FICO Scorecard Professional FAQ

Document created by lshahbazian@fico.com Advocate on Feb 7, 2018Last modified by Makenna.Brei on Feb 8, 2018
Version 2Show Document
  • View in full screen mode

Q: How do I change which users have access to my project?

A: Click on the project name and select user names (hold down 'Ctrl' key to select multiple names).

 

Q: How do I cut down on the project list I am viewing?

A: Use the radio buttons on the bottom of the screen to view just the projects you own or have access to. You can also remove zipped projects from the list by selecting the 'active' status radio button. You can also sort by the column headings (e.g. Last Use or Owner) to better navigate to those projects you're interested in.

 

Q: I imported data but forgot to change my performance variable data type to discrete. Can I do so after completing the import process?

A: No, but you can create a copy of the variable using the copy operator in the transformation segment

of the explore tab. Before running the data, change the data type of the copy by clicking on the variable name.

 

Q: I forgot to change the role of my performance variable during import. Can I change it later?

A: Yes, you can change the role of a variable at any time by clicking on the variable name.

 

Q: How do I generate new variables?

A: Click on the 'segment' drop down, and select the 'transformation' segment. This is where your generated (or transformed) variables live.

 

Q: I am creating a new variable, but have a long list of input variables to choose from. Is there an easier way to find the variable I want?

A: In the explore tab, the variable selection list does not have a search capability. However, you can click on a variable name in the list, then type a letter (or series of letters) and Xeno will jump to the next variable that starts with that letter(s). If that's not the variable you want, type the letter again and it will go to the next one. Continue until you get to the desired variable.

 

Q: I have a numeric variable with missing values. Can I set them to a special value?

A: Yes, you can do this with the apply condition operator. First, for numeric variables, create a variable using the 'isvalid' operator in the 'numeric 1-var' family of operators. This will output a 1/0 value, with 0 for invalid values. Next, create an apply condition operator, and choose the 'variable' radio button to output your original variable, setting the test to be when your 'isvalid' variable is '1'. Then set the next ruleset to output your special value.

 

Q: I am specifying a rule in my apply condition variable, but the variable I want to test on does not show up. Why not?

A: You may need to change the data type before selecting 'add rule'. The default data type is numeric, so if your rule variable is alpha or date/time, it will not appear.

 

Q: I am creating a variable using apply condition, but need several versions of the variable that are slightly different. Is there a way to do this easily?

A: You may copy the apply condition variable and edit it by clicking on the variable name, the variable specification tab, and the edit button. You can then output different variables or values, or change rule logic by using the E button within the rule box.

 

Q: I have a long list of variables. Is there a way to search for specific variables to make navigation through the explore tab easier?

A: Yes, you can search by label, using a unique string for the variable or variables, or you can search by role, variable number, or data type.

 

Q: How do I generate a score variable from a scorecard built in Xeno?

A: Use the apply scorecard operator.

 

Q: I would like to convert the score I built in Xeno into a probability. Is there an easy way to do so?

A: Yes, use the numeric 1-var family of operators, and select the 'ln(odds) to probability' operator.

 

Q: Can I rename a report I already created?

A: Yes, you can click on the report name and rename it.

 

Q: How do I use reports to validate a scorecard built in Xeno?

A: You should have built your original scorecard on a random subset of the data using a filter. Once you apply your scorecard using the apply scorecard operator, you will have a score variable that is populated for the entire set of records on the project. Set up a corresponding filter to capture the remaining random holdout subset, and run a score distribution, comparison, or evaluation report. If you wish to compare your scores on the development and holdout samples in the same report, you can do so by creating conditioned versions of your score or performance (the rule should match your development and holdout samples, respectively). The conditioned scores can be used in a score comparison, the conditioned performances can be used in a score evaluation report.

 

Q: I would like to run a score evaluation report with specific score breaks. How do I do so?

A: Format your score first, then choose the 'format' radio button in the upper left of the score evaluation specification.

 

Q: I'd like to check to see how a score works on multiple subpopulations. Is there an easy way to do so?

A: Create conditioned versions of your score variable. To do so, use the apply condition operator, select the 'variable' radio button and your score variable of interest, specify the subpopulation using the rule box, and leave the default as 'nan'. This will create a variable with your score just on the records of interest. Do this for each subpopulation, and use all the conditioned scores in a score comparison report.

 

Q: I would like to run a report on score deciles. Is there an easy way to format my score into deciles?

A: Yes. Set up a classing on the filter of your choice, and select the score(s) you wish to format. Use a performance outcome that covers the entire population (e.g. 1+0=all), or create a 'dummy' performance variable using a system operation to create a constant (remember to change the role to performance). Choose the 'percentile' classing algorithm, and run the classing. Output the formats onto these score variables by using the format button at the top of the screen. Now you can run a report using these formatted scores.

 

Q: I want to run an acceptance rate report, but nothing shows up in my variable list. Why not?

A: Only formatted variables are available in acceptance rate reports. Format the variables first (you can do this easily by going into a classing with the desired filter and predictors and push out the formats using the format button at the top of the classing screen.

 

Q: How do I download reports for presentation?

A: Use 'alt-print screen' to save the image into a document. You may wish to use an image editor to crop or otherwise edit the image first. All reports can be cut and paste into Excel or other programs as desired.

 

Q: How do I run a report on just a subset of records?

A: The most direct way to do so is to create a filter and write logic to keep only the records you're interested in. Alternatively, you can create variables using apply condition and use the conditioned variables in the report.

 

Q: How do I check my complex filter logic?

A: Run a filter count report.

 

Q: What is the difference between 'run' and 're-count'?

A: Run will generate new classing breaks, recount will apply the existing classing breaks. Usually, recount is only used when you have added performance variables, or if you have copied the classing and want to view additional/different performances but preserve the breaks from the original primary performance variable.

 

Q: After my classing has run, there are variables with 'inf' or 'nan' in the information value. Why?

A: It's likely you have zero counts in some levels, and Xeno cannot calculate the weight of evidence. If you go into the classing options (available in the action drop down) and check the box next to 'Exclude neutral levels from IV calc', Xeno will ignore neutralized levels. This can also happen in classings that you have 're-counted'. In these cases, you may need to manually group or neutralize levels.

 

Q: I see colored bars in my variable-level classing screen. What does that mean?

A: The colors are a visual cue about the volume of data. Red indicates very thin data, grey somewhat thin data, and bright blue indicates a large percentage of data. You can change the parameters that control when these colors show up in the classing options drop down (available in the action drop down).

 

Q: Can I rearrange the levels in my discrete variable classing?

A: Yes, click the arrows to the left of the variable label.

 

Q: I have grouped several levels in my discrete variable, and now my label is followed by '+++'. Can I change that?

A: Yes, you can edit a classing label by clicking on it.

 

Q: I see that I have multiple raw levels classed together. Can I see what's grouped together?

A: Yes, click on the level label and you will be able to see the values and distribution of the raw levels.

 

Q: I have special values in my data. Can Xeno automatically separate them in the classing?

A: Yes. In a new classing, before you run your classing the first time, go into the compression screen, specify the special values and add them to the list. If you have already run the classing, you can change the special values globally and run the classing again. Alternatively, if the special values apply to only a few variables, you can edit the compression settings from within the variable classing (the circle C in the upper left corner) and run just those few variables.

 

Q: I want to split out special values on multiple variables at once. Can I edit the compression setting on multiple variables without running them one at a time?

A: Yes, edit the global compression settings. Back on the classing variable list, click the box to the right of the variables where you want to edit special values, select 'run' from the variable level drop down and click the circle S button to submit. You should change the global settings back to make sure you don't inadvertently apply these settings to other variables or in a copy of the classing.

 

Q: Can I round my classings to remove the decimals?

A: Yes, the scale setting in the compression screen can be used to round variables. The default setting of '2' will give you two decimals of precision. Setting the scale to '0' will round the decimals. If you wish to round to 10's, use a scale of '-1', a scale of '-2' rounds to 100's, etc.

 

Q: I want to round all my balance variables by 10's. Can I edit the scale setting on multiple variables at once?

A: Yes, edit the global compression settings (an entry of '-1' in the scale setting will round to 10's). Back

on the classing variable list, click the box to the right of the variables you want to round, select 'run' from the variable level drop down and click the circle S button to submit. You should change the global settings back to make sure you don't inadvertently apply these settings to other variables or in a copy of the classing.

 

Q: I want to see a specific break in my classing. Can I force a certain break?

A: Yes, you can either do so by 'ungrouping' the desired range, or by setting a variable-level special value by clicking the circle C in the upper left of the classing and running. Alternatively, you can use the “Apply Format” classing option and override the data-suggested classing with a user-specified one.

 

Q: Can I use more than one performance variable in a classing?

A: Yes, you can select up to three. Classing breaks will be optimized based on the first (primary) performance.

 

Q: I have more than one performance variable. Is there a way to see the information values for all performances for all my variables side by side?

A: Run a classing report in the action drop down. When the report has finished, you can view or save it.

 

Q: I have selected my favorite variables and would like to run more classings on the same variables using different filters. Can I do so?

A: Yes, copy the classing using the circle C button on the right. Rename the classing by clicking on the name, and edit the classing by clicking on the icon. Run the classing to apply the changes.

 

Q: I have run a classing, then generated some new predictors. Can I add them to the classing?

A: Yes, go into the classing by clicking on the icon, then click the select button next to predictors to select the new variables of interest. Click the box to the right of the new variables, and from the variable level action drop down at the top of the list of variables, select run. You do not need to run the whole classing again unless other changes (e.g. filter, performance outcome) were made.

Q: I only see 50 variables in my classing list. Can I see more?

A: Yes, change the page size at the bottom of the page.

 

Q: My filter or performance variable has turned red in my classing. Why?

A: They probably contain generated variables, and your data needs to be run. Go to the explore tab and click the run button.

 

Q: What does 'Break and Heal' mean?

A: This is the automated algorithm used most often for scorecard development, designed to create optimal breaks and incorporate user parameters. Refer to the white paper entitled 'Automated Binning in Xeno: The Break and Heal Algorithm, FICO White Paper, 2008', available in the support tab of Xeno.

 

Q: I have submitted a long classing, and need to go home for the day. Can I log out of Xeno?

A: Yes, Xeno will keep running all processes after you log out (with the exception of data import).

Q: How do I select variables for a scorecard?

A: It's best to set the step1 variables from within the classing, where you can view the information values and predictive patterns. When you start a new scorecard and select a classing, the step1 (and step2, if you have them set) variables will be carried over into the initial classing specification. If you wish to choose the variables directly from the scorecard specification, you can use the search capabilities, in addition to visually finding the variable on the list. You can type in a specific string and all variables containing that string will appear. You can also search by role, data type, or storage.

 

Q: I have run a scorecard, and now want to make changes and rerun it. How do I do so?

A: Copy your scorecard by clicking the arrow to the right of the globe. You can make changes to the copy and run it again.

 

Q: I have a lot of variables in step2. How can I make my scorecards run faster?

A: After a scorecard is run, you can demote those step2 variables with little or no contribution. To do so, click the 'auto' button in the bottom right of step2, specify your threshold, and click 'ok'.

 

Q: How do I change my scorecard view?

A: From the scorecard page, click the 'view' link and click the boxes you'd like to see. All scorecards run from a copy of this one will retain that view. If you'd like to change the view permanently for all scorecards you run, click the options button at the top of the page and change the scorecard view there.

 

Q: Can I view an odds to score chart from within the model page?

A: Yes, when reviewing a scorecard, click the distribution page, and click the 'odds plot' link on the page.

 

Q: How do I scale a scorecard?

A: Once a scorecard is run, click on the distribution page, and edit the 'scale mode' parameters.

 

Q:  Once I've scaled a scorecard, I'd like to round the scores. How do I do so?

A: In your scorecard page click ' view' and click the 'round weights' option.

 

Q: I have an existing score on the dataset. I'd like to scale this score similarly. How can I tell how it's scaled?

A: While a score may not be in alignment with its original scaling, you can derive the current odds to score relationship by running a score evaluation report. The PDO is listed, and the target odds and score can be approximated from the odds to score plot (available by clicking the 'apply' button in the graph box). If you'd like a more exact target odds for a particular score, use the fitted slope and intercept reported in the score evaluation report and the equation ln(odds)=slope*score+intercept.

 

Q: How do I create a development and holdout sample to use in scorecard development?

A: If you don't have an indicator variable on the file, create a random number by using the system family of operators. Create a development and holdout filter using that variable (e.g. 'random number LT 7' and 'random number GE 7'). Use the development filter in the classing and scorecard, and the holdout filter in score validation reports.

 

Q:  How do I build a scorecard on a subset of the full population?

A: Create a filter to select the subset of interest (may also want to include a rule on a random 'development' population). Use this filter in the classing and the scorecard.

 

Q: I have created shape constraints in my classing, but they don't seem to be working. Why not?

A: Make sure you have applied the shape constrains using the check box at the bottom of an un-run scorecard.

 

Q: How can I see an ROC curve on my scorecard?

A: After building a scorecard in Xeno, apply the score using the apply scorecard operator. After running your data, go the analyze tab and select score comparison from the report drop down. After you have run the report, click the ROC button.

Attachments

    Outcomes