Discuss Databricks Databricks Machine Learning Associate Exam Topic 3 Question 23

A data scientist is performing hyperparameter tuning using an iterative optimization algorithm. Each evaluation of unique hyperparameter values is being trained on a single compute node. They are performing eight total evaluations across eight total compute nodes. While the accuracy of the model does vary over the eight evaluations, they notice there is no trend of improvement in the accuracy. The data scientist believes this is due to the parallelization of the tuning process.

Which change could the data scientist make to improve their model accuracy over the course of their tuning process?

AChange the number of compute nodes to be half or less than half of the number of evaluations.

BChange the number of compute nodes and the number of evaluations to be much larger but equal.

CChange the iterative optimization algorithm used to facilitate the tuning process.

DChange the number of compute nodes to be double or more than double the number of evaluations.

Suggested Answer: C

For large datasets, Spark ML uses iterative optimization methods to distribute the training of a linear regression model. Specifically, Spark MLlib employs techniques like Stochastic Gradient Descent (SGD) and Limited-memory Broyden--Fletcher--Goldfarb--Shanno (L-BFGS) optimization to iteratively update the model parameters. These methods are well-suited for distributed computing environments because they can handle large-scale data efficiently by processing mini-batches of data and updating the model incrementally.

Databricks documentation on linear regression: Linear Regression in Spark ML

Get Premium Databricks Machine Learning Associate Questions as Interactive Web-Based Practice Test or PDF Real Exam Format Detailed Explanations Instant Access Get Your Practice Test

34 comments · newest first

Tien 10 months ago

I think they should stick with fewer nodes for better accuracy.

Osvaldo 11 months ago

Changing the number of nodes could help, but not sure about doubling them.

Karima 11 months ago

But wouldn't more compute nodes just complicate things?

Paris 11 months ago

I agree, option C seems like the best choice!

Taryn 11 months ago

Sounds like they need a better algorithm for tuning.

Starr 12 months ago

I feel like option D might just complicate things further. More compute nodes don’t always mean better results, right?

Diego 12 months ago

This question reminds me of a practice problem where we had to balance compute resources. I wonder if option B could work if we scale everything up together?

Edelmira 12 months ago

I’m not entirely sure, but I think changing the optimization algorithm, like in option C, might help if the current one isn’t performing well.

Chun 1 year ago

I remember reading that sometimes using too many compute nodes can lead to less effective tuning, so maybe option A could be a good choice?

Adolph 1 year ago

I'm not totally sure about this one. The question mentions that the data scientist believes the lack of improvement is due to the parallelization, so I'm wondering if changing the optimization algorithm might be a better approach than just adjusting the number of nodes.

Beckie 1 year ago

Okay, I think I've got it. The question is asking us to identify a change that could improve the model accuracy over the course of the tuning process. Based on the information provided, option D seems like the best choice - changing the number of compute nodes to be more than double the number of evaluations.

Myong 1 year ago

Hmm, I think the key here is that the accuracy isn't improving over the eight evaluations, even though they're using eight compute nodes. Maybe changing the number of nodes could help with that?

Amber 1 year ago

I'm a bit confused by this question. It seems like the issue is with the parallelization of the tuning process, but I'm not sure which change would be the best solution.

Cherry 1 year ago

Option D, for sure! Gotta go big or go home, right? Double the nodes, double the fun!

Whitney 1 year ago

↳ replying to Cherry

Yeah, it's worth a try. Let's go big and see if it pays off in the accuracy of the model.

Candra 1 year ago

↳ replying to Cherry

I agree, doubling the number of compute nodes could make a difference in improving accuracy.

Virgie 1 year ago

↳ replying to Cherry

I think option D is the way to go. More compute nodes can definitely help.

Danica 1 year ago

I'd go with option A. Fewer compute nodes than evaluations might introduce some serial processing, but it could help the data scientist identify a more consistent trend in the results.

Ronny 1 year ago

↳ replying to Danica

It seems like a logical approach to improve the model accuracy during the tuning process.

Ciara 1 year ago

↳ replying to Danica

Changing the number of compute nodes to be half or less than half of the evaluations could be beneficial.

Felicidad 1 year ago

↳ replying to Danica

I agree, having fewer compute nodes might make it easier to track the trend in accuracy.

Valentin 1 year ago

↳ replying to Danica

I think option A is a good choice. It could help with consistency in the results.

Herman 1 year ago

Hmm, I'm not sure. The question mentions no trend of improvement, so maybe option B could work to explore a larger hyperparameter space. Worth a shot!

Lilli 1 year ago

I think changing the iterative optimization algorithm used could also help improve the model accuracy.

Maia 1 year ago

I disagree, I believe they should change the number of compute nodes to be double or more than double the number of evaluations.

Avery 2 years ago

Option D seems like the way to go. Doubling the number of compute nodes should allow for more parallel evaluations and potentially improve the model accuracy.

Gregoria 1 year ago

↳ replying to Avery

But wouldn't changing the iterative optimization algorithm also make a difference in improving the model accuracy?

Leatha 1 year ago

↳ replying to Avery

I agree, increasing the number of compute nodes could help with parallel evaluations.

Nydia 2 years ago

I think option C is the best choice. Changing the optimization algorithm could help the data scientist explore the hyperparameter space more effectively and potentially find a better model.

Denae 1 year ago

↳ replying to Nydia

I agree, trying a different optimization algorithm could make a big difference.

Felice 1 year ago

↳ replying to Nydia

That's a good point. It might lead to finding a better model.

Marge 1 year ago

↳ replying to Nydia

Changing the optimization algorithm could help explore the hyperparameter space more effectively.

Emogene 1 year ago

↳ replying to Nydia

I think option C is the best choice.

Niesha 2 years ago

I think the data scientist should change the number of compute nodes to be half or less than half of the number of evaluations.

Databricks Machine Learning Associate Exam - Topic 3 Question 23 Discussion