New parameter learning options

The main change here is in the parameter learning methods.  The parameters that are learned at first (i.e., if there is no evidence) are the distributions that are found directly in the data. I had to create or significantly modify several BNT files for this.

If there is evidence, the parameters are learned using a Dirichlet prior. This only required a couple of small changes to the BNW parameter learning files.

1 files changed, 20 insertions, 5 deletions

diff --git a/sourcecodes/help.php b/sourcecodes/help.php
index 683eeebe..92e44d67 100644
--- a/sourcecodes/help.php
+++ b/sourcecodes/help.php
@@ -96,7 +96,7 @@ BNW includes a structural constraint interface that provides users with options
 </tr>
 <br>
 <tr><td>
-  <p align="justify"> After structure learning is completed, BNW automatically performs parameter learning of the network model using the Kevin Murphy's <a href=http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.25.1216>Bayes Net Toolbox</a> (<a href=https://github.com/bayesnet/bnt>BNT</a>) and displays the network model. Discrete variables in the network are displayed as bar charts and continuous variables are displayed as Gaussian distributions. The networks can be used to make predictions after clicking on a node and entering a value for that variable. Specifically, click on either the blue bar for a discrete node or the blue line for a continuous node to bring up a pop-up box that can be used to enter a value for the node. After submitting a value for the variable, the distributions of the other nodes in the network will change, allowing for visualization of the impact of setting the variable to the given value. The distributions after the entered value is considered are shown in red, while the original distributions are shown in blue. The node for which data was entered is outlined in red.<br><br>
+  <p align="justify"> After structure learning is completed, BNW automatically performs parameter learning of the network model using the Kevin Murphy's <a href=http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.25.1216>Bayes Net Toolbox</a> (<a href=https://github.com/bayesnet/bnt>BNT</a>) and displays the network model. BNW has been recently been updataed and Dirichlet prior distributions are now used during parameter learning.<br><br>Discrete variables in the network are displayed as bar charts and continuous variables are displayed as Gaussian distributions. The networks can be used to make predictions after clicking on a node and entering a value for that variable. Specifically, click on either the blue bar for a discrete node or the blue line for a continuous node to bring up a pop-up box that can be used to enter a value for the node. After submitting a value for the variable, the distributions of the other nodes in the network will change, allowing for visualization of the impact of setting the variable to the given value. The distributions after the entered value is considered are shown in red, while the original distributions are shown in blue. The node for which data was entered is outlined in red.<br><br>
 Two prediction modes are available in BNW: evidence and intervention. In the evidence mode, entered values will alter the distributions of the other variables in the network, but will not alter the network structure. In intervention mode, the intervention alters both the distributions of the network variables and the network structure. Specifically, the intervened variable becomes independent of its parents. Evidence mode is appropriate when making predictions of other network variables after the value of one variable in the network is observed, while intervention mode is appropriate for predictions after experimental interventions that alter the values of some variables in the network. Further discussion of the difference between evidence and intervention prediction modes is given on the BNW <a href="faq.php">FAQ</a> page.       
 </td></tr>
 </table>
@@ -133,7 +133,7 @@ Two prediction modes are available in BNW: evidence and intervention. In the evi
  <b> 1.</b> If a variable contains 3 or fewer different values, the variable is considered to be discrete.<br><br>
  <b> 2.</b> If a variable contains more than 20 different values, the variable is considered to be continuous.<br><br>
  <b> 3.</b> If the ratio of the number of different values for a variable compared to the number of cases in the data set is large, the variable is considered to be continuous. Specifically, if this ratio is 1/3 or larger, the variable is considered to be continuous.<br><br>
- <b> 4.</b> If none of the first three rules apply, the data set is inspected to determine if any of the values for the variable contain a period (.). If at least one value contains a period, the variable is considered to be continous; otherwise, the variable is considered to be discrete.<br><br>
+ <b> 4.</b> If none of the first three rules apply, the data set is inspected to determine if any of the values for the variable contain a period (.). If at least one value contains a period, the variable is considered to be continuous; otherwise, the variable is considered to be discrete.<br><br>
   Users can examine whether or not BNW has correctly loaded input data files and classified variables by clicking on "View uploaded variables and data" on the left-hand menu after uploading a dataset. We believe that BNW should correctly classify variables in most cases, but users may occasionally need to add or remove a period to the data of some variables. <br><br>
     An example input data file for a file with 5 variables is given below. The network contains 2 discrete (Disc1 and Disc2) variables, which are given in the first two columns of the file, and 3 continuous variables (Cont1, Cont2, and Cont3). Disc1 is a discrete variable with two states (1 and 2), while Disc2 has two states (A and B). Although the samples of Cont2 are integral values, we wish to deal with this variable as continuous, not discrete. Therefore, the value of Cont2 for the first sample is given as '3.0' instead of '3' so that one of the values of Cont2 contains a '.', helping to ensure that Cont2 is interpreted as a continuous variable.
 </p>
@@ -209,12 +209,27 @@ Two prediction modes are available in BNW: evidence and intervention. In the evi
 <tr><td>
 <p align="justify"> BNW has recently been updated to add features and improve the user experience. These updates have included improving the network model visualizations and allowing users to more quickly load large data sets. Additionally, in a change that is invisible to users, BNW now uses Octave to perform parameter learning with the Bayes Net Toolbox. <br><br>
 Major changes and new features that have been added to BNW include:<br>
-<br><b>1)</b> Increased flexibility in formatting of uploaded data files. One major change is that BNW now allows for users to upload data files in which discrete variables have alphabetic values. For example, a file containing a node for the genotype of BXD mice can now have values of B and D; the values do not have to be recoded as integers. A full description of the proper format for input files in BNW considering these changes can be found <a href="help.php#file_format">here</a>.<br>
-<br><b>2)</b> Added "View uploaded variables and data" button to allow users to ensure that uploaded data sets are loaded and parsed correctly. After users upload a data set, clicking this button provides the ability to view either the uploaded data file directly or view a variable description file that shows how BNW has parsed the data. The variable description file lists the number of variables and cases (e.g., individuals or samples) in the data file. It also indicates whether each variable is discrete or continuous and the criteria that was used to make this determination. For discrete variables, the possible states of the variable are provided. For continuous variables, the mean and standard deviation of the variable is provided.<br>
-<br><b>3)</b> Added "View parameters" button to allow users to quantify network parameters. Users can now view the parameters of the network models considering the original data set that was uploaded by the user or the predicted parameters considering the evidence or intervention that has been entered by the user. This feature allows users to quantify predictions using the Bayesian network model.<br>
+<br><b>1)</b> Parameter learning settings have been modified. Specifically, network parameters (e.g., the distributions of states for discrete variables and means and standard deviations of Gaussian distributions) are now learned using Dirichlet prior distributions. In our testing, this has had a minimal impact on the parameters of most networks, but has helped reduce the impact of cases with rare combinations of states on the predicted distributions of some networks. In the original version of BNW, no priors were used.<br>
+<br><b>2)</b> Increased flexibility in formatting of uploaded data files. One major change is that BNW now allows for users to upload data files in which discrete variables have alphabetic values. For example, a file containing a node for the genotype of BXD mice can now have values of B and D; the values do not have to be recoded as integers. A full description of the proper format for input files in BNW considering these changes can be found <a href="help.php#file_format">here</a>.<br>
+<br><b>3)</b> Added "View uploaded variables and data" button to allow users to ensure that uploaded data sets are loaded and parsed correctly. After users upload a data set, clicking this button provides the ability to view either the uploaded data file directly or view a variable description file that shows how BNW has parsed the data. The variable description file lists the number of variables and cases (e.g., individuals or samples) in the data file. It also indicates whether each variable is discrete or continuous and the criteria that was used to make this determination. For discrete variables, the possible states of the variable are provided. For continuous variables, the mean and standard deviation of the variable is provided.<br>
+<br><b>4)</b> Added "View parameters" button to allow users to quantify network parameters. Users can now view the parameters of the network models considering the original data set that was uploaded by the user or the predicted parameters considering the evidence or intervention that has been entered by the user. This feature allows users to quantify predictions using the Bayesian network model.<br>
 <br>After performing structure learning and viewing their network model, users now have the ability to click a "View parameters" button on menu to the left of the model structure. This button provides a link to a file containing the original parameters of the model. For discrete nodes, the fraction of cases for each possible state in the variable is provided. For continuous nodes, the mean and standard deviation of the Gaussian distribution that best fits the data in the variable are provided.<br><br>
 If users have made predictions using either the evidence or intervention modes, a link to a file containing the parameters of the network considering the entered evidence or intervention is provided. The file lists the predicted fraction of states for discrete nodes and predicted mean and standard deviation of the Gaussian distribution for continuous nodes. Nodes for which evidence or intervention has been entered are also noted in the file.    
 <br>
+<br><b>5)</b> Added a "Use a network ID to return to a network" on the BNW home page. This button allows users to enter a network ID to return to a previously generated network model. The network ID can also be used to share the network model with collaborators. The network ID is a three character string that is listed in left menu of a BNW network page.
+<br> 
+<!-- 
+<br><b>6)</b> Added a "Cross validation" feature to test networks using leave-one-out cross-validation. This button is available on the left menu of a network page. <br><br>
+ Clicking "Cross validation" will open a pop-up window. Users can then enter the name of the variable that they want to test. After submitting the variable, the pop-up window will close. Users can then click the "Cross validation" button on the network page again to determine if their cross-validation calculation has completed.
+If it has completed, the cross-validation output contains the following information:<br><br>
+For discrete variables, the predicted likelihood of each state for the left-out case, given the values of its parent variables in the network, is provided.<br><br>
+For continuous variables, the predicted mean and standard deviation of the left-out variable, given the values of its parent variables in the network, is provided.<br>
+<br><b>7)</b> Added a method to make predictions on a test data set that can be accessed using the "Make predictions on test data set" button on the left menu of a network page. <br><br>
+Clicking this button will open a pop-up window. Users can then upload a file containing a test data set. The format of the file should follow the format of the input data file with two exceptions:<br><br>
+<b>a)</b> The first line of the file should containing the name of the variable that should be predicted.<br><br>
+<b>b)</b> "NA" can be used for missing data. The test data file can contain missing data. If data for a variable of a given sample or case is not known, an "NA" can be entered in the input file. Data can be missing for the variable that is to be predicted or for the variables that are to be used as predictors.<br><br>
+ -->
+<br>
 </table>