Overview
Unlike the classic random walk, Node2Vec random walk adopts a second-order random walk and controls whether the walk is biased toward BFS or DFS through parameters. The sequences generated by the Node2Vec Walk algorithm are used as samples for the Node2Vec graph embedding algorithm, as detailed in the chapter Node2Vec.
Results and Statistics
Perform Node2Vec random walk in the graph below for 3 times, start from node 3 with a depth of 5, each edge weight is 1, set p as 10000 and q as 0.01:
Algorithm results: 3 node arrays are contained in the returned walks
| walks |
|---|
| [3, 5, 3, 1, 2] |
| [3, 4, 6, 7, 8] |
| [3, 1, 2, 1, 3] |
Algorithm statistics: N/A
Command and Configuration
- Command：
algo(random_walk_node2vec) - Configurations for the parameter
params():
| Name | Type | Default |
Specification |
Description |
|---|---|---|---|---|
| ids / uuids | []_id / []_uuid |
/ | / | IDs or UUIDs of nodes to start the walk; all nodes to be selected if not set |
| walk_length | int | 1 | >=1 | Depth of each walk, i.e. the number of nodes walking through |
| walk_num | int | 1 | >=1 | Number of walks |
| p | float | 1 | >0 | return parameter; the larger the value, the smaller the probability of returning |
| q | float | 1 | >0 | in-out parameter that represents the probability of being to walk far away; >1 means tend to walk at the same level, >1 means tend to walk far away |
| edge_schema_property | []@<schema>?.<property> |
/ | Numeric edge property, LTE needed | Edge weight property/properties, schema can be either carried or not; nodes only walk along edges with the specified properties and the probability of passing through these edges is proportional to the edge weight; if edge has multiple specified properties, the edge weight is the sum of these property values; the weight of all edges is 1 if not set |
| buffer_size | int | 1000 | / | Number of results to return; < 0 means to return all results, ≧ 0 means to return partial results |
Example: Select nodes with UUID = 1,2,3 to perform Node2Vec random walk for 3 times with a depth of 5, set p as 10000 and q as 0.01
algo(random_walk_node2vec).params({
uuids: [1,2,3],
walk_num: 3,
walk_length: 5,
p: 10000,
q: 100
}) as walk
return walk
Algorithm Execution
Task Writeback
1. File Writeback
Configuration |
Data in Each Row |
Description |
|---|---|---|
| filename | _id,_id,... |
IDs of nodes that walked through |
Example: Select nodes with UUID = 1,2,3 to perform Node2Vec random walk for 3 times with a depth of 5, set p as 10000 and q as 0.01, write the algorithm results back to file named path
algo(random_walk_node2vec).params({
uuids: [1,2,3],
walk_num: 3,
walk_length: 5,
p: 10000,
q: 100
}).write({
file:{
filename: "path"
}})
2.Property Writeback
Not supported by this algorithm.
3. Statistics Writeback
This algorithm has no statistics.
Direct Return
| Alias Ordinal | Type | Description |
Column Name |
|---|---|---|---|
| 0 | []perWalk | Array of UUIDs of nodes that walked through each time | [_uuid, _uuid, ...] |
Example: Perform Node2Vec random walk in the whole graph for 10 times with a depth of 6, set p as 2 and q as 100, define algorithm results as alias named paths, and return the results
algo(random_walk_node2vec).params({
walk_num: 10,
walk_length: 6,
p: 2,
q: 100
}) as paths return paths
Streaming Return
| Alias Ordinal | Type | Description |
Column Name |
|---|---|---|---|
| 0 | []perWalk | Array of UUIDs of nodes that walked through each time | [_uuid, _uuid, ...] |
Example: Perform Node2Vec random walk in the whole graph for 10 times with a depth of 6, set p as 2 and q as 100, and specify edge weight locates on property @follow.level, return the results that walked more than 5 steps
algo(random_walk_node2vec).params({
walk_num: 10,
walk_length: 6,
p: 2,
q: 100,
edge_schema_property: @follow.level
}).stream() as walk
where size(walk) > 5
return walk
Real-time Statistics
This algorithm has no statistics.