-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathTODO
127 lines (105 loc) · 4.83 KB
/
TODO
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
FALSE-node bug (Netica Node object is replaced with FALSE).
Nasty bug where RReference gets overwritten in network. Not sure
where this is coming from, as they should be preserved.
Nodelist <- NetworkNodesInSet(LevelMotif,"EMVars")
NodeName(Nodelist$LeverTrohies) <- "LeverTrophies"
The corresponding element of the list is now FALSE and
NetworkFindNode(LevelMotif,"LeverTrophies")
returns FALSE.
I'm having trouble reproducing this with a small example. Maybe some
interaction with the garbage collector? I've added a check for bad
object caches in the Net/Node/UserData field, now check to make sure
the returned object is of the right class, and if not recreates it. I
wish I could more reliably reproduce the bugs so I can see if this
fixed the problem.
UPDATE: I've made some changes that may fix this. I've also now run
a couple of times using CHECK --gc-torture and not found the problem.
It may be fixed.
UPDATE 2: This is probably now irrelevant after going from S3 to R6
classes for Nodes and Nets. Will watch for it.
R stack error in BN_SetNodeProbs when passed a matrix of NA's as
probability table. This may be a problem with error reporting.
I cannot reproduce this.
Parse the junction tree report
I'm getting an error GetError_ns: deleted or damage report_ns passed
under Windows, 32-bit. The condition that seemed to trigger this was
compiling a network with uninitialized CPTs (which generates a
warning). This works correctly under Linux and Windows-64 bit. The
error seems to persist after it was first encountered.
UPDATE: I think I've found the problem. It has to do with passing a
"cleared" error as the previous error.
Seg fault if CreateNetwork_bn() is called with a null environment.
(Thanks to Elise Bethany Gould for spotting this).
Seg fault if NewNetworkTester_bn() is called with a null nodelist
(either one).
Functions for
## ReorderNodeStates, RemoveNodeState,
These seem to be redundant with current definition of NodeStates<-
## Update. Folded AddNodeStates and RemoveNodeState into
## NodeStates<-. Reorder Node States does something to the CPT, so
## There still might be a use for it.
Learning from Data (Both of these depend on Netica API updates)
## LearnCPTs -- update when Netica Fixes bugs
## Test Case streams with LearnCPTs and LearnFindings
Equations
Equation Suppot is currently pretty minimal. Probably not important
as it is easier to build CPTs in R than use equations.
Continuous Node support
Decision Net support
## EnterGaussianFinding,
## EnterIntervalFinding,
These two functions are giving unexpected results. Query out to
Norsys about this, i.e., they always give weight 1 to a single
state, rather than distributiong the output over a number of
states as expected.
Update: Worked around this by calling NodeLikelihood() instead of
the Netica API.
These two functions return NA in stituations where I'm expecting
a numeric value or a state.
## CalcNodeState, CalcNodeValue,
Testing
## NewNetTester_bn, DeleteNetTester_bn, TestWithCaseset_bn
## GetTestConfusion_bn, GetTestErrorRate_bn, GetTestLogLoss_bn
## GetTestQuadradicLoss_bn
## TestWithCaseset_bn,
Created testNetwork function and NetworkTester object to hold
results. The latter needs documentation.
Random Generation
Random generation implemented and tested, but probably very
inefficient in the simple case were we want to generate a set of
several hundred cases to a case file. Should reimplement that on the
C side so we don't need to constantly do the R->c conversions.
DBNs
## ExpandNet_bn, SetNodeInputDelay_bn, SetNodePersistance_bn,
GetNodeAtTime_bn
There are currently no corresponding Get functions for these.
Do Not implement
# NetworkUndo, NetworkRedo # These don't do anything sensible.
## SetLearnerRandomGen_bn #This is for future expasion (MCMC?)
Database Connectivity
## NewDBManager_cs, DeleteDBManager_cs
## InsertFindingsIntoDB_bn, AddDBCasesToCaseset_cs
## ExcecuteDBSql_cs, AddNodesFromDB_bn
## WriteCaseset_ns (The only use for this function is in conjunction
## with the database stuff.)
Ask Brent:
1. NodeLikelihood and MPE and FindingsProbability
2. IsNetCompiled function? -- Node size works.
* 3. GetNodesets for Node, Nodes in NodeSet?
* 4. CopyNodes does not copy NodeSet information?
Still outstanding.
* 5. EnterLikelihoodFindingClique
* 6. NetworkUndo, Redo
* 7. Learn_CPTs return value?
It has one, but what it is?
* 8. Mysterious behavior of CalcNodeValue
x [RGA] 9. Unexpected behavior of EnterGaussianFinding and EnterIntervalFinding
10. Error on NodeExpectedUtils
* 11. Learn_CPTs and case streams.
* 12. Crash in AbsorbNode
* 13. Getters for DBN properties
x ReadNetwork_bn has switch to suppress loading visual information.
* 14. Segfault on NewNodeTester_bn()
Issues with Version 6.07
* NodeExperience <- UNDEF_DBL
* GRADIENT_DESCENT_LEARNING