DTW based Affinity Propagation Clustering
AP Clustering using DTW distance for temporal sequences classification.
CharacterTrajectory
Time Consumption
| N | N(N-1)/2 | np_dtw (r=10) | np_dtw_parallel (r=10, cpu=15) | tf_dtw (batches) |
|---|---|---|---|---|
| 10 | 45 | 8.07s | 1.29s | 16.31s (1) |
| 50 | 1,225 | 229.88 | 26.60 | 72.30 (5) |
| 100 | 4,950 | 959.57 | 102.74 | 72.97 (5) |
| 200 | 19,900 | about 4k (1.1h) | 402.69 | 77.13 (5) |
| 1,000 | 499,500 | about 100k (27h) | about 10k (2.8h) | 795.20 (50) |
| 2,858 | 4,082,653 | about 784k (9d) | about 78.4k (21h) | 6476.16 (400) |
Results
Accuracy is 86.85% in train_set and 83.43% in test_set.
| DTW Matrix | Confusion Matrix |
|---|---|
 |
 |