Hacker News Reader

Yeah I should have been more explicit that this comment is in reference to ML code. Training is nothing but a loop so its unlikely you pass type A in iteration 1 and type B in iteration 200. If that happens most likely your training data is messed up and type safety cannot help you as you would have compiled and tested for type A.

View on HN

You could also have a complex training job that trains a shallow model for a while, then uses it to train a deeper model, or extract an embedding from some layer and train a classical prediction model that uses the embedding as the feature vector.

But the point is that the right way to ensure safety is with realistic fixture-based integration testing. That’s not what static typing is for in that type of use case and is not a de facto benefit of static typing.

Reply on news.ycombinator.com