mims-harvard / scikit-fusion Goto Github PK

Hi there,
I'm trying to use skfusion with my data, but I feel there is so little documentation describing functions and object types. I'm doing a little of reverse engineering from example data, so far I've got that fuser.factor(factor_name) returns the matrix for a specifed factor and fuser.backbone(relation_name) returns the backbone matrix for a specified relations. Options for ObjectType, Relation, FusionGraph objects are not described, neither Dfmf options are. Also, a good hint guide about choosing matrix ranks would help.

Two issues occur when working with small samples:

• (rank = 1) problem.
  Running:

import numpy as np
from skfusion import fusion

R12 = np.random.rand(30, 40)
R23 = np.random.rand(40, 50)

t1 = fusion.ObjectType('Type 1', 2)
t2 = fusion.ObjectType('Type 2', 9)
t3 = fusion.ObjectType('Type 3', 1)

relations = [fusion.Relation(R12, t1, t2),
             fusion.Relation(R23, t2, t3)]
fusion_graph = fusion.FusionGraph(relations)

fuser = fusion.Dfmf()
fuser.fuse(fusion_graph)

yields: KeyError: (ObjectType("Type 1"), ObjectType("Type 3"))
the code works if t3 rank is >= 2

• (dim < 5) problem.

Running:

import numpy as np
from skfusion import fusion

R12 = np.random.rand(30, 4)
R23 = np.random.rand(4, 50)

t1 = fusion.ObjectType('Type 1', 2)
t2 = fusion.ObjectType('Type 2', 9)
t3 = fusion.ObjectType('Type 3', 2)

relations = [fusion.Relation(R12, t1, t2),
             fusion.Relation(R23, t2, t3)]
fusion_graph = fusion.FusionGraph(relations)


fuser = fusion.Dfmf()
fuser.fuse(fusion_graph)

yield numpy warning and reconstructed matrices have NaN. This works if R12 has second dimension greater than 4.

In the file README.md: the links to the slides in the tutorial section are no longer functional.
The lines in question:
https://github.com/mims-harvard/scikit-fusion/blob/88dd02cc43a43c1c554602b5da6120de0128f188/README.md?plain=1#L128L129