Lunchbox Machine Learning

1

So, I got close, but it’s not quite working?

Capture
Capture.png3530×1380 211 KB

This ‘works’ but only if the number of rooms is less than my ‘test’ rooms, which is obviously not ideal…

image
image.png1649×629 114 KB

Also the outputs aren’t right…! I tried plugging everything in the other way round and playing with the various other inputs but couldn’t get it working.

I realise I could use IF statements for this.

The inspiration came from John Pierson and Nate Miller here… https://www.youtube.com/watch?v=e6WglyUB-7A though it’s all in grasshopper so took some interpretation.

Any thoughts appreciated.

Cheers,

Mark

Machine Learning Rooms-1.dyn (29.4 KB)

3 Likes
2

I’m actually submitting a Pull Request this week to their repo with the sample file. (And another one for the Gaussian nodes) :grin:

Right now I am training the neural network with the room width, depth, area, and perimeter (derived from specific rooms in an isolated selection). I am then placing a new room and using that data as the test data. Took me a bit too, but after dissecting their workflow on the video and looking at the source code, the inputs started to make more sense.

2 Likes
3

Thanks John,

Let us know when it gets approved :slight_smile:

Perhaps my test is not complex enough to warrant using it.

It’s interesting in that currently we have hard prescribed sizes… A Studio flat is upto 50m²… In a world of ML, we could move to… This flat is 49m² but suitable for a 1 Bed flat.

Cheers,

Mark

4

Got my Pull Request in here, https://bitbucket.org/archinate/lunchboxml/pull-requests/1/dynamo-neural-network-sample/diff

It still needs reviewed and all of that but the files are up there now for your viewing!

2 Likes
5

Hello @john_pierson, I am so curious that I have already tried and explored your sample which is so amazing. And, I going to try to train room shape data. Thanks john :grinning::hugs::sunglasses:

room_test
room_test1920×1044 197 KB

1 Like
6

Thanks John,

That’s a really smart graph :slight_smile:

I have rebuilt it (I presume your file is 2.1 and my 2.0.3 can’t open it perhaps?) and it appears to work (still quite mysterious!)

Would you mind confirming/correcting my train of thought below?

  • I would have expected to remap data sets per attribute

  • Instead we remap the data set per room

  • This gives an appropriateness per room

  • This ‘average appropriateness’ is what is used to select the most likely match.

Thanks again,

Mark

Capture
Capture.png9242×1982 705 KB
Machine Learning Rooms-4.dyn (107.0 KB)

7

image
It should be in Dynamo 2.0.3, and even then if it was in 2.1.0, it should be compatible per semantic versioning. even then, I did not use anything that would be specific to 2.1.0.

Are you getting an error of some sort?

1 Like
8

This is what I did…

Test
Test.gif1257×834 1010 KB

Probably user incompetance!

image

image

Cheers,

Mark

9

Oh that is just downloading the webpage with the name of the DYN.

Any time you see that corrupted error, do some digging. Opening in Notepad++ showed me what was wrong pretty quick.

image
image.png1331×893 68 KB

1 Like
10

Thanks! should I copy and paste the code into notepad and save it as a dyn?

11

That will do it.

Or: View Raw > Save As > Rename to .dyn

1 Like
12

Success :smiley: That’s Great!

Any comment on the below would be awesome…

  • We remap the data set per room
  • This ‘average appropriateness’ is what is used to select the most likely match.

Cheers,

Mark