The evidence in one of the cases has zero probability. It is “impossible” under the model you are using. The analysis wizard in version 7.0 does not support such cases.
If you have constructed your model from data using EM learning, the you can ensure a non-zero probability of all cases using experience counts.
Yes, the HUGIN GUI allows the user to save and load structural constraints between nodes in a file. Structure constraints can be saved in and loaded from a file in the “Structure Constraints”-step of the Learning Wizard. You use the “save network with constraints” button to save structure constraints in a network specification file and the “import model information” button to load constraints from a network specification file.
For instance, if you want to disallow a certain direction on a link, then you would use the “No Arrow Constraint Tool” to define a constraint that a node X is not allowed to be parent of node Y. You can add multiple constraints in one operation by selecting a set of nodes and pressing the right mouse button on another node.
node D
{
label = “D”;
position = (210 110);
states = (“yes” “no”);
HR_Constraint_X = “NoArrow”;
}
Specifies that node D is not allowed to be a parent of X.
Maybe you can use the variance, which can be displayed in the monitor window of the node in the Graphical User Interface (GUI)?
Display of the variance in the GUI can be enabled in “Network->Network Properties” under the “Monitors”-tab.
This functionality is not available in the API.
P(e) is the probability of the evidence. E.g., if you have entered X=x and Y=y (assuming discrete nodes), then P(e)=P(X=x,Y=y).
It is calculated as the normalization constant after propagation of evidence in the junction tree.
For interval nodes, the values specified for state i and state i + 1 are the left and right endpoints of the interval denoted by state i (the dividing point between two neighboring intervals is taken to belong to the interval to the right of the dividing point), e.g., [0;1[.
Using the HUGIN .NET API it is possible to make use of HUGIN functionality in Matlab. The appropriate HUGIN dll file should be loaded as an assembly into Matlab.
The following example illustrates how to load the ChestClinic example from the Samples directory of the HUGIN installation and print the belief in node L:
Code:
ghapi = NET.addAssembly (‘C:\Program Files\Hugin Expert\Hugin Researcher 7.4\HDE7.4CS\Lib\hugincs-7.4-2.0-x64.dll’);
d = HAPI.Domain (‘C:\Program Files\Hugin Expert\Hugin Researcher 7.4\Samples\ChestClinic.net’, HAPI.DefaultClassParseListener);
L = d.GetNodeByName (‘L’);
d.Compile ();
L.GetBelief (0)
ans =
0.0550
L.GetBelief (1)
ans =
0.9450
The next code fragment shows how to select a state of a node, propagate the evidence and show the posterior beliefs in node L:
Code:
X = d.GetNodeByName (‘X’);
X.SelectState (0);
d.Propagate (HAPI.Equilibrium.H_EQUILIBRIUM_SUM, HAPI.EvidenceMode.H_EVIDENCE_MODE_NORMAL);
L.GetBelief (0)
ans =
0.4887
L.GetBelief (1)
ans =
0.5113
HUGIN uses floating point to support the representation and use of real numbers (e.g., probabilities), see e.g., http://en.wikipedia.org/wiki/Floating_point. On the wiki-page there is useful information on the use of floating point and its limitations.
In floating point a real number is represented using a (limited) number of significant digits. The number of significant digits depends on the data type used to represent a real number, i.e., the precision. In HUGIN we use either single (4 bytes) or double precision (8 bytes).
In single precision you have 7 significant digits. This means that single precision has sufficiently many digits to store, for instance, the number 0.1234567 in a variable with seven significant digits. A number such as 0.99999999 is, on the other hand, rounded to 1 and if you add 0.00000001, the number 0.00000001 is lost due to the limited precision (only seven significant digits).
As an example consider using HUGIN ActiveX server (single precision) in Excel to add the numbers above (i.e, 0.99999999 + 0.00000001). If Excel uses double precision, the sum becomes 1.00000001 (due to the fact that 0.99999999 is rounded to 1 by HUGIN due to the single precision nature of the HUGIN ActiveX server).
There are other examples which cannot be solved. For instance, consider computing the sum 1/3 + 1/3 + 1/3 using floating point in any (finite) precision. This will never sum to one no matter how many digits of precision you have. Thus, in a computer probabilities do not always sum to one.
The number of free parameters in a discrete CPT is (n-1) * m, where n is the number of states in the child and m is the number of parent configurations.
What could have caused the error in your model is that somehow the learning algorithm encounters cases that are impossible according to the probabilities in the model.
If you manually change the probabilities in the model prior to running the EM learner that may be what causes the problems. For instance, in your model you may have A –> B and some state in A which has prior probability 0 or a state in B which has conditional probability 0 given some state in A.
There are important differences related to influence diagrams between HUGIN 6.9 and HUGIN 7.x.
In HUGIN 7.0 we introduced “LIMIDs” (Lauritzen & Nilsson’01). The support for LIMIDs changes the semantics of the information arcs in the diagram and the solution algorithm.
In the traditional influence diagram (HUGIN 6.9) we assume the decision maker to be non-forgetting and use the Jensen, Jensen & Dittmer (94) algorithm to solve the diagram. This implies that the decision maker is assumed to recall all past observations and decisions. By this assumption some information arcs are assumed present. E.g., for the last decision all observations prior to the first decision are assumed observed due to not-forgetting. The diagram is solved by solving for each decision in reverse time ordering.
In the LIMID, all information arcs should be explicitly drawn in the diagram. There is no assumption about perfect recall (non-forgetting). Thus, an information arc from node X into the first decision does not imply that we assume an information arc from node X into any later decision to be implicitly present. Hence, we may model that the decision maker is forgetful.
Also, the solution algorithm is changed. The solution algorithm for LIMIDs is Single Policy Updating where we iteratively solve for each decision (in reverse time order, if an ordering is present). The user has to press the “SPU” button in the toolbar to run Single Policy Updating after compiling the network in order compute updated decision policies.
To change your model into a traditional influence diagram (if you want), you will have to add information arcs to the diagram.
The adaptation algorithm processes the cases individually. It assumes that the conditional distributions (one for each parent configuration) follow the Dirichlet distribution. When a new case arrives, it computes the updated distribution. However, if the new case is incomplete, the updated distribution is in general a mixture of Dirichlet distributions. In order to avoid exponential growth of the number of terms of the distribution, the updated distribution is replaced by a single Dirichlet distribution that has the same means and average variance (of the Dirichlet parameters) as the correct updated distribution.
The EM algorithm, on the other hand, considers all cases before it updates the CPTs. This is repeated until the CPTs converges to a stable maximum (stabilityis determined by a sufficiently small change of the log-likelihood between two iterations).
If all cases are complete, there is usually no difference between the methods –with one exception: the adaptation algorithm needs to have a valid starting distribution (this means that experience counts must be positive), whereas the EM algorithm is happy to start with experience counts equal to zero.
The adaptation algorithm is primarily intended to continuously updatethe CPT parameters when a system is already online (for example, it is useful if the parameters can change over time).
The EM algorithm will generally give the best results, but it is also more costly to run (since it processes all cases, and often does that several times). It is therefore not suitable if you want online updating of CPT parameters.
The default package is strictly a per user installation. Only the windows user account that installed HUGIN can run the program.
To install just log on to the target windows user account and double-click the installation file, afterwards only this windows user account can run the program. (DO NOT right-click-run-as-administrator the installation file – this will result in the program being installed for the administrator Windows user account and not the target Windows user account as desired).
The ‘admin’ package is a machine wide installation. Any windows user account on the computer can run the program. If you do not have the ‘admin’ installation package and a corresponding ‘admin’ license key, then please contact sales-at-hugin.com.
The following command-line is known to work (April 2019):
msiexec /i HuginDist87(x64).msi /L*v SilentInstallation.log /quiet PATH_LICENSE=”C:\TMP\LICENSE-8.7″
A couple of words on the installation:
Try the following and see if it helps:
1. Select a node
2. In the menu, click View -> Open tables
3. The CPT for the selected node should appear on screen
On macOS, it is necssary to install the LICENSE file using the dmg-installer. Here, you should make sure that the file is placed in a folder that the operating system allows the installer to read. For instance, the “Downloads” folder cannot be read by the installer.
Placing the LICENSE file in the folder “Documents” usually works. Do not place the LICENSE file on an USB or an external drive. This will not work.
Alternatively, the LICENSE file can be copied to (adjusted to your own system):
/Users/anders/Library/Application Support/com.hugin
