LSCD
Progress
Dataset
- GMM
- PGnorta
- Bikeshare
- other dataset
Baseline
- KDE
- GMM
Generator ML model
- RNN
- Attention layer
- Implement the Poisson count simulator as a PyTorch layer
- NoisyReinforcedMLP
- MLP
- Generator penalty
Discriminator ML model
- Geoless (compute and minimize W-distance directly)
- MLP
Evaluation metric
- Run through queue
- pairwise plot
- marginal mean
- marginal variance
- Pierre correlation
- Gaussian approximated W-distance
Others
- Gaussian gradient approximation precision (when lambda is small)
Timeline
The tasks are listed in order
| Task | Approximated time needed |
|---|---|
| Finalize the C version Run through queue code | 1 day |
| Link the C version Run through queue code to python | 1 day |
| Implement NoisyReinforcedMLP | 2 days |
| Check the old generator penalty code | 0.5 day |
| Compare the performance of (MLP + Geoless) (NoisyReinforcedMLP + Geoless) | Writing the code is fast, but there’s a lot of uncertainty in the performance of these model. |