Appearance
Topic Drill 3.2.1: Activities in Data Preparation
About this drill
LO AI-3.2.1 (K2): Explain the activities related to data preparation.
Lesson: Ch 3, section 3.2.1
This drill covers every point in section 3.2 / 3.2.1:
- Why data preparation is crucial and resource-intensive; its link to the data pipeline
- Data acquisition: data types, gathering from sources, labeling and verifying labels
- Data preprocessing: cleaning (defects, duplicates, outliers, imputation, anonymization), transforming, scaling, augmenting (incl. adversarial examples and synthetic data), sampling
- Feature engineering: selecting and extracting features; EDA in parallel
- Iterative and manual nature; operational data matching the training data
Topic Drill: 3.2.1
Q1. A team collects temperature and vibration readings from factory sensors through a REST API. Which data preparation activity is this?
AData acquisition
BFeature engineering
CData preprocessing
DExploratory data analysis
Q2. Before training, customer names and account numbers are removed from the dataset. Which activity is this, and why?
AFeature engineering, to improve accuracy
BData acquisition, to identify data types
CData preprocessing (cleaning), to protect privacy and comply with regulations
DExploratory data analysis, to find anomalies
Q3. An image dataset is enlarged by adding rotated, flipped and slightly darkened copies of existing images. Which activity is this?
ASampling subsets
BAugmenting data (data preprocessing)
CFeature extraction
DLabeling data
Q4. From 80 available columns, a team keeps the 12 that contribute most to model performance, to reduce training time. Which activity is this?
AData acquisition
BData cleaning
CExploratory data analysis
DFeature engineering
Q5. A data analyst plots histograms of every feature and notices an unusual spike of "age = 0" values. Which activity is this?
AExploratory data analysis (EDA)
BData augmentation
CFeature selection
DData sampling
Q6. Why might a team deliberately add adversarial examples to its training data?
ATo reduce training time
BTo label the data automatically
CTo enhance robustness against adversarial attacks, as part of data augmentation
DTo anonymize personal data
Q7. What is the purpose of "sampling subsets" in data preprocessing?
ATo increase the sample size
BTo reduce training times and computational costs
CTo label ambiguous data points
DTo detect bias
Q8. Which statement about operational data and data preparation is CORRECT according to the syllabus?
AOperational data never needs to resemble the training data
BThe preparation steps used in production must be identical to the manual steps used for training
COperational data is used to label the training data
DOperational data should match the characteristics of the training data (e.g., distributions, feature ranges), but the preparation steps may be adjusted for production efficiency and scalability
Q9. Which TWO statements about data preparation are CORRECT?
Select 2 options.
AThe activities are always performed in a fixed order and none may be skipped
BPreparing training data is typically iterative and often manual
CEDA is performed only after the model has been trained
DLabeling data is part of feature engineering
EData preparation typically takes a significantly larger share of the effort than model selection and building
Q10. Labeling data for a supervised learning task and verifying the labels' accuracy and consistency belongs to which data preparation activity?
AFeature engineering
BData preprocessing
CData acquisition
DModel evaluation
0 / 10 answered
← 3.1.4 Pretrained Models, Fine-Tuning, RAG · Next drill: 3.2.3 Datasets →