MLA-C01 Data Preparation for Machine Learning Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Configure the imputation transform in the Data Wrangler flow so it learns the fill statistic (such as the mean) from the training data during the flow run.

  Why this is correct

  Learning the fill statistic from the training data ensures the imputation parameters are derived from the correct distribution and not from inference data, which would leak information. Data Wrangler captures these learned parameters as part of the flow, making the transformation reproducible. This is the foundation for applying the same imputation consistently at inference time.

• ✗

  Replace missing numeric values with zero in both the training and inference datasets using a custom Python script.

  Why it's wrong here

  Filling with a constant zero is a valid but arbitrary choice that can distort distributions and does not learn from the training data. A custom script run separately on training and inference risks divergence if the logic is edited between runs. It also bypasses Data Wrangler's captured parameters, undermining reproducibility.

• ✓

  Export the Data Wrangler flow to a SageMaker Pipeline and include the generated processing step so it runs on both training and inference inputs.

  Why this is correct

  Exporting the flow into a SageMaker Pipeline turns the transforms, including the learned imputation parameters, into a processing step that can be reused. Running that same step on inference data applies the identical fill statistics and transformation logic. This guarantees consistency between training and inference rather than relying on manual re-implementation.

• ✗

  Compute the mean of each numeric column on the full dataset including inference data, then hard-code those values into the pipeline.

  Why it's wrong here

  Including inference data when computing imputation statistics causes data leakage and makes the pipeline dependent on the inference batch, which is not available at training time. Hard-coding values breaks reproducibility if the training data changes. This approach also cannot scale to a stream of new inference records.

• ✗

  Delete all rows that contain missing values from the training dataset before training.

  Why it's wrong here

  Dropping rows with missing values discards potentially useful data and can bias the model if missingness is non-random. It also does not produce any reusable imputation logic to apply at inference time, so an inference record with a missing value would still need handling. This action reduces the dataset without ensuring consistency.

Go deeper

Related to this question