News
|
|
|
www.bayesia.com does not fully support your browser (Internet Explorer 6).
We suggest upgrading to IE 7 or downloading Firefox for a more enjoyable web experience.
We suggest upgrading to IE 7 or downloading Firefox for a more enjoyable web experience.
Release of BayesiaLab 4.5
December, 01 2008
The main new features of this release are:
- Filtered/Censored States: this new feature allows to rigorously take into account the variables that are contingent on context (usage/configuration). As before, to deal with this kind of variable, we add a new state (eg. NA) that indicates that the variable can be not applicable in particular situations. However, we modified our learning algorithms to implement a "local select" in order to evaluate the probabilistic relations on "compatible observations" only, i.e. those implying no Filtered states.
- Network Comparison (Tools menu): this tool allows comparing the structure of two Bayesian networks (or their equivalence classe). It creates a report summarizing the differences (in terms of arcs, edges and V-Structures). It also gives access to a new Graphical Tool that creates a colored graphical structure representing all the differences between the networks.
- Arc Confidence Analysis (Tools - Cross Validation menu): by using the JackKnife method, this tool allows computing a frequency value that measures how robust each arc is. A report is generated to summarize the comparison of all the learned Bayesian networks with the reference one. This report gives access to the Graphical Tool to visualize the synthesis structure and all the networks. It is also possible to automatically extract the Bayesian network based on user defined frequencies.
- Targeted Cross-Validation (Tools - Cross Validation menu): based on the K-folds method, this tool allows to measure the performance of a set of networks to predict the target node. It generates the classical BayesiaLab's Target Evaluation window with a tab for each learned network (with occurrence matrices, gain, lift and ROC curves), and another tab to synthesize the results (mean performance, statistics on the nodes used in each network). A report is also available with the synthesis of the results and an arc confidence analysis. This report gives access to the Graphical Tool to visualize the different networks, and allows to automatically extract the Bayesian network based on user defined frequencies
- Fixed Probabilities: the tool for entering evidences as probability distributions allows now to fix these distributions. The likelihoods then are recomputed after each new pieces of evidence gathered on the other nodes (hard evidences, likelihood distributions, and fixed probability distributions) so that the fixed probability distributions of the nodes remain the same as initially defined.
- Evidence Scenarios: set of evidences (hard evidences, likelihood distributions and fixed probability distributions) can be associated to a network. These scenarios can then be used for interactive inference, interactive updating, and all the batch processes. They can be externally defined through a text file, or directly defined by using the monitors. A comment can be associated to each evidence scenario. It is displayed in the network status bar during the interactive use to identify the current scenario.
- Node Exclusion (Node Contextual menu): it is now possible to temporarily exclude some nodes in order to ignore them during structural learning.
- Data saved with the network: The data associated with the network can be saved in the network file. When a loaded network contains a database, the database is automatically loaded too (unless the user explicitly specifies not to do so);
- and, as usual, increased ergonomics of the interface and performance improvements.