| pyvgx Vertex Vector Methods | Description |
|---|---|
Assign a feature vector to the vertex. |
|
Return |
|
Return the vertex vector. |
|
Remove the vertex vector. |
|
Return 32-bit lsh region not overlapping with lsh segment used for projection key |
1. pyvgx.Vertex.SetVector()
Assign a feature vector to the vertex.
1.1. Syntax
pyvgx.Vertex.SetVector( elements ) -> pyvgx.Vector1.2. Parameters
elements: List of 2-tuples [ (feature, weight), (feature, weight), … ]
-
feature is a string of no more than 27 characters (silent truncation will occur)
-
weight is a number constrained to the range [0.0078125 – 1.875] (silent capping at either end will occur)
-
The maximum length of a vector is 48 features.
1.3. Return Value
A new pyvgx.Vector object instance representing the vertex feature vector will be returned. The returned object does not wrap the vertex vector, i.e. there is no way to modify a vector once set. To change the vertex vector one must call pyvgx.Vertex.SetVector() again with a different element list.
The returned vector is useful for inspection and comparison processing.
1.4. Remarks
Setting the vertex vector enables similarity matching for the vertex.
Being enabled for similarity matching means the vertex may be a candidate for retrieval (or positive filter match) for a query that includes a similarity threshold condition.
| The vector is subject to truncation and quantization. Vector feature strings will be truncated beyond the 27th character and weights will be constrained to the interval [0.0078125 – 1.875] and quantized according to a set of pre-defined values. (Please see pyvgx.Vector for technical details.) |
| If feature strings are utf-8 encoded it is possible for truncation to occur in the middle of a utf-8 byte sequence resulting in invalid string data if you try to print the string after retrieving it from a vertex. (Internal processing and comparison will continue work without issue.) For this reason, if you plan to insert vectors with utf-8 encoded features make sure the encoded byte-sequence is shorter than 27 bytes or pre-truncate to avoid cutting off a utf-8 encoded unicode character in the middle. |
1.5. Example
from pyvgx import *
g = Graph("g")
A = g.NewVertex("A")
B = g.NewVertex("B")
vA = A.SetVector( [("this",1.3),("is",1.1),("a",0.9),("vector",0.7)] )
vB = B.SetVector( [("this",1.7),("is",1.2),("another",0.9),("vector",0.5)] )
# Inspect
vA.Magnitude() # 1.95
vB.Magnitude() # 2.22
# -> [('this', 1.25), ('is', 1.0), ('a', 0.875), ('vector', 0.6875)]
vA.External()
# -> [(27871805, 61), (43031197, 57), (37390563, 54), (51389292, 48)]
vB.Internal()
# Compare
g.Cosine( vA, vB ) # 0.80859375
g.Jaccard( vA, vB ) # 0.527343752. pyvgx.Vertex.HasVector()
2.1. Syntax
pyvgx.Vertex.HasVector()2.2. Return Value
Returns True if the vertex has a similarity vector, otherwise False.
3. pyvgx.Vertex.GetVector()
Return the vertex vector.
3.1. Syntax
pyvgx.Vertex.GetVector() -> pyvgx.Vector3.2. Return Value
If the vertex has a vector a new pyvgx.Vector object representing that vector is returned. The returned object does not wrap the vertex vector, i.e. there is no way to modify a vector once set. To change the vertex vector one must call pyvgx.Vertex.SetVector() with a different element list.
If the vertex does not have a vector a new pyvgx.Vector object representing the null-vector is returned.
The returned vector is useful for inspection and comparison processing.
3.3. Example
from pyvgx import *
g = Graph("g")
A = g.NewVertex("A")
A.SetVector( [("this",1.3),("is",1.1),("a",0.9),("vector",0.7)] )
V = A.GetVector()
# -> <PyVGX_Vector: len=4 mag=1.9486 fp=0xDEF40C4CB2D466A1
# elem=[('this', 1.25), ('is', 1.0), ('a', 0.875),
# ('vector', 0.6875)]>
V4. pyvgx.Vertex.RemoveVector()
4.1. Syntax
pyvgx.Vertex.RemoveVector()4.2. Remarks
Removes the vector from the vertex.
5. pyvgx.Vertex.ArcLSH()
Return 32 lsh bits from a region of the 64-bit LSH not overlapping with LSH segment used to generate a projection key.
5.1. Syntax
pyvgx.Vertex.ArcLSH( lsh64 ) -> lsh325.2. Parameters
lsh64 is the 64-bit LSH from which we will extract 32 bits to return. The vertex must have been created by pyvgx.Similarity.CreateProjectionSets() for this to have the desired effect.
5.3. Return Value
A 32-bit segment extracted from lsh64 is returned.
5.4. Remarks
Call this method to obtain a 32-bit integer suitable for use with arc modifier pyvgx.MLSH.
Vector index implementations may want to set this value on arcs from projection nodes to items holding vectors to make neighborhood queries more efficient. Such queries may want to use a hamming distance filter before traversing the arc to the target item node, thereby reducing expensive memory access and vector similarity computations.
5.5. Example
from pyvgx import *
system.Initialize( "vectors", euclidean=1 )
g = Graph("ann")
# Initialize index structure
g.sim.CreateProjectionSets(1,10)
# Create a random vector
V = g.sim.NewVector( [random.random() for i in range(768)] )
# Get LSH for seed 0
lsh = V.Fingerprint( 0 )
# Extract 32 bits from lsh not overlapping with the segment
# of lsh used for projection 0x1234
g["_1234|000"].ArcLSH( 0 ) # -> 32 bit integer