Getting Started¶

Installation¶

pip install bayesprop

Or with uv:

uv pip install bayesprop

For development (from source):

git clone https://github.com/AVoss84/bayesProp.git
cd bayesprop
uv venv --python 3.13
uv sync
source .venv/bin/activate

Your first A/B test¶

Non-paired design (independent groups)¶

Use this when group A and group B consist of independent observations (i.e. different items or subjects in each group). Input arrays can be binary (0/1) or real-valued on (0, 1) — continuous scores are automatically binarized at a configurable threshold.

from bayesprop.resources.bayes_nonpaired import NonPairedBayesPropTest
from bayesprop.utils.utils import simulate_nonpaired_scores

# Simulate binary outcomes
sim = simulate_nonpaired_scores(N=100, theta_A=0.85, theta_B=0.70, seed=42)
y_A, y_B = sim.y_A, sim.y_B

# Fit the model
model = NonPairedBayesPropTest(seed=42, n_samples=20_000).fit(y_A, y_B)

# Print summary
model.print_summary()

# Unified decision (Bayes factor + P(H₀) + ROPE)
d = model.decide()
print(f"BF₁₀ = {d.bayes_factor.BF_10:.2f}  →  {d.bayes_factor.decision}")
print(f"ROPE: {d.rope.decision}  ({d.rope.pct_in_rope:.1%} in ROPE)")

# Visualise
model.plot_posteriors()
model.plot_savage_dickey()

Paired design (same items, two conditions)¶

Use this when both conditions (e.g. treatment A and treatment B, or version A and version B) are evaluated on the same items or subjects.

from bayesprop.resources.bayes_paired_laplace import PairedBayesPropTest
from bayesprop.utils.utils import simulate_paired_scores

# Simulate paired binary data (y_A[i] and y_B[i] refer to the same item/subject)
sim = simulate_paired_scores(N=100, delta_A=0.5, seed=42)
y_A, y_B = sim.y_A, sim.y_B

model = PairedBayesPropTest(seed=42).fit(y_A, y_B)
model.print_summary()

# Unified decision
d = model.decide()
print(f"BF₁₀ = {d.bayes_factor.BF_10:.2f}  →  {d.bayes_factor.decision}")
print(f"ROPE: {d.rope.decision}")

model.plot_savage_dickey()

For exact MCMC inference with convergence diagnostics:

from bayesprop.resources.bayes_paired_pg import PairedBayesPropTestPG

# Reuse paired data from above
model = PairedBayesPropTestPG(seed=42, n_iter=2000, burn_in=500, n_chains=4)
model.fit(y_A, y_B)
model.print_summary()

# MCMC diagnostics
diag = model.mcmc_diagnostics()
print(f"R-hat (δ_A): {diag.delta_A.r_hat:.3f}")
print(f"ESS (δ_A):   {diag.delta_A.ess:.0f}")

Sample-size planning¶

Use Bayes Factor Design Analysis (BFDA) to determine how many observations you need:

from bayesprop.utils.utils import bfda_power_curve, plot_bfda_power

curve = bfda_power_curve(
    theta_A_true=0.85, theta_B_true=0.70,
    sample_sizes=[20, 50, 100, 200, 500],
    n_sim=500, seed=42,
)
plot_bfda_power(curve, theta_A_true=0.85, theta_B_true=0.70)

Return types¶

All inference methods return Pydantic models with typed, validated fields:

Method	Return type
`model.test()`	`NonPairedTestResult`
`model.fit()` → `model.summary`	`NonPairedSummary` / `PairedSummary`
`model.decide()`	`HypothesisDecision`
`model.savage_dickey_test()`	`SavageDickeyResult`
`model.posterior_probability_H0()`	`PosteriorProbH0Result`
`model.rope_test()`	`ROPEResult`
`model.ppc_pvalues()`	`dict[str, PPCStatistic]`
`model.mcmc_diagnostics()`	`MCMCDiagnostics`

See Data Schemas for full field documentation.