Scientific Computing with F#
summary
Implements an Ising spin glass with a Metropolis annealer and strong domain types using units of measure, producing the core model in less than 150 lines and preventing type errors like treating spins as booleans. Reports a performance chase: initial run showed F# 40× faster then 2× faster after correcting build modes, and later F# wins again after fixing redundant recomputation. Also implements linear regression with linear algebra libs, composing four variants of gradient descent in about 250 lines; notes missing parametric modules for matrix-parameterized models.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:30Presentation
- 8:30Why Fsharp is good
- 9:30Improving Fsharp
- 10:30Performance
- 13:20Gradient Descent
- 18:45Example
- 19:40Composition
- 20:37Conclusion
- 21:46Questions
related talks
01h 02mMachine Learning with F#, Redux
43mHow to look like a statistician (...)
53mApplied Functional Programming in Computational Genomics
44mFast Neural Networks… a no brainer!
36mA developer's guide to probabilistic programming
26mTight Algorithmic Thresholds for Spin Glass Hamiltonians
51mAgile experiments in Machine Learning with F#
46mBeing Creative with Genetic Algorithms and Typeclasses