iCAUSALBAYES USER MANUAL
Contents
1. Since we have already created the Explanatory Variables for Melanoma we can ignore the Prior Distribution and set these up right away Tap Explanatory Variables swipe to make both Sunscreen and Melanin Index explanatory variables and enter 823 as the coefficient for Melanin Index and 415 as the coefficient for Sunscreen We ll use the Error Term Mean and Standard Deviation defaults here too Tap Done in the top right corner to close the menu and add the variable to the network lt Edit Variable Explanatory Variables Done EXISTING EXPLANATORY VARIABLES Melanin Index Index 0 823 Sunscreen Mls per Person 0 415 ERROR TERM Mean 0 Standard Deviation 1 POTENTIAL EXPLANATORY VARIABLES Now we have arrows from Sunscreen and Melanin Index to Melanoma This is something that might be learned by an ordinary regression analysis and it says that the incidence of melanoma varies with both Melanin Index and Sunscreen use Arrange and resize the variables so you can see everything Save the network if you like by tapping Save As and entering a suitable name At this point you may also wish to set the Display Options You can set Display Options for the network as a whole or for each node individually Our small network is complete Melanoma Incidence per 100 u 0 Scale 1 Sunscreen Mis per Person s Melanin Index Index N Scale 2 Scale 1 OBSERVING AND DOING Ther2. POTENTIAL EXPLANATORY VARIABLES Melanin Index Index Tap the button to make Melanin Index an Explanatory Variable lt Edit Variable Explanatory Variables Done EXISTING EXPLANATORY VARIABLES POTENTIAL EXPLANATORY VARIABLES Make _ xplanatory p Melanin Index will move from the Potential Explanatory Variables to Existing Explanatory Variables and an Error Term will appear lt Edit Variable Explanatory Variables Done EXISTING EXPLANATORY VARIABLES Melanin Index Index 0 8 ERROR TERM Mean 0 Standard Deviation 1 POTENTIAL EXPLANATORY VARIABLES Enter 0 8 as the coefficient for Melanin Index Leave the Error Term Mean and Standard Deviation at their default values Tap Edit Variable in the upper left corner to return to the previous screen Back in the Edit Variable view you will notice that the Prior Distribution is gone Tap Done Sunscreen Mls per Person s Melanin Index Index Scale 2 Scale 1 Now there is an arrow from Melanin Index to Sunscreen We have changed the meaning of both variables The arrow indicates that Melanin Index affects Sunscreen use The coefficient of Melanin Index tells us how people with a lower Melanin Index use more Sunscreen This is something a Bayes Net learning algorithm might discover Our third and last variable will be called Melanoma For Units we ll use Incidence per 100 Once again we ll leave Upper and Lower Bounds on Auto
3. respectively The calculation is as follows P M S P M DS P D S P M ADS P D S DOING We can approach this in another way instead of Observing what happens with different distributions of Sunscreen and Melanin Index we can try Doing something so that Sunscreen Is not affected by Melanin Index A real world analogy would be an experiment or a law that enforced Sunscreen use regardless of skin pigmentation so that Melanin Index Is eliminated as a confound Some people will find the idea of a sunscreen law objectionable and anyone who enacted such a law would have to contend with problems related to compliance and enforcement But never mind all that This is just a what if scenario What if we enforced sunscreen usage What if sunscreen use had nothing to do with skin pigmentation Tap the eye icon a second time to put the Sunscreen variable into Doing mode indicated by a wrench icon Doing means we intervene to fix the value of Sunscreen in such a way that Melanin Index has no effect on it you will notice that the arrow from Melanin Index to Sunscreen becomes a dashed arrow Now drag the observed value line around As Sunscreen use goes down Melanoma increases aS Sunscreen use increases Melanoma rates go down Melanin Index is unaffected because we are manipulating Sunscreen use directly there is no relationship between Sunscreen use and Melanin Index or skin pigmentation lf you perform these manipulations of Sunscr
4. so the colored area can be seen Choose a Background Image Selecting this option opens a new menu screen that allows you to choose an image to set as the background of a node A new image can be selected by choosing from your device s photo library or by capturing an image with the camera The Stretch and Scale options alter the way the image fits in the node If Stretch is selected the entire image will be visible in the node although the aspect of the image may be distorted due to the stretching If the Scale option is selected the aspect of the image is maintained but there in no guarantee the entire image will be shown 14 lt Variable Display Options Done Stretch Scale Take a Photo Choose a Photo Using the Camera From Photo Library NETWORK DISPLAY OPTIONS To change the display options for a network tap Display Options in the upper right of the main window Carrier F l 11 47 AM l 100 m Back Save SaveAss Email Sunscreen Example Display Options i 15 Cancel Network Display Options Done NETWORK Shadows C Arrow Display Mode Simple Network Background Color a Arrow Color NODE Orientation Landscape Portrait Header Color GRAPH Ghosting Mean Line Display Mode Line Both Background Color Line Color The display options available for the network are the same as those for individual variables listed on page 13 and 14 with the addition of Shadows Turn shado
5. ICAUSALBAYES USER MANUAL INTRODUCTION You can use this app to build a causal Bayesian network and experiment with inferences We hope you ll find it interesting and helpful We expect most of our users will be Al students with a basic knowledge of probability and Statistics You don t need a deep understanding of regression to use the app but you do need to be aware of some terminology and the basic regression equation a response variable RV depends on one or more explanatory variables EV plus an error term ET The basic regression equation then is RV C EV Ci EV C EV ET CREATING A NETWORK A TUTORIAL There are two ways to create a network 1 Build a network in the app recommended 2 Create and import an xml or zip file describing the network IMPORTING A NETWORK This app is capable of importing a network by opening a properly formatted zip or xml file from another app By pressing the Email button the app can generate and send zip files of the required format After receiving the email select the attachment and choose Open in then select iCausalBayes It is possible to create your own network by writing or editing an xml file with a text editor then zipping that file along with a folder containing any images the network should use if any are used To see the format of these files use the app to email any existing network to yourself then unzip the attachment We recommend using the ap
6. e is a long standing controversy over sunscreen There are those who say that people who use sunscreen end up spending more time in the sun and that the increased sun exposure negates and even outweighs any protection sunscreen might provide Some argue that sunscreen users have lower levels of vitamin D which protects against skin cancer as well as other cancers Others maintain that some sunscreen ingredients increase the risk of melanoma and other cancers On the other hand most dermatologists and cancer organizations say that if you want to reduce your risk of melanoma you should use sunscreen consistently Suppose the observational data suggests that sunscreen use doesn t much affect the probability that you will develop melanoma Now let s investigate OBSERVING Tap the icon on the upper right corner of the Sunscreen variable to put it into Observing mode you know you are in Observing mode when you see an eye icon in the upper right of the variable A blue line representing an observed value will appear in place of the distribution curve Leave the other two variables in Graph mode indicated by a sparkline Now drag the line around to see the values of the other two variables for a given value of the Sunscreen variable You will see that the distribution for Melanoma doesn t move very much It would seem that Sunscreen use doesn t much affect the incidence of Melanoma However you might also observe somethi
7. een while Observing different values for Melanin Index you will notice that while increased Sunscreen use always lowers melanoma rates the mean for Melanoma Is higher for people with lighter skin and lower for people with darker skin Melanin Index in this scenario has no effect on Sunscreen but it is still a factor when it comes to the observed incidence of Melanoma The lighter the skin the greater the susceptibility to Melanoma However Sunscreen use always reduces the incidence of Melanoma 10 Again we can conclude that Melanin Index is aconfound and that Sunscreen use reduces the incidence of Melanoma But this time we were able to reach that conclusion much more quickly and directly How would we calculate P M S in this case The calculation above where we are Observing is not correct here In this scenario where we manipulate Sunscreen in such a way that it does not depend on skin pigmentation it would an error to multiply PLM DS by P D S There is a relationship between Melanin Index or in binary terms D and D and Melanoma rate but Sunscreen use is independent of skin pigmentation So in the Doing case P M DS should be multiplied by the proportion of persons who are dark skinned and P M DS by the proportion that are light skinned which gives us this 11 P M S P M DS P D P M ADS P D In graphical terms the difference between the Observing calculation and the Doing calculation is the presence Observ
8. ing or absence Doing of an arrow between Melanin Index and Sunscreen MENU ITEMS NAVIGATION While in any menu screen pressing the Done button located in the top right of the menu bar on will apply any changes made since opening the menu and return you to the network The button on the upper left of the menu bar will take you back to the previous screen If the button in the upper left reads Cancel the menu will be closed without applying the changes made COLOR AND GRAPHICS To change the display options for a variable tap and hold the variable then tap Color and Graphics Cancel Edit Variable Done GENERAL Name Sunscreen Unit Mis per Person GRAPH Lower Bound 4 Auto q Upper Bound 4 Auto C PRIOR DISTRIBUTION Mean 0 E Standard Deviation 1 Explanatory Variables Color and Graphics ee BY Mas lat le 12 VARIABLE DISPLAY OPTIONS Turn Use Network Settings off to over ride the network settings for this variable For network settings see NETWORK DISPLAY OPTIONS below lt New Variable Variable Display Options Done NETWORK Use Network Settings NODE Orientation Landscape Header Color GRAPH Ghosting Mean Line Display Mode Both Background Color Line Color Area Color AAAS a VIA E Use Background Image Greyscale Background Choose a Background Image If you over ride the network settings you wi
9. ll have access to a number of options that affect the look of each variable Each option is explained further below Orientation Each node can have either a portrait or landscape aspect ratio Nodes in portrait mode show a bar graph rather than the normal distribution curve displayed in nodes with landscape orientation Header Color The color of the top and bottom of the node the area where the name of the variable is displayed Ghosting When Ghosting is turned on the prior distribution for a variable appears as a ghost behind any posterior distribution 13 Mean Line Show or hide a vertical line positioned at the current mean of a variable This line is only visible for variables in landscape orientation Display Mode Choose whether the graph is shown as a line a filled area or both Background Color The color of the node behind the distribution curve This color is only shown if a background image is not being used Line and Area Color The colors of the distribution curve Use Background Image Turn this option on to show an image in the node behind the distribution curve Greyscale Background When this mode is turned on only the area of the image that is under the distribution curve is shown in color the rest of the image is shown in black and white as shown below This mode is only available for variables in landscape orientation Note be sure to set the graph Display Mode to Line
10. ng we mentioned earlier when we set up Melanin Index as an explanatory variable for Sunscreen when Sunscreen goes up Melanin Index goes down when Sunscreen goes down Melanin Index goes up It appears that people with darker skin use less Sunscreen overall while people with lighter skin use more intuitively this makes sense So what happens if we observe the effect of different values for Melanin Index Tap to Observe the Melanin Index variable and drag the observed value line to the left We are observing values for people with little skin pigmentation Now drag the line for Sunscreen to the right It appears that for people with little skin pigmentation as Sunscreen usage increases the incidence of Melanoma goes down Conversely as Sunscreen usage decreases the incidence of Melanoma goes up Let s see what happens when Melanin Index has a higher value Drag the line to the right and then observe the effect of different values for Sunscreen ON Melanoma Again more Sunscreen means less Melanoma and less Sunscreen means more Melanoma though the mean has moved to the left It would seem that Melanin Index is a confound and that Sunscreen use reduces Melanoma rates It will be useful to consider how the probability of Melanoma M given a particular observed value of Sunscreen S is calculated For the sake of simplicity we ll pretend we re dealing with discrete binary variables D and D represent dark skin and light skin
11. nscreen variable on your screen The Sunscreen node shows that MIs per Person is approximately normally distributed with a mean of 0 At this point this node says on average people use x mls of Sunscreen Sunscreen Mis per Person s Scale 1 Now let s add another variable This one is will be an Explanatory Variable for Sunscreen We ll call it Melanin Index This variable is a measure of skin pigmentation as determined by a reflectometer For the Unit we ll use Index Again we ll leave the Upper and Lower Bounds on Auto Again we ll use the defaults for the Prior Distribution Position the variables as you like Use the familiar two finger pinch inside a node to make it bigger or smaller or outside to expand or contract the network as a whole Now welll make Melanin Index an explanatory variable for Sunscreen Touch and hold inside Sunscreen to edit the variable Sunscreen Mis per Person s Scale 1 In the Edit Variable view tap Explanatory Variables Cancel Edit Variable Done GENERAL Name Sunscreen Unit Mis per Person GRAPH Lower Bound 4 Auto Upper Bound 4 Auto PRIOR DISTRIBUTION Mean 0 Standard Deviation 1 Explanatory Variables Ape Color and Graphics DAGCA EUEI In the Explanatory Variables view swipe left to expose the Make Explanatory button lt Edit Variable Explanatory Variables Done EXISTING EXPLANATORY VARIABLES
12. p to create networks as writing your own xml document can be very error prone BUILDING A NETWORK IN THE APP We ll use the same 3 node throughout this manual This network comes bundled with the app but here we ll build it from scratch Tap the button at the top right of the home Your Networks screen to create a new network iPad 12 05 PM 98 Gas Your Networks Edit Success Example Last Modified May 29 2014 11 00 AM Sunscreen Last Modified May 27 2014 7 42 PM Touch and hold anywhere on the screen to create a new variable at that location Use the New Variable view to configure the variable Cancel New Variable Done GENERAL Name Sunscreen Unit Mis per Person GRAPH Lower Bound Auto Upper Bound PRIOR DISTRIBUTION Mean 0 Standard Deviation 1 Explanatory Variables Color and Graphics Give your variable a name Sunscreen For the Unit enter Mls per Person You can enter a Lower and Upper Bound for the graph or leave these one or both on Auto This determines how many Standard Deviations will be displayed for the variable We ll leave these on Auto You can always come back and change this In fact you can always come back and change anything For the Prior Distribution leave the Mean and Standard Deviation at 0 and 1 respectively Tap Done in the top right of the New Variable view to apply these changes You should see the Su
13. ws behind nodes on or off Arrow Display Mode Set the Arrow Display Mode to Simple or Color Coded If you choose Color Coded the intensity of the color of an arrow will reflect the strength of the relationship between two variables Network Background Color and Arrow Color Select the color of the background and arrows between variables 16 SAVE SAVE AS and EMAIL Tap Save located in the upper left of the main window to save any changes that have been made to the network This option will not be available until the network has a name Tap Save As to save a network under a new name The first time a network is saved Save As must be selected in order to set a name for the new network Tap Email to email a zip file containing the xml file describing a network and any images that are used in the network The Email button will not be available until the network has been saved under a name using Save As Note any unsaved changes will not be sent in the email Carrier gt 11 47 AM i 100 mmm Back Save SaveAs Email Sunscreen Example Display Options i a HELP Tap the icon in the upper right corner of the main window to review the gestures described above iPad 1 34 PM 3 CD Back Save SaveAs Email Sunscreen Example Display Options 17
Download Pdf Manuals
Related Search