Exam DP-100 topic 3 question 80 discussion

Answer is correct as Grid Sampling doesn't support continuous hyperparameter configuration

Correct!

B. Grid sampling D. Random sampling Grid Sampling does not support continuous value but we can define a finely spaced range for learning rates and use that as hyperparameter learning_rate_values = np.linspace(0.001, 0.1, num=10).tolist()

On exam 18 Aug 2023

https://learn.microsoft.com/en-us/azure/machine-learning/v1/how-to-tune-hyperparameters-v1

BD. Both Grid sampling and Random sampling can be used for configuring the sampling method for the Hyperdrive experiment. Grid sampling will allow you to define a search space with different combinations of hyperparameters, and it will train the model with every possible combination of hyperparameters. This approach works well when the search space is relatively small. Random sampling will randomly sample hyperparameters from the defined search space. This method can cover a larger search space more efficiently than grid sampling, and it can be useful when there are a large number of hyperparameters to tune. Bayesian sampling is not an option in this case, as it requires an early termination policy to be applied in order to determine which hyperparameters to explore next based on the results of previous runs. Since an early termination policy is not allowed in this scenario, Bayesian sampling is not a viable option

CD is correct

The answer is correct. Grid sampling only supports discrete hyperparameters. Refer to this link in Microsoft documentation: https://docs.microsoft.com/en-us/azure/machine-learning/how-to-tune-hyperparameters#grid-sampling

Bayesian and Default don't permit early termination. Random and Grid support early termination

Correct answers

In my opinion, the answer is A C. we need to select the sampling which is not supporting early termination. Random and Grid both are supporting the early termination

on 19Oct2021

On Exam 01 Oct 2021

Bayesian Sampling - https://docs.microsoft.com/en-us/python/api/azureml-train-core/azureml.train.hyperdrive.bayesianparametersampling?view=azure-ml-py#:~:text=Bayesian%20sampling%20does%20not%20support,leave%20off%20the%20early_termination_policy%20parameter. Random Sampling - https://docs.microsoft.com/en-us/azure/machine-learning/how-to-tune-hyperparameters#:~:text=Random%20sampling%20supports%20discrete%20and,termination%20of%20low%2Dperformance%20runs. Grid Sampling - https://docs.microsoft.com/en-us/azure/machine-learning/how-to-tune-hyperparameters#:~:text=Grid%20sampling,-Grid%20sampling%20supports&text=Supports%20early%20termination%20of%20low%2Dperformance%20runs.,-Grid%20sampling%20does Random and Grid Sampling supports early termination policy for low-performance runs in the documentations. A and C maybe?

the answer is B, D

"Random sampling supports discrete and continuous hyperparameters. It supports early termination of low-performance runs."

It's correct.