Applying Reinforcement Learning Pipelines to Time Series datasets

June 24, 2022

CHICAGO – June 22, 2022 – NorthGravity, a leading Data Automation/AI software, today announced the release of an Reinforcement Learning model within the NorthGravity Platform (NGP).  By integrating Reinforcement Learning, clients are able to test and integrate deep learning alongside the currently established 100+ core task for data collection, analytics, and distribution within the NGP. 

"We're excited about the release of Reinforcement Learning capabilities in the NorthGravity Platform," said Travis Nadelhoffer, CEO of NorthGravity. "Reinforcement Learning allows users to apply deep learning and customize what success looks like, ultimately, achieving a more focused result."

Travis Nadelhoffer - CEO of NorthGravity
What is Reinforcement Learning?

Reinforcement Learning is a deep learning training method based on rewarding desired behaviors and/or punishing undesired ones. In general, a reinforcement learning agent is able to perceive and interpret its environment, take actions and learn through trial and error.

Reinforcement learning has many applications including NLP, Autonomous Cars, Image Processing, Business Management, Education, Energy, Finance and many more.    In the context of forecasting time series with in finance, reinforcement learning is achieved by having an agent trade an asset for a specified period of time within the environment (which in this case is a realistic trading simulator), and then check what kind of trade logic led to the biggest profit, which is the reward function in this case.

North Gravity Reinforcement Learning Pipeline
How is it different from traditional Machine Learning methods?

Machine Learning based trading signals are based on two types of algorithms: classifiers and regressors. Classifiers put current market situations into bullish or bearish categories, outputting Long and Short signals respectively, while regressors estimate price of an asset for a specified time in the future. Classifiers are optimized to maximize the number of times they are right, and regressors try to get as close as possible to the actual future price with its predictions.  

For reinforcement learning, the “reward” is customizable - it can be programmed to maximize profit, Sharpe ratio, information ratio and others. Therefore, we may see that Reinforcement Learning produces lower accuracy,  but the winning trades may outperform the losing ones, resulting in greater potential rewards. Customization of reward functions is definitely one of the biggest upsides to applying reinforcement learning.

What will production Reinforcement Learning look like within NorthGravity ?
  1. Pick your input dataset
  2. Select the target 
  3. Run the Optimization
  4. Receive the trained model which can be used in forecast mode.
  5. Schedule or trigger forecasting pipeline
  6. Receive the results on the desired schedule  

Within the NorthGravity Platform this can be setup with no code or can be configured with code based on a user's desire.   

Why should you care?

Successful Machine Learning adoption has a proven positive impact on revenue and a clear cost reduction fueling companies to lap their competitors.

About NorthGravity

The NorthGravity Platform (NGP) helps companies utilize Data Flow Automation and  Machine Learning at scale creating a competitive advantage for their business.   The NorthGravity Platform includes a cloud data warehouse, extract load transform framework,  dataflow automation, self-service ML insights, and global data governance.  NGP fuels data-driven decisions across organizations, and the adoption of machine learning and data. The NGP includes a cloud data warehouse, ELT framework, dataflow automation, AutoML, self-service ML insights and global data governance.

Contact us and find out more about NorthGravity and reinforcement learning solutions for your company

Grow your business
Reduce the time to predictive results and increase value at lower cost and risk.