Costs Associated with Using AI Factor

AI Factor is an advanced, premium tool which has some costs in addition to your Portfolio123 subscription. This article explains those costs.

Training Time Costs

Training AI models and predictors require substantial compute resources. The training of an individual model could take only a few seconds or it could run for more than a day depending on the combination of the algorithm selected and it's hyperparameter settings, universe size, number of features, number of dates in the training period, frequency of the training data and the number of folds in the validation model.

When training models or predictors, you choose the "worker" on which your training will run. There are various workers to choose from, each with its own cost per hour of training time. More powerful workers cost more per hour, but the training will complete sooner. Additionally, certain models require a worker with significant memory, otherwise the training may fail. These models have the #highmem tag.

When you enable AI Factor in your account, you will receive a $50 credit to use for worker training time. As you train models or predictors, the training time costs are deducted from your available credits. For example, if you train a set of models and select a worker with a cost of $.75 per hour, and the total combined training time for those models is 10 hours, $7.50 will be deducted from your AI credits. If your AI credit balance falls below $0, then during the nightly billing process, you will be charged an amount which restores your credit balance to $50.

Resource Units

Each subscription level is allocated a certain number of "Resource Units" (RUs). When you save any work on Portfolio123, a small number of RUs is deducted from your balance. Loading datasets, training models, and saving prediction data all use Resource Units. If these items are later deleted, the RUs are returned to your available RU balance. There is no monetary cost associated with RUs unless you exceed the allocation for your account. If you do exceed it, you can either purchase additional RUs or free up space by deleting old work.

Prediction Add-On

Up until this point, you have been developing and testing your models, with the only cost incurred being the training costs. Once you are ready to use the AIFactor function in your live strategies, screens and ranking systems, you will need to subscribe to the Prediction function add-on. You will see this add-on and its cost displayed on the AI Factor page in Account Settings. This add-on allows you to use the AIFactor() function, which returns the current predictions from your model(s).