Model embedded in Application - Architecture

Data Science, AI and ML
1

A typical flow of this architecture is to extract a pre-trained model (DE serialize) as a pickle file and embed it into the application framework such as Django.

Advantages:

  1. Easy architecture
  2. Predictions on the fly

Disadvantages:

  1. Software processes, and deployment process has to be carefully extracted. For e.g. extracting pickle file on 64 bit and using it on 32 bit might have some issues.
  2. Model re-training and deployment will require an app update and a new release. With dynamically changing trends in datasets, this can become a problem.