Launching Smart Zone Settings ⚙️ - Announcements - Rachio Community
Launching Smart Zone Settings
You have selected 0 posts.
3.7k views 79 likes 2 links 20 users
post by james.foley on Apr 30, 2020
james.foley
Machine Learning Engineer @ Rachio
Hey Rachio community . I’m James, a machine learning engineer at Rachio, and I’m excited to tell everyone about the first machine learning (ML) feature we’re launching today… Smart Zone Settings. The TLDR is that this feature will help optimize your controller settings for Flex schedules, keeping your lawn healthy while saving you water. Keep reading for more details on how the models work and how we measure water savings.
Advanced watering schedules like Flex Daily require a variety of lawn settings to operate efficiently. Some examples include soil type, crop type, root depth, crop coefficient, sun exposure, slope, and spray head type. These settings can be applied through the mobile app, but they’re often tricky to set accurately.
Naturally, three of the most vital settings for advanced schedules also happen to be the most challenging for users to set: soil type, root depth, and crop coefficient. Since these are the most difficult, we’ve chosen to optimize these three components for advanced schedules through our new ML models. The main purpose of these ML models is to apply smart settings for a newly registered controller without requiring users to do so, making for a streamlined user experience. (Note: users are still free to modify their lawn settings at any point in time; these ML generated Smart Zone Settings are merely the default values initially set at device startup time.)
Behind the scenes, we’ve trained ML models using millions of data points. The models consume the controller’s climate region and the zone’s crop type to deliver optimal predictions for each of the three components. These updated default settings are personalized and smarter than the previous defaults and according to testing, we’ve found they will save our users water without adversely affecting crop growth.
Evapotranspiration (ET) is the process by which water is transferred from the land to the atmosphere by evaporation from the soil and by transpiration from plants. We take into account ET and precipitation to determine how much additional irrigation a lawn needs to remain healthy. Using these historic measurements we can run an irrigation simulation across Rachio controllers that estimate how much water would’ve been irrigated if the Smart Zone Settings were applied. We compare those numbers to the observed irrigation numbers using the current lawn settings and see up to 15% in water savings for Flex schedules, depending on the month.
Based on this thorough evaluation, we’re confident that these machine learned Smart Zone Settings will not only remove the challenging step of initially tuning your Flex watering schedules, but will also yield water savings without hurting your crops.
- Latest upgrade seems to have been created in a vacuum
- I find rachio to be way more complicated than it needs to be
post by laura.bauman on Apr 30, 2020
post by benpaige on Apr 30, 2020
This is so cool James.
post by franz on Apr 30, 2020
post by Kubisuro on Apr 30, 2020
This sounds amazing. Can’t wait to learn more about Smart Zone Settings and hopefully try it out soon!
post by Gene on Apr 30, 2020
Sounds great, anyway to see the suggested settings for those of us already using flex schedules? Would recreating the schedule set the defaults based on the new models?
post by tmcgahey on Apr 30, 2020
I messaged him with the same question @Gene. I might have to set up one of my old controllers and create a fake yard. I assume this will work on Gen1’s?
post by gizbug on Apr 30, 2020
So how do we get these new settings?
post by joylove on May 1, 2020
Would it be possible to back up our existing settings, and then try the new ones? Like you could flip a switch to go Smart, but if you unflipped it, your old settings where not lost.
post by robross0606 on May 2, 2020
Same question about existing systems. Without Rachio addressing this, it feels more like a cash-grab than a new feature for the community. Does this only work with new controllers?
post by andylon on May 2, 2020
Please include guidance from local authorities in the mix. With drought-like conditions returning every summer, the local authorities restrict use of water (on days and the amount to be applier). Happy to help if any more inputs are required.
post by franz on May 2, 2020
The easiest way to see how it generates new zone defaults is to pick a zone and change the crop type. The system will render new root zone depth and crop coefficient values. For initial controller onboarding you could see different soil types. Also want to call out that there were noticeable differences in different climate regions for annuals and garden crop type root zone depths, but for the other crop types in most locations the variance was about 5%-10%. Very much impactful in the larger picture but for individuals it might not be as noticeable. Excited to start rolling machine learning technology to other aspects of our system (i.e. scheduling and other impactful areas).
Hope this helps.
post by Kubisuro on May 2, 2020
Aha! Just changed the crop type from Cool season grass and back- and now see the ML values. For my grass they seem reasonable.
post by Anthony on May 2, 2020
Same here @Kubisuro - I saw how the crop type changes but for this time of year isn’t crop for cool season grass should be 0.95
post by Linn on May 2, 2020
A BIG thanks for calling this Machine Learning, and not AI — we need to see more of this!
post by gizbug on May 2, 2020
I see nothing that says “crop type” when editing one of my zones… using latest IOS app with R3.
post by Linn on May 2, 2020
I just played with it. Pretty sure @franz meant to say “Zone Type”
post by scorp508 on May 2, 2020
If the team has started moving into machine learning does this mean “finish before sunrise” is no longer “hard” to fix?
post by franz on May 2, 2020
Well played @scorp508, well played.
post by johnny2678 on May 3, 2020
Lol, keep fighting the good fight @scorp508
post by scorp508 on May 4, 2020
I thought I’d give this a go so I changed my zone type from cool season grass, to warm season grass, and back to cool season grass on all of my zones. I did see current soil moisture percentage points change a few points per zone after the flip/flop. For example if I was at 46% moisture it was something like 49-50% moisture after flipping from cool to warm to cool. It went up a small number of points in all zones.
Would all previous custom advanced settings be retained? I’m sort of curious if I should reset all of my zones to default other than zone type, spray head, soil type, exposure, and slop to see how the new smart zone settings pan out.
post by inkahauts on May 4, 2020
I’d like to know as more and more info is added will it auto adjust the settings over time should it make sense for it to do so?
post by Fixer on May 4, 2020
I live in Tampa Fl and have St. Augustine grass. I just switched my zone type from warm season grass to cool season grass and back to warm season to see what the ML smart zone setting will change to. Originally I had my root depth set to 5 in. After changing zone type the ML setting set my root depth to 9.8 in. I think 9.8 is way too deep for St. Augustine root depth. Appreciate any input from any experience Florida landscapers or lawn care specialist.
post by james.foley on May 5, 2020
Users will have more information specific to their lawns than our ML models do, so you should definitely use that local information to update your settings if you can. These ML models are not perfect, but they’ve proved better than our original default settings.
post by james.foley on May 5, 2020
If we adjust (re-train) the models to make them better as we get more information, we’ll make another announcement about it and similarly not silently apply the new settings but allow existing users to opt in.
post by scorp508 on May 5, 2020
A couple questions
How would customers who don’t visit the forum know? Or even as someone who visits the forum how would I know which version of a ML’d model I’m using?
Maybe some in-app notifications “We’ve updated our models, go here to learn more or click here to apply new defaults.” along with some kind of revision # or date (applied vs most current) in the zone may be useful.
p.s.
When a new model is applied to a zone what setting(s) may be updated, and can the app detect custom values then prompt the user if they wish to override or retain custom values?
post by mnichols on May 6, 2020
This whole thing is still way too complicated. It’s clever that a super geek gardener can tinker with these variables, but I have no idea what my root depth, soil type, crop coefficient, trans evaporation rate, etc., but I know when my plants aren’t getting enough water. Just give me an option to say “My plants aren’t getting enough water” and let that feed the ML algorithm.
post by james.foley on May 6, 2020
I’m glad to see support for this idea from the community, as we’re already planning out the details of this exact kind of feedback loop functionality now. I might follow up for more feedback on this soon.
post by tmcgahey on May 6, 2020
Awe, come on! That is where all the fun is at!!!
post by mnichols on May 6, 2020
Lol I don’t disagree that some people find it fun to dial in these variables. I’m the lazy, too busy user that just wants it to work (you can probably already guess that I’m an Apple fan boy also). Not trying to offend the great work that’s gone into this, but I think it can be simpler for dummies like myself.
post by Gizbug on Jul 14, 2020
What’s the next step in the ML process here and when should we expect to see it rolled out ?
post by james.foley on Jul 14, 2020
We’re working on the next version of ML driven watering schedules right now, focusing on simplicity and efficiency. We’ll keep the community updated as we approach the Beta testing period.
post by crosbyg on Aug 15, 2021
Just curious…. How is this AI or ML if it doesn’t have a feedback loop to actually learn something? Isn’t this just big data analysis? I am asking because there actually seems to be a big opportunity for AI and ML. Add soil testing and intake simple phone pictures for lawn health trends? Send everyone a soil probe and a test kit, stop pushing thrive too hard, and I’d bet you’d have a horde of loyal fans added to the ranks.
And seriously, why is “stop before sunset” hard again?
post by Gene on Aug 15, 2021
I think hard part is having to rewrite a bunch of legacy code to support a fixed end time, rather than a scheduled start time. The math is not hard, you are right, when you choose a stop before sunset option, Rachio does exactly that, finds sunset time, subtracts the maximum time a schedule can run and sets that as a scheduled time.
post by crosbyg on Aug 15, 2021
Ahh. Maybe perfect is the enemy of good here? I don’t think we are going to whip out our stopwatches and watch for the green lantern… oh wait, that’s sunset…
At any rate, I learned something new today; I hadn’t realized this feature is there!
Cheers!