UoI Var

[arokem]

This is Yao's recent accelerated versions.

[Copilot]

Pull request overview

This PR introduces Yao's accelerated versions for UoI-VAR (Union of Intersections for Vector Autoregressive models), implementing significant enhancements to support VAR model fitting with Poisson GLM and ADMM-based distributed optimization.

Key Changes:

• Replaced mean-based log-likelihood calculation with sum-based calculation for consistency with Poisson loss function in VAR case
• Added FDR (False Discovery Rate) based edge selection method as an alternative to frequency-based intersection
• Implemented ADMM (Alternating Direction Method of Multipliers) MPI solver for distributed Lasso and Poisson regression
• Added comprehensive VAR model support including vectorization utilities and parameter packing/unpacking functions

Reviewed changes

Copilot reviewed 98 out of 100 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/pyuoi/utils.py	Modified log-likelihood calculation to use sum instead of mean for VAR consistency
src/pyuoi/mpi_utils.py	Added explicit int64 casting for array sizes and displacements in MPI operations
src/pyuoi/linear_model/utils.py	Added magnitude thresholding, FDR-based intersection, and VAR parameter utilities
src/pyuoi/linear_model/sparse_comm_util.py	New utility for sparse matrix MPI communication
src/pyuoi/linear_model/poisson.py	Major refactor: added ADMM support, feature weights, time step parameter, and vectorized coefficient handling
src/pyuoi/linear_model/logistic.py	Replaced deprecated log_logistic with stable_log_logistic implementation
src/pyuoi/linear_model/lasso.py	Added ADMM solver support and timing instrumentation
src/pyuoi/linear_model/base.py	Major refactor: added VAR model support, vectorization functions, and FDR-based selection
src/pyuoi/linear_model/admm_mpi_poisson.py	New ADMM solver implementation for Poisson regression with MPI support
src/pyuoi/linear_model/admm_mpi_dale.py	New ADMM solver for Dale's principle constrained models
src/pyuoi/linear_model/admm_mpi.py	New ADMM solver for standard Lasso with MPI support
src/pyuoi/lbfgs/_lowlevel.pyx	Replaced deprecated np.product with np.prod
examples/*	New example scripts and utilities for VAR model simulation and evaluation
causal_net/*	New causal network analysis tools and visualization utilities

Comments suppressed due to low confidence (6)

src/pyuoi/linear_model/poisson.py:1

• Hardcoded value of 20 for p should be calculated from the input coefficients array. The correct value should be p = int((-1 + np.sqrt(1 + 4 * len(coefficients))) / 2) based on the relationship len(coefficients) = p * (p + 1).
  examples/uoi_var_poisson_test.py:1
• Invalid syntax: uoi_poisson.self.l1_support_count should be uoi_poisson.l1_support_count. The self keyword is unnecessary and causes an attribute error.
  src/pyuoi/linear_model/utils.py:1
• Documentation states default is 0.05, but the function signature shows min_w=0.01. Update documentation to match the actual default value of 0.01.
  examples/var_utils.py:1
• Malformed header decoration contains 'r[i]' which appears to be a typo. Should be a consistent line of '=' characters.
  src/pyuoi/linear_model/poisson.py:1
• Corrected extraneous 'f' in docstring comment.
  src/pyuoi/linear_model/lasso.py:1
• Corrected spelling of 'default' in docstring.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.