Lorenz 84
- Average works as well as minimum for non-normalized, and works better for normalized.
- Normalizing features improves results: 3 clusters are identified even for more numics.
Several seeds for normalized and with average
Optimizing epsilon, fixed min neighbors
Works very well for all seeds. In some, works perfectly.
500 ics
5000 ics
Epsilon via elbow method, fixed min neighbors
Worse, but still ok. Finds a few attractors in FP clusters. Also sometimes in the other clusters. Optimal radius is a bit small. Is much faster though
500
5000