Table 2 Next-element prediction performance of all models
From: Predicting variable-length paths in networked systems using multi-order generative models
BMS1 | FIFA | MSNBC | WIKI | AIR | TUBE | |
|---|---|---|---|---|---|---|
(a) Cross-entropy loss [bit] for prefixes up to length 6 | ||||||
MOGen | 6.14±.07 | 7.32±.12 | 2.85±.02 | 7.90±.06 | 4.20±.01 | 1.81±.00 |
AKOM | 19.60±.30 | 8.53±.11 | 6.01±.04 | 14.40±.05 | 13.27±.01 | 7.09±.00 |
CPT+ | 19.33±.28 | 10.66±.02 | 7.33±.04 | 17.42±.02 | 14.84±.01 | 10.42±.00 |
NET | 19.60±.30 | 8.69±.11 | 6.13±.04 | 14.40±.05 | 13.34±.01 | 7.57±.00 |
MOM | 19.60±.30 | 8.51±.12 | 6.01±.04 | 14.40±.05 | 13.20±.01 | 6.34±.01 |
RND | 19.68±.24 | 9.63±.04 | 6.73±.03 | 15.79±.03 | 14.11±.01 | 11.94±.00 |
(b) MOGen: detected and best performing maximum order | ||||||
detected | 1 | 1 | 2 | 1 | 2 | 6 |
best | 1 | 1 | 2 | 1 | 2 | 6 |