Table 7 Number of trainable parameters and training time of each time-varying graph representation learning model for LyonSchool and the two synthetic datasets. The embedding dimension is fixed to 128, technical specifications of the computing system and hyper-parameters configuration are reported in Additional file 1
From: Time-varying graph representation learning via higher-order skip-gram with negative sampling
Model | Dataset | |||||
|---|---|---|---|---|---|---|
LyonSchool | OpenABM-2k-100 | OpenABM-5k-20 | ||||
\(|\mathcal{V}|=242\), \(|\mathcal{T}|=104\) | \(|\mathcal{V}|=2000\), \(|\mathcal{T}|=100\) | \(|\mathcal{V}|=5000\), \(|\mathcal{T}|=20\) | ||||
Tr. parameters | Tr. time | Tr. parameters | Tr. time | Tr. parameters | Tr. time | |
DyANE | 4,396,544 | \( 62~\text{s} \) | 50,825,472 | \( 1014~\text{s} \) | 25,591,296 | \( 448~\text{s} \) |
DynGEM | 459,270 | \( 516~\text{s} \) | 1,867,428 | \( 10\text{,}765~\text{s} \) | 4,270,428 | \( 23\text{,}307~\text{s} \) |
DynamicTriad | 3,221,632 | \( 1131~\text{s} \) | 25,600,128 | \( 17\text{,}191~\text{s} \) | 12,800,128 | \( 12\text{,}625~\text{s} \) |
DySAT | 98,336 | \( 18\text{,}323~\text{s} \) | 323,232 | \( 152\text{,}976~\text{s} \) | 707,232 | \( 8958~\text{s} \) |
ISGNS | 61,952 | \( 381~\text{s} \) | 512,000 | \( 5895~\text{s} \) | 1,280,000 | \( 3062~\text{s} \) |
\(\text{HOSGNS} ^{(\text{stat})}\) | 75,264 | \( 316~\text{s} \) | 524,800 | \( 548~\text{s} \) | 1,282,560 | \( 724~\text{s} \) |
\(\text{HOSGNS} ^{(\text{dyn})}\) | 88,576 | \( 303~\text{s} \) | 537,600 | \( 565~\text{s} \) | 1,285,120 | \( 734~\text{s} \) |