AI Model Changelog

Our AI Model is always growing and changing to improve performance and staying up to date. To learn more about our AI Model, check out this page, or if you are looking to find out changes across the boberdoo system, we have our regular changelog.

Version 1.5 (September, 2022)

Optimal win rate recalculation in real time (ping by ping)
Optimized process of code maintenance

Version 1.4 (March, 2022)

Improved model reaction time - changing trends in optimal win rates and bids are channeled back to the model every 10 minutes instead of hourly
Optimized code structure for faster updates and client setup

Version 1.3 (October, 2021)

the Model Aggressiveness Rate can now be set on the filter level
We use data imputation to deal with inactive periods of the filter - either due to status change, limit reach or budget issue. The average accuracy, symmetric mean absolute percentage error (SMAPE, relative error) and correlation between actual and predicted ping volume have improved by 35-40%
We also improved the weekend and Monday predictions
We now use Redis.io for optimized logging. This lowers our costs and need for expensive resources.
We improved the data preparation process by per-aggregating the data through AWS. This results in faster reaction time to intraday changes in the market.
New model, script and function parameters are used for faster client setup

Version 1.2 (May, 2021)

Added new feature: Model Aggressiveness Rate

Version 1.1 (March, 2021)

Improved model reaction time - changing trends in buying rate are channeled back to the model hourly instead of daily
Added new ping logs for better predictions