Summary

In this chapter, we discussed using the keyed prediction pattern when serving models. This is an important pattern when our serving environment is making predictions on a large scale asynchronously. In this case, we risk losing the order of the predictions and we will not be able to map the inputs to the corresponding predictions. To solve this problem, we should tag the training data with keys and get predictions with the same keys tagged. In this way, we can reproduce the exact mapping of input to output.

We also discussed situations in which keyed prediction is needed and looked at some examples. Then, we discussed the techniques of keyed prediction. We concluded with some ideas for the creation of keys during keyed prediction.

In the next chapter, we will start talking about a different group of patterns for serving models. These serving patterns fall into the second category of patterns as discussed in Chapter 2. These patterns deal with the exact serving mechanism...