NzoCs/Learning-point-processes
New-LTPP is a modern, advanced framework for Temporal Point Process (TPP) research and development. Originally inspired by EasyTPP, this project has evolved into a comprehensive toolkit with significant enhancements in performance, usability, and research capabilities.
What's novel
New-LTPP is a modern, advanced framework for Temporal Point Process (TPP) research and development. Originally inspired by EasyTPP, this project has evolved into a comprehensive toolkit with significant enhancements in performance, usability, and research capabilities.
Code Analysis
11 files read · 4 roundsA comprehensive PyTorch framework for training, simulating, and statistically evaluating neural temporal point process models (NHP, THP, ODETPP, etc.) with novel MMD-based distribution matching tests using custom point process kernels.
Strengths
Genuinely deep implementations of multiple TPP models with correct mathematical formulations (continuous-time LSTM decay, thinning-based rejection sampling, proper log-likelihood with MC integration), plus a novel statistical testing framework using M-kernels and MMD permutation tests specifically designed for variable-length event sequences.
Weaknesses
Some code has French comments mixed with English, the test suite appears limited relative to the project's scope (only a few test files visible), and the simulation loop is inherently sequential (one event at a time) which limits throughput for long sequences.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
182
Contributors
1
Files
184
Active weeks
26
Repository
Language
Python
Stars
1
Forks
0
License
—