- NGMRL : Solving the Neoclassical Growth Model with Reinforcement Learning
- The VFI project: coding the Value Function Iteration algorithm in multiple languages
Note: vfi.ipynb is an implementation of Value Function Iteration in Google JAX. Executed on COLAB with T4 GPU yields impressive speed gain in large-dimensional problems. The results of this simulation, run on Google Colab are as follows.
| Hardware x Dimensionality | Pure Numpy | Pure JAX | Numpy with compiled JAX function |
|---|---|---|---|
| CPU, N=50 | 76.7ms | 18.4ms | 46.8ms |
| CPU, N=5000 | 1min56s | 30.4s | 26.8s |
| GPU, N=50 | 169ms | 135ms | 219ms |
| GPU, N=5000 | 1min20s | 274ms | 326ms |