       |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
News Letter: May 2006
|
|
Data |
|
| The list of the available target variables that can be used for the discretization with decision trees contains now the continuous variables for which the discretization as been defined. | |
| An importation and association report is available at the end of the loading process. | |
| It is now possible to choose the encoding in
order to import file encoded with non default platform encoding (Shift_JIS,
UTF-8, UTF-16, UTF-16BE, UTF-16LE, iso, ibm, windows and many others are
supported). BOMs (Byte Order Masks) are supported for the UTF-8 and UTF-16
(BE and LE) encoding.
|
|
| The settings allow to choose the encoding of
the output files such as the saved or generated databases as well as the
generated html files (such as reports and others). This encoding format can
be changed in the setting panel through the "File encoding" combo box in the
option "Database>Save format". An encoding indicator is also added into the
meta of the html files.
|
|
| The path of the imported or associated data file is now kept in order to be quickly reused through the Recent database menu. | |
Dictionaries of costs, classes and modality
values have been added.
The dictionaries can be also exported.
|
|
| If the name of a node or a modality contains a
blank, it must be replaced by an underscore in the dictionary file. For
example, if you want to link a cost (10) to a node named "Node 1", the
dictionary file must contain the line: Node_1 10 |
|
| All the images associated to the network such as background image and node images are now included in the network xbl file. | |
Graphs |
|
The results can be displayed along four modes:
|
|
Node editor |
|
| The node editor has a new interface. All the
properties of a node can be directly edited by selecting the convenient
thumbnail.
|
|
| A new mechanism allowing to create classes of
nodes has been added. A class is a subset of nodes with a given name. A node
can belong to several classes at the same time. It is very useful to manage
properties shared by several nodes (costs, temporal indices, colors,
images,...) and to create arc constraints between nodes.
The classes are managed with the node editor:
|
|
| Color, image, temporal index and cost can be
edited in this panel. It is also possible to propagate a property to the
nodes that belong to the same classe.
|
|
|
When there is at least a node with associated values, the expected total
value of the network and the mean value of the nodes having associated
values are displayed below the joint probability.
The values are used quite like Utility nodes. Indeed, an expected numerical value can be obtained by associating an Utility node to each node, except that the modalities without values can't be represented with this kind of node. Thus, these values are used to evaluate the network, to measure the impact of such lever on the quality of the network. However, unlike Utility nodes, these values are not taken in account during the action policies learning. This is these values that are used to compute the Pearson's linear correlation coefficient.
|
|
| The comment can be edited within the node
editor.
|
|
Network |
|
| A new editor allows to create and manage
constraints over the arcs. The created constraints will be taken into
account by the learning algorithms.
You can forbid probabilistic relation, in one or both directions, between two nodes, between a node and a class of node, between two classes of node or between a class of node and a node.
|
|
| A constant editor has been added. These
constants will be used in the formulas that generate the conditional
probability tables. A constant has a type (real, integer, boolean and
string) and a value. The constants are managed by this editor:
To create a new constant you must choose a name that is not already used by a node or another constant, a type and a value. Once the constant is created, its value can be modified and the conditional probability tables will be regenerated according to the formulas that use the modified constants.
|
|
Interface |
|
 |
|
| The tool that allows to search nodes and arcs
takes not only into account the node nameS but also the classes that have
been defined..
|
|
| The node selection tool available from the node
contextual menu allows to select all the nodes that belong to the same
classe.
|
|
| The alignment tool available from the node
contextual menu allows to define a layout where the selected nodes are
equally distributed, horizontally or vertically.
|
|
| The not observable nodes and their monitors are
displayed now with a mauve color in order to be identified immediately.
|
|
| Now, you can display an image instead of the
default node representation.
The chosen image can be propagated to the nodes of the same classes if needed. The images are saved in the network file. We can switch between standard display and images display (if there are images) with the button:
|
|
| An image preview has been added into the file chooser when the selected file is a valid image format. The dimensions of the image are also displayed. | |
| The comments of the nodes and arcs can be
displayed separately.
|
|
| All node's properties can be directly edited
from the contextual menu.
|
|
| The Inference menu has been split into two
menus: the Analysis menu containing all the graphic and report analysis and
the Inference menu containing the batch labeling and inference and the
adaptive questionnaire.
|
|
Security |
|
| When the automatic validation process is used, you can configure the use of a proxy for the Internet connection by giving the login and password for authentication. | |
Monitors |
|
| The joint probability of the network is
displayed at the top of the monitor panel. If the network has a database
associated, the number of cases is also displayed. If the different
modalities of the nodes have associated values, the total value of the
network and its mean are displayed.
|
|
| If a color is associated to a node, it will be
displayed as a border in the corresponding monitor.
|
|
| The time indicator is no more represented by a
node but has been included as an icon at the bottom right of the network
frame. Clicking on it allows to remove the use of the time in the network.
|
|
Learning |
|
| The EQ and SopLEQ algorithms have been completely rewritten and are much more efficient. The learning time has been reduced by 10% in general. | |
| Switching to different completion mode is
faster. The completion methods has been improved during learning. |
|
| A compression rate is available in the console.
This new indicator measures the data compression obtained by the network
with respect to the previous network (usually, the unconnected network).
This rate then not only gives an indication on the probabilistic links that
are in the network, but also the strength of these links. For example, with a database containing two binary variables that are strictly identical, the corresponding network will link these variables and describe in the conditional probability table that the value of the second variable is deterministically defined by the first variable. The compression rate will be then equal to 50%. |
|
Inference |
|
| In the Data menu, you can use the new
Imputation menu that allows to perform imputation on the currently loaded
database. You can choose the values according to the law or according to the
maximum probability. The generated database will be saved in a file.
|
|
| The interactive bayesian updating allows to use
the associated database as a file of observations. This file can then be
used to update the probability distribution of the nodes that have been
declared as "not observable", with respect to the observations that are
interactively read from this file. Whereas the probability distribution of
all the unobserved nodes can be impacted by these observations, we just
update those of the "not observable" nodes after each observation. This mode
displays a new toolbar that allows to perform a step by step updating and
also complete updating over the database:
The button |
|
| The interactive inference allows to use the
associated database as a file of observations. This mode displays a new
toolbar that allows to navigate through the different cases contained in the
database:
The button At each example, the nodes are observed with the corresponding value in the database except if this value is missing or if the nodes are declared as not observable or as target node. The probability distributions of these unobserved nodes are computed and displayed in the monitors. When a node is unobserved but has a corresponding value in the database, this value is indicated in the monitor with the blue sky bar. The joint probability and the corresponding number of cases are also computed again. In the following picture, Cancer is the target node (pink background) and is not observed. The corresponding value in the database is No (blue sky) and corresponds to the value predicted by the network (99,97%). The node TbOrCa is not observed because it is declared as not observable (mauve background) et the corresponding value in the database is False (blue sky). The node Smoking is not observed because the corresponding value is missing in the database:
This mode allows then to see interactively the behavior of the network and to check its validity. |
|
| The batch inference has been added in order to infer the probability distributions of the nodes declared as "not observable" based on the cases that are describe in a database. The batch inference process can be interrupted at any time without loosing the computed data. The already generated data are saved in the output database. | |
| The batch labeling process can be interrupted at any time without loosing the computed data. The already generated data are saved in the database. A node with "not observable" cost is not observed, even if its values are in the file. | |
| The gain curve has been extended ("Performance
of the network" analysis toolbox) in order to automatically analyze the
expected economical gains with the evaluated model. These computations
follow the definition of the unit costs corresponding to the treatment of
each individual (x-axis), of unit gains corresponding to each positive
answer (y-axis), and finally of a target population's size. The economical
gain is then defined as the difference between the profit corresponding to
the treatment of x% of the population and the profit corresponding to the
treatment of the whole population. As the following screen captures shows
it, the result is displayed as a curve (blue curve) and as a gradient of
color (the closer we are to the yellow, the more we are close to
optimality).
The economical parameters can be modified with the following dialog:
|
|
| We have developed a new algorithm to reduce the
complexity of the graph that are too connected to allow the construction of
the junction tree, and then, that prevent exact inference. This algorithm
incrementally simplifies the network structure until the exact inference can
be performed.
After reduction, a report containing all the removed arcs is displayed. |
|
| The association of values to nodes' modalities allows to compute R the Pearson's linear correlation coefficient between two nodes linked by an arc. If the modalities don't have associated values, default values are defined in order to compute R (from 0 to n-1 for a node with n modalities). The thickness of an arc is directly proportional to the absolute value of R, its color represents the sign of R (blue if positive and red otherwise). The exact value of the correlation for each arc is temporary displayed in the comment of the arc. The Pearson's correlation has also been added to the relationships analysis report. | |
| In Validation mode, the network can be
displayed without the head of the arcs in order to avoid any erroneous
causal analysis of the direction of the arrows. This option is activated by
pressing the convenient button in the toolbar:
|
|
Settings |
|
An option has been added in the Settings to
choose the behavior of the software when a node has been created:
|
|
It is possible to set default values for some
options of the importation wizard:
|
|
| It is possible to modify the parameter of the
network complexity reducer algorithm. Based on the available RAM memory, one
can increase or decrease the reduction rate.
|
|
allows to come back to the first example of the database and to reset the
probability distributions of the "not observable" nodes. The button
performs an
updating from the current index to the last in the database. This process
can be stop while running by clicking on the red light of the Status bar.
The button 
steps
to the next example. The text field indicates the index of the current
example. It is possible to enter an index in the field to perform updating
from the current index to the new index. If the new index is lower than the
current one, the probability distributions are reset and the updating goes
from the index 0 to the specified one. The button
validates the
updated conditional probability tables. The button
stops the
interactive updating and reinitialize the conditional probability tables. It
also removes all the observations.
goes to the
previous example if it is possible and the button
