Collaborative repository for optimizing the placement of temperature sensors in the Great Lakes using the DeepSensor machine learning framework. Aiming to enhance the quantitative understanding of surface temperature variability for better environmental monitoring and decision-making.
MIT License
0
stars
0
forks
source link
Explore active learning features of DeepSensor for sensor placement optimization #24
Issue Description:
The task involves exploring the active learning capabilities of the DeepSensor package to strategically place temperature sensors across the Great Lakes, as part of the Great Lakes Summer Fellows Program. You will be running through an example in the DeepSensor Active Learning Documentation and considering how to apply these techniques to our ongoing project on temperature variability within the lakes.
Start with the pre-trained ERA5 spatial interpolation ConvNP model from the DeepSensor documentation's training page.
Follow the step-by-step guide in the documentation's active learning section to use the model for proposing new sensor placement locations based on specific acquisition functions and algorithms provided by DeepSensor.
Key Concepts to Understand:
What is an Acquisition Function, and what role does it play in active learning within the context of sensor placement?
How does the Greedy Algorithm operate, and why is it referred to as 'greedy' in the context of this problem?
Running the Example:
Use a GPU-enabled virtual environment (deepsensor_env_gpu) on the Great Lakes HPC system.
The example will utilize pre-trained ConvNP weights and DeepSensor’s active learning functionality with the GreedyAlgorithm class.
You'll interact with classes like AcquisitionFunction and Stddev, which assign utility values to query locations.
Engage with the documentation to ensure a full understanding of each step of the active learning process, including initializing the GreedyAlgorithm and calling it with a Task and AcquisitionFunction.
Deliverables:
A quick summary documenting your understanding of each step of the active learning example from the DeepSensor documentation.
A report detailing any issues encountered and how you addressed them, including any deviations from the documentation's guidance.
Any visual outputs (such as sensor placement proposals and acquisition function visualizations) generated by following the documentation.
Issue Description: The task involves exploring the active learning capabilities of the DeepSensor package to strategically place temperature sensors across the Great Lakes, as part of the Great Lakes Summer Fellows Program. You will be running through an example in the DeepSensor Active Learning Documentation and considering how to apply these techniques to our ongoing project on temperature variability within the lakes.
Background Reading: Please refer back to the paper 'Environmental sensor placement with convolutional Gaussian neural processes' (EDS, 2023), which provides a foundational methodology used in DeepSensor for active sensor placement, informing our quantitative framework.
The Task:
Key Concepts to Understand:
Running the Example:
deepsensor_env_gpu
) on the Great Lakes HPC system.GreedyAlgorithm
class.AcquisitionFunction
andStddev
, which assign utility values to query locations.GreedyAlgorithm
and calling it with aTask
andAcquisitionFunction
.Deliverables: