Does anyone know if they are still working on the barge in San Francisco?
Apparently they bought 4 barges, so they had big plans for this project (which doesn't seem smart without at least testing it out once first, also doesn't allow them to use lessons learned from the first build on the second barge).
What a crazy project, kudos to Google for thinking outside the box. Even if it doesn't work out I'm sure they've learned a lot from this build. Hopefully they'll share some construction video, designs, and lessons learned someday.
For a company the size of Google, 4 barges probably DOES count as a tiny initial test. I suppose 4 simultaneously has some advantage over 1 at a time because it allows you to do A/B/C/D testing.
Anyone else would have a striped + mirrored RAIB. But I suppose at Google's scale, they need geographically distributed barges to cover for a failure at one port.
You'd want aabc, at least. You need to test your treatment against itself as a control. If you see significance between control A and treatment A, you have noise.
Comments
Does anyone know if they are still working on the barge in San Francisco?
Apparently they bought 4 barges, so they had big plans for this project (which doesn't seem smart without at least testing it out once first, also doesn't allow them to use lessons learned from the first build on the second barge).
Here is a good video of the one in SF: https://www.youtube.com/watch?v=cLueLIcUrwc
Here is a closeup of one side: http://www.vosizneias.com/wp-content/uploads/2014/03/Google-...
What a crazy project, kudos to Google for thinking outside the box. Even if it doesn't work out I'm sure they've learned a lot from this build. Hopefully they'll share some construction video, designs, and lessons learned someday.
For a company the size of Google, 4 barges probably DOES count as a tiny initial test. I suppose 4 simultaneously has some advantage over 1 at a time because it allows you to do A/B/C/D testing.
Anyone else would have a striped + mirrored RAIB. But I suppose at Google's scale, they need geographically distributed barges to cover for a failure at one port.
You'd want aabc, at least. You need to test your treatment against itself as a control. If you see significance between control A and treatment A, you have noise.