Model implementation

Implementing the model is, after all, the most important step in our pipeline. In a way, we have built the whole pipeline for this step. Apart from building the network architecture, there are numerous details we need to consider to optimize our implementation (in terms of effort, time, and perhaps code efficiency as well).

In this session, we will discuss profiling and bottleneck tools available in the PyTorch package itself and ignite, a recommended trainer utility for PyTorch. The first part covers bottleneck and profiling utility, which is essential when the model starts underperforming and you need to know what went wrong where. The second part of this session explains ignite, the trainer module.

A trainer network is not really an essential component, but it is a good-to-have helper utility that saves a lot of time writing boilerplate and fixing bugs. Sometimes, it can reduce the number of lines of your program by half, which also helps to improve...