np.random

import numpy as np

# 1) Create — pass a seed for reproducibility
rng = np.random.default_rng(seed=42)

# 2) Sample distributions
rng.integers(0, 100, size=5)                       # uniform int [0, 100)
rng.random(size=(3, 4))                            # uniform float [0, 1)
rng.normal(loc=0, scale=1, size=(2, 3))            # normal
rng.standard_normal(size=10)                       # N(0, 1) shortcut
rng.binomial(n=10, p=0.3, size=100)
rng.poisson(lam=4.0, size=1000)
rng.exponential(scale=1.0, size=100)

# 3) Choice — sample with / without replacement
rng.choice(['red', 'green', 'blue'], size=10)
rng.choice(10, size=3, replace=False)              # 3 distinct picks 0..9
rng.choice(10, size=5, p=[0.5, 0.1, 0.1, 0.05, 0.05, 0.05, 0.05, 0.05, 0.025, 0.025])

# 4) Shuffle / permutation
arr = np.arange(10)
rng.shuffle(arr)                                   # in place
rng.permutation(10)                                # NEW array
rng.permutation(arr)

# 5) Reproducibility — seeding
rng1 = np.random.default_rng(0)
rng2 = np.random.default_rng(0)
assert (rng1.random(5) == rng2.random(5)).all()    # identical

# 6) Splitting — independent streams for parallel work
from numpy.random import SeedSequence
seeds = SeedSequence(42).spawn(4)
rngs = [np.random.default_rng(s) for s in seeds]   # 4 independent generators

# 7) Pandas — sample rows
import pandas as pd
df = pd.DataFrame({'x': range(100)})
df.sample(n=10, random_state=42)
df.sample(frac=0.1, random_state=42)

# 8) Train/test split — sklearn (which uses RandomState under the hood)
from sklearn.model_selection import train_test_split
Xtr, Xte = train_test_split(df, test_size=0.2, random_state=42)

# 9) Legacy API — avoid in new code
np.random.seed(42)
np.random.rand(3)         # global state; tests / parallel code → bugs

import numpy as np
rng = np.random.default_rng(seed=42)
print(rng.integers(0, 10, size=5))
print(rng.normal(loc=0, scale=1, size=(2, 3)))