Google Professional Machine Learning Engineer Exam - Topic 10 Question 31 Discussion

Actual exam question for Google's Professional Machine Learning Engineer exam

Question #: 31
Topic #: 10

[All Professional Machine Learning Engineer Questions]

You are building a model to predict daily temperatures. You split the data randomly and then transformed the training and test datasets. Temperature data for model training is uploaded hourly. During testing, your model performed with 97% accuracy; however, after deploying to production, the model's accuracy dropped to 66%. How can you make your production model more accurate?

ANormalize the data for the training, and test datasets as two separate steps.

BSplit the training and test data based on time rather than a random split to avoid leakage

CAdd more data to your test set to ensure that you have a fair distribution and sample for testing

DApply data transformations before splitting, and cross-validate to make sure that the transformations are applied to both the training and test sets.

Show Suggested Answer

Suggested Answer: D

by Jess at May 05, 2022, 11:38 AM

Limited Time Offer

25%

Off

Get Premium Professional Machine Learning Engineer Questions as Interactive Web-Based Practice Test or PDF

Contribute your Thoughts:

Submit Cancel

Madonna

4 months ago

Not sure if adding more test data will really fix the issue.

upvoted 0 times

...

Freeman

4 months ago

Normalizing separately could help, but I think cross-validation is key too.

upvoted 0 times

...

Rene

4 months ago

Wait, how did the accuracy drop so much after deployment?

upvoted 0 times

...

Mariann

4 months ago

I agree, option B seems like the best choice here.

upvoted 0 times

...

Rene

5 months ago

Sounds like a classic case of data leakage!

upvoted 0 times

...

Alecia

5 months ago

I recall a practice question where we applied transformations before splitting the data. Option D seems like a solid approach to ensure consistency across datasets.

upvoted 0 times

...

Ryan

5 months ago

I think adding more data to the test set, as mentioned in option C, could help, but I'm not sure if it addresses the drop in accuracy after deployment.

upvoted 0 times

...

Viola

5 months ago

I'm not entirely sure, but I feel like normalizing the data separately for training and testing could lead to inconsistencies. Maybe option A isn't the best?

upvoted 0 times

...

Jame

5 months ago

I remember we discussed the importance of avoiding data leakage, so I think option B makes sense to prevent that issue.

upvoted 0 times

...

Veronica

5 months ago

Hmm, I'm not sure if table permissions alone will be enough to meet the goal. I'll need to think through the authentication and authorization requirements more carefully.

upvoted 0 times

...

Billye

5 months ago

This one is tricky! I feel like I read about the differences in accreditation somewhere, but I can't remember the details right now.

upvoted 0 times

...

Chandra

5 months ago

I've got this! The answer is clearly C - the ability of one service to consume another service forms the basis of service composition. That makes total sense to me.

upvoted 0 times

...

Penney

5 months ago

I definitely remember something about Datastore Control Center, but I can't recall if it was related to debugging or something else entirely.

upvoted 0 times

...