Scaling and Tuning MACHINE-LEARNING-PROFESSIONAL Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   In Optuna, what is the role of a 'study' object?

   • AIt is a specific algorithm (e.g., TPE, Random) used to suggest new hyperparameters.
   • BIt manages the entire optimization process, including the search space, the trials, and the results.
   • CIt is a container for the training dataset used in the objective function.
   • DIt defines a single hyperparameter configuration to be evaluated.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A team needs to perform inference using a pre-trained model on a large dataset of 100 million rows. The model is a scikit-learn model that accepts a pandas DataFrame. They want to parallelize the inference across a Spark cluster. Which approach should they use?

   • ACollect the data to the driver and run inference locally.
   • BUse SparkML's model.transform() method.
   • CUse a Pandas UDF (applyInPandas) to apply the model's predict function to each group of the DataFrame.
   • DUse a regular UDF that calls the model's predict function on each row.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   In data parallelism, how is the model updated after each batch of training data is processed?

   • AEach worker independently updates its own copy of the model, and the copies are never synchronized.
   • BEach worker computes gradients on its data shard, and the gradients are aggregated (e.g., averaged) before updating the shared model parameters.
   • CThe driver node trains the model on the entire dataset and sends the final weights to the workers.
   • DThe model is split into layers, and each worker trains a different layer.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A data scientist is using Optuna to run a hyperparameter search. They want to track the results of each trial and log the best model to the MLflow tracking server. What is the recommended way to integrate Optuna with MLflow?

   • AOptuna cannot be integrated with MLflow; you must use a separate tracking tool.
   • BUse the MLflowCallback class provided by Optuna to automatically log metrics and parameters for each trial.
   • CUse mlflow.autolog() which automatically detects Optuna studies and logs all trials.
   • DManually call mlflow.log_param() and mlflow.log_metric() inside the objective function for each trial.
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   A deep learning model is too large to fit into the memory of a single GPU. The team decides to split the layers of the neural network across multiple GPUs. What parallelization strategy is this?

   • AHorizontal scaling
   • BData parallelism
   • CModel parallelism
   • DPipeline parallelism
   Reveal Solution Discussion Select an answer first