Open ilsebreedvelt opened 2 years ago
@ilsebreedvelt to organize meeting for brainstorming
Max and Ilse have met to discuss design for the Advice page - Max was working from that, but has vacation this week. Ilse to set up another review - meeting planned for August 23rd
Meeting on August 26th to discuss all the designs and finish the details
Still work in progress.
@max-arbezcheung is the design complete or still work in progress for the advice page? Thanks
@ilsebreedvelt @ilfreedom Hi, here are the design elements for the Farm dashboard & Crop advice page:
Design changes to the advice page - we need to choose what to do first, what is the easiest to do?
Input Ilse wrote on July 22: We currently have the following advice in the Liquid Prep app:
To water or not, if water, how much --> can we improve with mm? Inches?
Soil moisture is low, medium, high - we probably need to also indicate the % as measured. Would this indicator of low, med, high be influenced by the need of the crop? Yes
Next watering - is there information on why they should water then… does it make sense to have this if we already provide them watering advice for now? Yes
Also, in the new design direction, we want farmers to indicate if they actually gave water. This was for later on the roadmap, but that should feed into the ML model.
Another idea we had was that this advice page needs to have - clickable/selectable elements, so users can go deeper and see more details. What kind of information on why a certain advice is given can be provided by the ML output?
Can the ML model provide “Predicted soil moisture”? Yes
Can the ML model provide “Expected harvest date”? Yes
Fearghal answered: The short answer is that for Texas pilot we can do all these things. Since we have relatively detailed information on the crop planting, harvesting, and management scheduling. We are currently developing a reinforcement learning model for crop management that identifies appropriate management actions (when to water and how much, when to apply fertilizer, etc.) We can output these decisions for liquidprep to use. The challenge there is that the more sophisticated machine learning models are pretty data hungry when we try to move to different locations. We need to balance here the data that's coming from the farmer (e.g. crops planted, planting time, etc) that are then fed to the ML model to generate predictions (probably of soil moisture but not sure if more is needed), and consequently inform crop management. The prototyping of this is pretty advanced, the key challenge will be the automated deployment to different regions that we need to consider