• Joe_Black
  • Junior Member
  • Email
  • Posts 11
  • Reputation 1
  • Member Since
  • Last Active
  • Name Joe_Black
All Posts Topics Started Likes
how to create a dynamic network and import temporal resolved data
Hi Andrea:

I have attached an example of a csv file and some screenshots for the workflow.  The data is random.  Only the score column is time-varying.  

You have to make sure to have an ID column to delimit the different time series. 


Then select the option "Use Identifiers to Define Time Series" under network temporalization (assuming that you have more than 1 time series).  If you have only 1 time series then this is not necessary. 


Also if you only have one variable and several time series you can also chose to organize your data such that different times are the rows and the individual series are in columns.  See the S&P 500 stock market bayesian network example as reference.

Hope this helps,
0 0
This forum
Hi.  I'm having some issues seeing some of the posts in this forum.  Not sure if that is due to my browser config or otherwise but below is an example of the problem.  I get this sort of characters in many places:  [*[redface]odppaqu].

Also you will see the first post in below topic is probably there for original testing purposes and may now need to be deleted.  Thanks!


Latent variable (or Factor) is a variable that is not contained in the original dataset. It is also frequently referred to as a Hidden variable. Such nodes are always shown in white in BayesiaLab.You can manually create Latent variables can by adding a node to the current network, and then defining its number of states. Furthermore, you will then need to manually define the connections with the other variables and use either Probabilities Learning (Learning | Parameter Estimation), which uses the Expectation-Maximization (EM) algorithm, or manually define the distributions via the Node Editor.The other way to create Latent variables is to use the BayesiaLab's automatic Data Clustering algorithms (Learning | Clustering) on a selected set of nodes:[*[redface]odppaqu]Data Clustering: the objective of this EM algorithm is to create a Latent variable that summarize the joint probability distribution defined by the selected nodes. This algorithm can be used to search the optimal number of states.[/*:m[redface]odppaqu][*[redface]odppaqu]K-Means: this EM algorithm consists in partitioning the (numerical) observations corresponding to the selected nodes into K clusters in which each observation belongs to the cluster with the nearest center/mean. The number of centers K is defined by the user. We have illustrated this algorithm for [url=https://forums.bayesialab.com/viewtopic.php?f=10&t=34[redface]odppaqu]data discretization[/url[redface]odppaqu].[/*:m[redface]odppaqu][*[redface]odppaqu]Binary Clustering: this algorithm is a deterministic tool for generating binary (or boolean) latent variables by applying a deterministic function on a set of selected nodes. [/*:m[redface]odppaqu]The Multiple Clustering tools can also be used for creating Latent variables per node Class. To be used for the automatic induction of Latent variables, such Classes of nodes must follow the format, [Factor_i]Classes can be set manually, e.g. utilizing expert knowledge, via the Class Editor in the Contextual Menu. Alternatively, Classes can be automatically defined using BayesiaLab's Variable Clustering tool.
1 0
Parameter Estimation
Thanks Dan.  I get it now - if you build a network manually (or add nodes to an existing network manually) Parameter estimation will learn the network weights from the data without changing the structure of the network.
0 0
Augmented Markov Blanket
You can add whatever knowledge you extracted from unsupervised learning to your data file and then use that modified input file with supervised learning algos.  For instance you could add some clusters/factors or manual nodes to your dataset.

Also please take a look at this very useful explanation showing the use of fixed arcs and virtual datasets to express prior knowledge:

0 0
Supplementing a learned network with utility and decision nodes
After you add the new nodes manually you can re-associate the dataset ensuring that you have new columns in the dataset for the manually added nodes if need be.
0 0
Ignore Columns in Data Import
Concerns - import dataSometimes you have several columns which can serve as identifiers - not just one. Currently if you select several columns as identifiers, BL will merge them into a single column during output. It would be very useful to have the option to keep them separate. So these columns would not be discretized but just passed from inputs to outputs.Example: One column is Date and the other is Car ID. Together they uniquely identify each row. Mixing them works but when you export results (through inference for instance) it gets difficult to work with the data if you want for instance to sort by date only or car ID only.
0 0
Automatic naming of latent factor variables in multiple clustering
I was able to find how to do this. In fact the default behavior is for this to happen in the comments (show comments shortcut: M). However, one must be careful to check the option for "compute manifest variable contribution to latent variables".
0 0
Automatic naming of latent factor variables in multiple clustering
After variable clustering and then multiple clustering is there a way to get the newly created factors to reflect the name of the most important observed Manifest variable in the factor (based on mutual info for instance)?Thanks
0 0
Automatic Complexity Reducer
I find myself having to crash BayesiaLab often after it won't respond during a passage to Validation mode. I find the automatic complexity reducer feature to be quite helpful and believe it would be great to have it available in modelling mode with some kind of threshold parameter. Otherwise it's difficult to know how to manually simplify a network without already being in Validation Mode (so you can run analysis reports on Node and Arc force).Additionaly it would be great if you could abort the passage to validation mode - that would be a step up from having to crash the software and lose your work as well as provide us the ability to get back into modeling mode to try and reduce complexity. Thanks
0 0
Parameter Estimation
Hello. What does Learning Menu -> Parameter Estimation do?Thanks
0 0
Crash/Error reporting
Hello. Whenever Bayesialab crashes we get the option to send an email. Unfortunately I have never been able to send an error report because I don't have a mail client installed on the PC on which I have Bayesialab installed. In fact I no longer use email clients as I have office365. Many people also only use Gmail web interface.To allow all those people to send you crash reports, would it be possible to send the message directly from BayesiaLab? A service like Amazon SES (simple email service) comes to mind.Many thanks
0 0
count post selected

Add a Website Forum to your website.