Introduction to GeneralizationGeneralization is the act of calculating the generalization of the overall predictions which allows you to accurately predict the correct results over the span of multiple samples rather than just one individual sample. While you can preform generalization calculations on your own program natively, CoffeeHouse provides the developers with the ability make this process easier by storing the generalization data on our servers and having CoffeeHouse use a pointer-like mechanism to calculate the generalization without any additional performance loss or overhead.
This works by storing all the predictions on a table then by adding up all the predictions and dividing the final results by the total size of the prediction table. So while there may be one or two false predictions, the overall prediction is accounted for by the dominate label which in most cases is the true and correct prediction. This is great for determining the final prediction of a set of a samples or to accurately determine the detected language of a chat conversation. While not all CoffeeHouse features supports generalization, the documentation will indicate when generalization is supported and what labels to expect. This part of the documentation will explain how to initialize generalizations and understand the results.