Sampling

Enables selecting a representative subset of data to reduce processing load while preserving statistical validity.

Data analysis: extracting a 10% random subset of a large dataset for quick prototyping Machine learning: creating a training set from a massive log file to fit within memory constraints A/B testing: sampling user traffic to run experiments without affecting all users

Data pipelines, statistical analysis scripts, machine learning model training, big data frameworks such as Spark or Pandas.

define population → choose sampling_rate → apply random_selection → obtain sample

sample = population * sampling_rate

• random sampling
• stratified sampling
• systematic sampling
• reservoir sampling
• bootstrap

• simple random sampling → easy to implement but may miss rare categories stratified sampling → preserves class distribution at the cost of added complexity systematic sampling → fast but can introduce periodic bias if data has hidden cycles

Assuming a small random sample automatically reflects rare classes → leads to biased models Using systematic sampling on sorted data without shuffling → introduces periodic bias Neglecting to set a random seed for reproducibility → results cannot be replicated

sampling strategy vs sampling algorithm: the former is the high-level plan, the latter is the concrete implementation sampling vs data sharding: sampling selects a subset, sharding partitions the whole dataset sampling vs downsampling: downsampling reduces resolution of data, sampling selects representative items

Coming from SQL: using LIMIT without ORDER BY may return non-representative rows → ensure deterministic ordering before limiting Coming from Python: using random.choice on a list repeatedly can bias selection if the list changes during iteration → copy the list first

• random sampling
• stratified sampling
• systematic sampling
• reservoir sampling
• bootstrap

Do not use a simple random sample when rare event detection is critical; consider stratified or oversampling techniques Sampling reduces memory and CPU usage proportionally to the sample size, but improper sampling can increase variance of estimates When the population size is unknown or streaming, reservoir sampling is required to maintain uniform probability

Correctly applied sampling strategies allow teams to prototype faster, train models on feasible data sizes, and run experiments with controlled exposure, ultimately reducing costs and time-to-market.

Think of a chef tasting a spoonful of soup to gauge the whole pot – a small sample informs the whole.

Always fix a random seed when reproducibility is required, especially in scientific experiments or CI pipelines.

Advanced sampling techniques such as importance sampling or stratified reservoir sampling