离散、多元与经验采样
Bernoulli
sr_sample_bernoulli(p, k)
签名:sr_sample_bernoulli(p DOUBLE, k BIGINT) -> LIST(DOUBLE)
每个样本都是 0 或 1:
SELECT sr_sample_bernoulli(0.5, 5)
Binomial(二项)
sr_sample_binomial(p, n, k)
签名:sr_sample_binomial(p DOUBLE, n UBIGINT, k BIGINT) -> LIST(UBIGINT)
SELECT sr_sample_binomial(0.5, 10, 12)
sr_sample_binomial_algorithm(p, n, algorithm, k)
签名:sr_sample_binomial_algorithm(p DOUBLE, n UBIGINT, algorithm DOUBLE, k BIGINT) -> LIST(UBIGINT)
显式选择算法:algorithm 为 1.0 = automatic,2.0 = inversion,3.0 = rejection。
SELECT sr_sample_binomial_algorithm(0.5, 10, 2.0, 4)
Negative binomial(负二项)
sr_sample_negative_binomial(r, p, k)
签名:sr_sample_negative_binomial(r DOUBLE, p DOUBLE, k BIGINT) -> LIST(UBIGINT)
SELECT sr_sample_negative_binomial(2.0, 0.5, 10)
Poisson(泊松)
sr_sample_poisson(lambda, k)
签名:sr_sample_poisson(lambda DOUBLE, k BIGINT) -> LIST(UBIGINT)
SELECT sr_sample_poisson(3.0, 10)
Geometric(几何)
sr_sample_geometric(p, k)
签名:sr_sample_geometric(p DOUBLE, k BIGINT) -> LIST(UBIGINT)
SELECT sr_sample_geometric(0.5, 10)
Hypergeometric(超几何)
sr_sample_hypergeometric(population, successes, draws, k)
签名:sr_sample_hypergeometric(population UBIGINT, successes UBIGINT, draws UBIGINT, k BIGINT) -> LIST(UBIGINT)
SELECT sr_sample_hypergeometric(10, 5, 4, 8)
Categorical(类别)
sr_sample_categorical(probabilities, k)
签名:sr_sample_categorical(probabilities DOUBLE[], k BIGINT) -> LIST(UBIGINT)
每个样本是 0 .. len(probabilities) - 1 内的一个类别下标。
SELECT sr_sample_categorical([1.0, 2.0, 1.0], 5)
Discrete uniform(离散均匀)
sr_sample_discrete_uniform(min, max, k)
签名:sr_sample_discrete_uniform(min BIGINT, max BIGINT, k BIGINT) -> LIST(BIGINT)
SELECT sr_sample_discrete_uniform(1, 6, 10)
Multivariate normal(多元正态)
sr_sample_multivariate_normal(mean, covariance, k)
签名:sr_sample_multivariate_normal(mean DOUBLE[], covariance DOUBLE[], k BIGINT) -> LIST(LIST(DOUBLE))
返回 k 个点的列表,每个点的长度与 mean 一致。
SELECT sr_sample_multivariate_normal([0.0, 0.0], [1.0, 0.0, 0.0, 1.0], 5)
Multivariate Student's t(多元 t)
sr_sample_multivariate_students_t(location, scale, freedom, k)
签名:sr_sample_multivariate_students_t(location DOUBLE[], scale DOUBLE[], freedom DOUBLE, k BIGINT) -> LIST(LIST(DOUBLE))
返回 k 个点的列表,每个点的长度与 location 一致。scale 是行主序展平的尺度矩阵
(len(location)² 个元素),长度不符报查询错误。
SELECT sr_sample_multivariate_students_t([0.0, 0.0], [1.0, 0.0, 0.0, 1.0], 3.0, 6)
Dirichlet
sr_sample_dirichlet(alpha, k)
签名:sr_sample_dirichlet(alpha DOUBLE[], k BIGINT) -> LIST(LIST(DOUBLE))
返回单纯形上的 k 个点:每次抽样都是与 alpha 等长、分量和为 1 的向量。
SELECT sr_sample_dirichlet([1.0, 2.0], 8)
Multinomial(多项式)
sr_sample_multinomial(probs, trials, k)
签名:sr_sample_multinomial(probs DOUBLE[], trials UBIGINT, k BIGINT) -> LIST(LIST(UBIGINT))
返回 k 个计数向量,长度与 probs 一致;每个计数向量的分量和恒等于 trials。
SELECT sr_sample_multinomial([0.3, 0.7], 10, 8)
Empirical(经验分布,聚合)
sr_sample_empirical(v, k)
签名:sr_sample_empirical(v DOUBLE[], k DOUBLE) -> LIST(DOUBLE)
唯一的一个聚合采样器:它把整列 v 收集起来,从它的经验分布里重抽 k 个点。
SELECT sr_sample_empirical(v, 6)
FROM (VALUES (1.0), (2.0), (3.0)) t(v)
错误
未知的 algorithm code:
SELECT sr_sample_binomial_algorithm(0.5, 10, 9.0, 4)
-- error: sr_sample_binomial_algorithm: the algorithm must be 1 (automatic), 2 (inversion) or 3 (rejection), got 9