESENTER 1: --Refinery [INAUDIBLE] find it there.

JACQUELINE ROHRMANN: Yeah, well, I can't find the-- how do I enable the hidden folders here?

PRESENTER 1: [INAUDIBLE]

JACQUELINE ROHRMANN: Wait, let me-- I'll just go here.

PRESENTER 1: You can do it in that one.

JACQUELINE ROHRMANN: So frame and then view the hidden items. And then here app data. And then it's in roaming? OK--

PRESENTER 1: Refinery--

JACQUELINE ROHRMANN: And then here for every study, you will have four different sets of-- OK, and then for every study that you created, so you can see here that every study you created has a distinct combination, like a name. And so here you can find four Excel files under the name. And as Lily said, the hall of fame are the solutions that you end up seeing in the Refinery [INAUDIBLE], so the best solutions they have found.