The process of developing AI-models is iterative.
The Training process uses a time period in the data which is defined.
The Verification period used cannot include the same period as the Training process.
Once the Model is verified and approved, it can be deployed into Production.
AI-models (Elliot)
The steps below illustrates the use-case for generating a AI-model for Elliot, but must be managed by us as a service.
Deep Reinforcement Learning
The Gym framework can be used for generic generation of AI-models in a clustered Cloud solution using Deep Reinforcement Learning.
Learning process
The following methods are supported:
- Proximal Policy Optimization (PPO)
- Deep Q-Networks (DQN)
Decide on Neural Network settings.
Add the instruments you would like to train on.
Define training period.
Add the feature parameters you would like to add including normalization etc using functions from the Calculator API.
Generate Gym file (contains all training data).
Setup training with Gym file and configuration.
Start up the training cluster in the Cloud using Kubernetes.
View training using Tensorboard.
When the training has completed, deploy the new model to production environment.
Configure the strategy to use the new model.
Run simulation batches to verify result of new model.
