Katkat
Back

Implement GLM Training and Prediction from an Explicit Family and Link

Quantbot TechnologiesData Scientist · Online Assessment
HardData ScientistOnline AssessmentMachine Learning

Last updated

Explain a from-scratch generalized linear model through family and link selection, consistent loss and gradients, numerical stability, and prediction semantics.

Explain how you would implement training and prediction for a generalized linear model (GLM) from scratch. Define the model, objective, optimization procedure, and prediction behavior for one explicitly chosen illustrative family and link function.

Constraints

The response distribution, link, input schema, regularization, and numerical tolerances are not specified. Clarify them before claiming a unique executable implementation. Use your chosen GLM as an example rather than implying that the original task required logistic regression, linear regression, or another particular family.

Clarifying Questions

  • What is the target's domain, and which distribution and link are required?
  • Is an intercept included, and is regularization part of the objective?
  • Are predictions expected as conditional means, probabilities, or thresholded labels?
  • What convergence, initialization, and invalid-input behavior should be supported?

Hints

Hint 1

Connect the three GLM components: Identify the response family, the linear predictor, and the link between that predictor and the conditional mean before deriving an update.

Follow-up questions

  • How would changing the response family change the loss and allowable predictions?
  • What behavior would you expect from unregularized logistic regression on separable data?