##### Scenario Hyper-parameter optimisation service must enumerate every grid-search combination. ##### Question Write a generator that yields all combinations of values from a dict such as {'learning_rate':[0.1,0.2], 'feature':['A','B'], 'batch':[10,20]}. What alternatives exist to reduce explosion when grids are huge? ##### Hints Use recursive backtracking or itertools.product; discuss random search, Bayesian optimisation, Sobol sampling.

Evaluates Python grid-search generation and hyperparameter optimization strategy. Strong answers yield lazy Cartesian-product combinations and discuss random search, Bayesian optimization, pruning, early stopping, and budget limits.

How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

What difficulty level is this interview question?

This is a medium difficulty Machine Learning question, commonly asked during Onsite rounds at Pinterest.

What role is this question designed for?

This question is commonly asked for Data Scientist candidates at Pinterest during technical interviews.

Optimize Hyper-parameter Search to Prevent Combinatorial Explosion

Enumerating Grid Search and Avoiding Hyperparameter Explosion

You are building a hyperparameter optimization service that must enumerate every grid-search combination. The input is a Python dict mapping parameter names to candidate values, for example {'learning_rate': [0.1, 0.2], 'feature': ['A', 'B'], 'batch': [10, 20]}.

Constraints & Assumptions

The enumerator should be lazy and memory efficient.
It should support any number of parameters.
The output for each combination should be a dict mapping parameter name to chosen value.
Discuss alternatives when the full Cartesian product is too large.

Clarifying Questions to Ask

Should the generator preserve input key order?
How should empty grids or empty candidate lists be handled?
Are candidate values hashable or serializable?
Is exhaustive grid search required, or can the service sample candidates?

Part 1 - Python Generator

Write a Python generator that lazily yields all combinations as dictionaries.

What This Part Should Cover

Use itertools.product or recursive backtracking.
Avoid materializing all combinations in memory.
Yield one dict per combination.
Explain time complexity as the product of candidate-list lengths and space complexity per yielded item.

Part 2 - Avoiding Combinatorial Explosion

Discuss practical alternatives for huge grids.

What This Part Should Cover

Include random search, Bayesian optimization, Hyperband, successive halving, Sobol or Latin hypercube sampling, adaptive search, and coarse-to-fine grids.
Use early stopping, pruning, parallelism, warm starts, and budget constraints.
Prioritize high-impact hyperparameters and narrow ranges using prior knowledge.
Track experiment metadata for reproducibility.

Follow-up Questions

How would you count combinations without enumerating them?
What should happen if one parameter has an empty candidate list?
When is random search better than grid search?

Enumerating Grid Search and Avoiding Hyperparameter Explosion

Constraints & Assumptions

The enumerator should be lazy and memory efficient.

It should support any number of parameters.

The output for each combination should be a dict mapping parameter name to chosen value.

Discuss alternatives when the full Cartesian product is too large.

Clarifying Questions to Ask

Should the generator preserve input key order?

How should empty grids or empty candidate lists be handled?

Are candidate values hashable or serializable?

Is exhaustive grid search required, or can the service sample candidates?

Part 1 - Python Generator

Write a Python generator that lazily yields all combinations as dictionaries.

What This Part Should Cover

Use itertools.product or recursive backtracking.

Avoid materializing all combinations in memory.

Yield one dict per combination.

Explain time complexity as the product of candidate-list lengths and space complexity per yielded item.

Part 2 - Avoiding Combinatorial Explosion

Discuss practical alternatives for huge grids.

What This Part Should Cover

Include random search, Bayesian optimization, Hyperband, successive halving, Sobol or Latin hypercube sampling, adaptive search, and coarse-to-fine grids.

Use early stopping, pruning, parallelism, warm starts, and budget constraints.

Prioritize high-impact hyperparameters and narrow ranges using prior knowledge.

Track experiment metadata for reproducibility.

Follow-up Questions

How would you count combinations without enumerating them?

What should happen if one parameter has an empty candidate list?

When is random search better than grid search?

Optimize Hyper-parameter Search to Prevent Combinatorial Explosion

Quick Overview

Optimize Hyper-parameter Search to Prevent Combinatorial Explosion

Enumerating Grid Search and Avoiding Hyperparameter Explosion

Constraints & Assumptions

Clarifying Questions to Ask

Part 1 - Python Generator

What This Part Should Cover

Part 2 - Avoiding Combinatorial Explosion

What This Part Should Cover

Follow-up Questions

Write your answer

Optimize Hyper-parameter Search to Prevent Combinatorial Explosion

Quick Overview

Optimize Hyper-parameter Search to Prevent Combinatorial Explosion

Enumerating Grid Search and Avoiding Hyperparameter Explosion

Constraints & Assumptions

Clarifying Questions to Ask

Part 1 - Python Generator

What This Part Should Cover

Part 2 - Avoiding Combinatorial Explosion

What This Part Should Cover

Follow-up Questions

Write your answer