rename log_gamma module to sbc

Author: daniel-habermann

bayesflow/diagnostics/metrics/__init__.py

@@ -4,4 +4,4 @@
 from .expected_calibration_error import expected_calibration_error
 from .classifier_two_sample_test import classifier_two_sample_test
 from .model_misspecification import bootstrap_comparison, summary_space_comparison
-from .log_gamma import log_gamma, gamma_null_distribution, gamma_discrepancy
+from .sbc import log_gamma

bayesflow/diagnostics/metrics/log_gamma.py renamed to bayesflow/diagnostics/metrics/sbc.py

@@ -19,7 +19,7 @@ def log_gamma(
     Log gamma is log(gamma/gamma_null), where gamma_null is the 5th percentile of the
     null distribution under uniformity of ranks.
     That is, if adopting a hypothesis testing framework,then log_gamma < 0 implies
-    a rejection of the hypothesis of uniform ranks at the 5\% level.
+    a rejection of the hypothesis of uniform ranks at the 5% level.
     This diagnostic is typically more sensitive than the Kolmogorov-Smirnoff test or
     ChiSq test.
 

tests/test_diagnostics/test_diagnostics_metrics.py

@@ -108,20 +108,20 @@ def run_sbc(N=N, S=S, D=D, bias=0):
         ranks = np.sum(posterior_draws < prior_draws, axis=0)
 
         # this is the distribution of gamma under uniform ranks
-        gamma_null = bf.diagnostics.metrics.gamma_null_distribution(D, S, num_null_draws=100)
+        gamma_null = bf.diagnostics.metrics.sbc.gamma_null_distribution(D, S, num_null_draws=100)
         lower, upper = np.quantile(gamma_null, (0.05, 0.995))
 
         # this is the empirical gamma
-        observed_gamma = bf.diagnostics.metrics.gamma_discrepancy(ranks, num_post_draws=S)
+        observed_gamma = bf.diagnostics.metrics.sbc.gamma_discrepancy(ranks, num_post_draws=S)
 
         in_interval = lower <= observed_gamma < upper
 
         return in_interval
 
     sbc_calibration = [run_sbc(N=N, S=S, D=D) for _ in range(100)]
-    lower_expected, upper_expected = binom.ppf((0.005, 0.995), 100, 0.95)
+    lower_expected, upper_expected = binom.ppf((0.0005, 0.9995), 100, 0.95)
 
-    # this test should fail with a probability of 1%
+    # this test should fail with a probability of 0.1%
     assert lower_expected <= np.sum(sbc_calibration) <= upper_expected
 
     # sbc should almost always fial for slightly biased posterior draws