Embedded Laplace: fix the GLM likelihoods for multiple observations per group and vector dispersion - #3424
Merged
SteveBronder merged 4 commits intoOct 1, 2026
Conversation
…ons per group The Poisson and Bernoulli grouped likelihoods aggregate observations with a loop bounded by theta.size() instead of the number of observations, so any data set with more observations than latent variables is silently truncated. The negative binomial likelihood only works with a scalar dispersion, while the Stan signature requires a vector, making the built-in unusable from Stan programs. The new tests compare against the general laplace_marginal with per-observation prim lpmfs.
Poisson and Bernoulli: the aggregation loop ran over theta.size() instead of the number of observations, silently dropping observations beyond the latent dimension (and reading out of bounds when there were fewer observations than latent variables). The loop now runs over y_index. The Poisson normalizing constant is computed per observation (lgamma(y_i + 1)) instead of per group so the returned value is the actual marginal of y; the two agree for one observation per group, which is all the previous tests exercised. Negative binomial: the dispersion eta was silently assumed scalar; a vector eta hit a matrix product and an elementwise op with mismatched shapes, so the Stan-facing signature (vector) never worked. eta may now be a scalar shared by all groups or a vector with one entry per group, checked against the latent dimension. The summary variant has no per-observation index, so it now requires a scalar eta explicitly. The latent rng variants reuse these likelihood structs and are fixed by the same change.
Contributor
Jenkins Console Log Machine informationDistributor ID: Ubuntu Description: Ubuntu 20.04.3 LTS Release: 20.04 Codename: focal CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian Address sizes: 43 bits physical, 48 bits virtual CPU(s): 256 On-line CPU(s) list: 0-255 Thread(s) per core: 2 Core(s) per socket: 64 Socket(s): 2 NUMA node(s): 2 Vendor ID: AuthenticAMD CPU family: 23 Model: 49 Model name: AMD EPYC 7742 64-Core Processor Stepping: 0 Frequency boost: enabled CPU MHz: 1498.470 CPU max MHz: 3416.0681 CPU min MHz: 1500.0000 BogoMIPS: 4491.85 Virtualization: AMD-V L1d cache: 4 MiB L1i cache: 4 MiB L2 cache: 64 MiB L3 cache: 512 MiB NUMA node0 CPU(s): 0-63,128-191 NUMA node1 CPU(s): 64-127,192-255 Vulnerability Gather data sampling: Not affected Vulnerability Indirect target selection: Not affected Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Not affected Vulnerability Old microcode: Not affected Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection Vulnerability Spec rstack overflow: Mitigation; Safe RET Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsa: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Mitigation; IBPB before exit to userspace Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es G++: g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Clang: clang version 10.0.0-4ubuntu1 Target: x86_64-pc-linux-gnu Thread model: posix InstalledDir: /usr/bin |
SteveBronder
enabled auto-merge
September 30, 2026 21:15
SteveBronder
approved these changes
Sep 30, 2026
Contributor
Jenkins Console Log Machine informationDistributor ID: Ubuntu Description: Ubuntu 20.04.3 LTS Release: 20.04 Codename: focal CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian Address sizes: 43 bits physical, 48 bits virtual CPU(s): 256 On-line CPU(s) list: 0-255 Thread(s) per core: 2 Core(s) per socket: 64 Socket(s): 2 NUMA node(s): 2 Vendor ID: AuthenticAMD CPU family: 23 Model: 49 Model name: AMD EPYC 7742 64-Core Processor Stepping: 0 Frequency boost: enabled CPU MHz: 1497.132 CPU max MHz: 3416.0681 CPU min MHz: 1500.0000 BogoMIPS: 4491.56 Virtualization: AMD-V L1d cache: 4 MiB L1i cache: 4 MiB L2 cache: 64 MiB L3 cache: 512 MiB NUMA node0 CPU(s): 0-63,128-191 NUMA node1 CPU(s): 64-127,192-255 Vulnerability Gather data sampling: Not affected Vulnerability Indirect target selection: Not affected Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Not affected Vulnerability Old microcode: Not affected Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection Vulnerability Spec rstack overflow: Mitigation; Safe RET Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsa: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Mitigation; IBPB before exit to userspace Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es G++: g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Clang: clang version 10.0.0-4ubuntu1 Target: x86_64-pc-linux-gnu Thread model: posix InstalledDir: /usr/bin |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
AI disclosure: The code as well as this commentary was made with the help of claude Fable.
It's motivated by real issues I encountered.
Summary
Two separate bugs in the built-in Laplace GLM likelihoods, both invisible to the existing tests because those only ever use one observation per group (
y_index = 1:n) and a scalar dispersion.poisson_log_likelihood,bernoulli_logit_likelihood): the aggregation loop ranfor (i = 0; i < theta.size(); i++)while indexing the per-observation arraysy[i],y_index[i]. With more observations than latent variables the extra observations were silently dropped (e.g. builtin −5.18 vs. correct −13.17 on the new Poisson test case); with fewer it read out of bounds. The loop now runs overy_index.size(). The Poisson normalizing constant is also now-sum(lgamma(y_i + 1))per observation instead of per group; the two agree in the one-observation-per-group case, so no previously-correct result changes.neg_binomial_2_log_likelihood): the dispersionetaonly worked as a scalar (multiply(n_per_group, eta)is a matrix product,log_sum_exp(theta_offset, log_eta)is elementwise,add(y_map, eta)is per observation — no vector length satisfies all three), while the Stan-language signature requiresvector eta, making the built-in unusable from Stan (stanc3master, CmdStan 2.39).etamay now be a scalar shared by all groups or a vector with one entry per group (size-checked againsttheta); the binomial-coefficient term indexes the per-group dispersion throughy_index. The Stan Functions Reference documentsetaasreal, so a companion stanc3 PR (branchbugfix/laplace-neg-binomial-scalar-eta) changes the signature toreal; accepting a per-group vector on the Math side additionally keepsrep_vector(phi, M)working with already-released stanc3.The
_rngvariants reuse these likelihood structs and are fixed by the same change. The C++-only_summaryvariant has no per-observation index, so a per-group dispersion is not computable there; itsetais now explicitly constrained to a scalar (require_stan_scalar_t).Tests
All new tests fail without the patch.
laplace_marginal_{poisson_log,bernoulli_logit,neg_binomial_log}_lpmf_test.cpp: a shared setup with 6 observations, 3 latent variables, group sizes 3/2/1 and a non-zero vector mean, compared against the generallaplace_marginalwith a per-observation reference likelihood built on the prim*_lpmf. Without the patch, Poisson and Bernoulli return wrong values (observations dropped) and the negative binomial with vectoretadoes not even compile fordouble(from Stan,varinstantiations threw at runtime).etawith unequal group sizes matches the reference; a constant vectoretamatches the scalar result; a wrong-lengthetathrowsstd::invalid_argument;expect_adover vectoretaand vector mean verifies derivatives (both must stay differentiable for use underoptimize).expect_adw.r.t. the mean with more observations than latents.Ran the eight affected suites (the three above plus
..._summary_lpmf, the three_rngtests andlaplace_latent_solve_test): 174 tests, 0 failures.Side Effects
neg_binomial_2_log_likelihood_summaryno longer accepts a non-scalareta(it never produced a usable result with one). This needs as stanc3 fix as well.Release notes
Fixed the embedded Laplace GLM likelihoods (
laplace_marginal_{poisson_log,bernoulli_logit,neg_binomial_2_log}_lpmfand the corresponding_rngfunctions) to handle multiple observations per group: observations beyond the latent dimension were previously ignored. The negative binomial dispersion now accepts a scalar or one value per group, making the function usable from Stan, whose signature requires a vector.Checklist
Copyright holder: Jachym Barvinek
The copyright holder is typically you or your assignee, such as a university or company. By submitting this pull request, the copyright holder is agreeing to the license the submitted work under the following licenses:
- Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause)
- Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
the basic tests are passing
./runTests.py test/unit)make test-headers)make test-math-dependencies)make doxygen)make cpplint)the code is written in idiomatic C++ and changes are documented in the doxygen
the new changes are tested