Random sampling with SciPy

Use dist.rvs(size=..., random_state=...) for reproducible simulation. SciPy distributions complement np.random with named families used in tests and Monte Carlo workflows.

Reproducibility

• Pass random_state integer for repeatable labs
• Seed NumPy with np.random.default_rng(seed) when mixing APIs
• Document seed in notebooks and tests

Simulation patterns

• Draw samples → compute statistic → compare to theory
• Bootstrap resampling (manual loops on indices)
• Power analysis prototypes before real experiments

Examples

import numpy as np
from scipy import stats

rng = np.random.default_rng(42)
normal = stats.norm.rvs(loc=0, scale=1, size=5, random_state=rng)
uniform = stats.uniform.rvs(0, 1, size=5, random_state=42)
print('normal:', normal)
print('uniform:', uniform)

Important interview questions and answers

1. Q: rvs vs np.random.normal?
   A: Both work; stats.rvs ties sampling to named distribution objects used in inference.
2. Q: Why random_state?
   A: Reproducible lessons, tests, and debugging—same seed → same draws.

Self-check

1. How do you draw 10 samples from Normal(0,1) reproducibly?
2. Name one simulation use case for rvs.

Tip: Pass random_state in lessons and tests so playground output matches MCQ explanations.