Open Climate Fix Improves Solar Forecasting With Weights & Biases

“We’re still a small team working on developing our workflows, and W&B has been a big help in optimizing and automating a lot of those key processes.”
Jacob Bieker
Senior Machine Learning Research Engineer

Renewable Energy Uncertainty

If a snowstorm is coming, we check the weather app to find out just how much snow is headed our way and when it’s going to hit. The same can’t be said for forecasting solar energy. When unexpected clouds roll in or temperatures take a turn, it can cause a significant dip in solar power generation, forcing grid operators to rely on fossil fuel plants to generate excess energy as backup.

Based in London, Open Climate Fix (OCF) is a non-profit research and development lab working on solving that precise problem with hopes of delivering better solar forecasts through the use of machine learning. Tackling such a challenge is part of the organization’s mission of reducing carbon emissions as rapidly as possible. 

As a small team with diverse backgrounds and experience in ML, ensuring their tech stack was flexible and capable of automating a large part of their workflow was a high priority. Designed to manage the end-to-end ML lifecycle, Weights & Biases became the platform of choice with its real-time collaborative workspace, robust system for experiment tracking, and extensive integration with popular ML frameworks and tools—scaling up the team’s collaborative and agile experimentation. 

Connected and Aligned

“The nature of our projects often means engineers are working in silos and running their own experiments,” said Jacob Bieker, Senior Machine Learning Engineer at OCF. “Having these siloed pockets of ML makes it difficult to share knowledge and collaborate effectively across the team.”

OCF wanted to break down those silos and develop better models at greater velocity and scale—and with the climate crisis rapidly intensifying, speed and impact are key. It was essential for the team to find a way to share findings and results in one place and gain visibility into each other’s work, creating a culture of sharing and idea generation while removing the opacity of ML work across the entire team. 

Built for team collaboration, Weights & Biases unifies everything from models and pipelines to experiments and datasets in a single system of record for OCF. The team can easily track all of their experiments with metrics and logs, ensuring reproducibility and transparency. This makes it easy to detect and debug issues, compare the results of different runs, and gauge whether the model they’re building is going in the right direction or not. Leveraging Weights & Biases as a centralized source of truth helps keep the team aligned, establishes a place for communication and provides documentation for the key decisions made.

 
 

Seamless Integrations

Committed to open-source technology in their forecasting tools, OCF firmly believes open code and open data are crucial to addressing the climate challenge. All of the models trained by the team are hosted on HuggingFace. Coupled with Weights & Biases integration, the team can quickly train and monitor models for full traceability and reproducibility without compromising that ease of use. 

“In my work, for each trained model which is on HuggingFace, it’s linked to a wandb run where you can actually see the entire training log,” said James Fulton, Machine Learning Researcher at OCF. “If something happens downstream of a model build, we can easily trace the issue back to the wandb run, saving us time in debugging.”

 

The Future is Bright

As the world we live in grapples with challenges like climate change, rising energy needs, and dwindling reserves of finite fossil fuel resources, solar energy emerges as a promising solution to address these pressing issues. This makes what the OCF team is working on all the more important, as it helps build reliability and accuracy in solar forecasting. 

To accelerate research and train models at scale, OCF needs tools that can optimize collaboration, support the iterative model development process, and complement its existing tech stack. Leveraging Weights & Biases enables the team to fuel innovation with more transparency, standardization, and centralized operations. 

“We’re still a small team working on developing our workflows, and W&B has been a big help in optimizing and automating a lot of those key processes,” said James.