See how to:
- Understand the goal: to identify root cause of semiconductor lots with <70% yield and provide overall graphical summary of yield distribution, root cause analysis, suggested corrective action and graphical reports supporting corrective action
- Use Predictor Screening to identify possible key factors re: yield
- Distribution of key factors
- Analysis (ANOVA and Bivariate fit) of yield relationship to key factors
- Various graphs, heatmaps, and box plots showing relationship to key factors to support possible corrective actions
Questions from the session answered by Jason Wiggins @Jason_Wiggins Mark Zwald @mzwald.
Q: What happens to missing data during Join function?
A: Missing values will be in the joined table provided the match conditions are met.
Q: Can the Y be categorical in predictor screening?
A: Yes
Q: How is Predictor Screening different from Linear Regression?
A: Predictor Screening use a Bootstrap Forest approach to variable contributions (identify predictors - Xs that influence responses – Ys), whereas Linear Regression ranks the significance of the coefficients of the polynomial equation used in fit. One benefit to Bootstrap Forest is that it does not require an assumed model. Predictor Screening does not do a regression, so it's not looking at a correlation and computing a p-value the way Response Screening does. Predictor Screening and Response Screening kind of answer the same question but in two completely different mathematical ways. Predictive screening takes different random subsets of your data table, generates a decision tree on each subset, and then averages over those trees. So, it's not actually computing a p- value between an X, Y correlation and then ranking the effects by those p-values. It uses a Random Forest Model to looks at how much of the variation of the responses comes from those factors. You can also do this using Analyze>Predictive Modeling>Partition to create a single decision tree. Predictor Screening does it automatically (Bootstrap Forest) on subsets by column and generates a decision tree for each, then creates an average. Predictor Screening doesn’t miss factors that may be weak on their own but strong through interaction. You can also use Response Screening and compare results.
Q: What is the meaning of the Contribution and Portion values in the Predictor Screening Report?
A: The report shows the list of predictors with their respective contributions and rank. Predictors with the highest contributions are likely to be important in predicting Y. The Contribution column shows the contribution of each predictor to the bootstrap forest model. The Portion column in the report shows the percent contribution of each variable.
Q: In predictor screening, are factor interactions considered?
A: Yes. A good way to be convinced of that is to perform the same analysis in JMP Pro using the Bootstrap Forrest platform. Evidence of interactions and higher order behavior can be found in the Profiler. This is useful because we do not have to have an assumed model like we do with regression. We get all that information for free.
Q: Do I need JMP Pro for Predictor Screening?
A: No, although JMP Predictor Screening gives you the Column Contribution Report that JMP Pro Bootstrap Forest provides. With JMP Pro, when you run Bootstrap Forest you can run the model and save the model results, which you can't do using only JMP Predictor Screening.
Q: In the One Way and Bivariate Analyses, can you color the dots by CMP?
A: Yes. Simply assign value colors as a column property for the CMP column. Rows > Color or Mark by Column. Then choose Colors and Save to Column Property.
Q: Re: publishing reports? Can you publish your report to a private company location?
A: Yes, you can publish in interactive HTML to a private location and organizations that license JMP Live can publish to any external or internal server and set up a full range of securities, groups, etc.
Q: Do you need a special version of JMP to publish to JMP Live?
A: JMP Live was offered with JMP 15. Your site needs to license JMP Live, establish security, etc.
Q: Is there a way to save interactive reports without JMP live?
A: Yes, there are many options using File>Save As. One interactive option for may reports is to save as interactive html. Some capabilities, like Column Switcher and Data Filter may not be available without saving to JMP Live.
Q: Do the associated data filters save with the data table?
A: Yes, from the right triangle under reports with data filter, you can Save script>To Data Table. You can also Save script>To Data Table (All Objects)
Q: If you add rows, will the graphs be updated?
A: Many platforms allow you to choose Automatic Recalc for changes to row states (added, deleted, hidden, excluded). To turn on Automatic Recalc for a report window, click the platform red triangle and select Redo > Automatic Recalc. To turn it off, deselect the same option. Platforms that support Automatic Recalc include a preference in the Platforms preferences. To turn on Automatic Recalc for a report window, click the report red triangle and select Redo > Automatic Recalc. To turn it off, deselect the same option. From any platform red triangle, you can also choose Redo>Redo Analysis.
Resources:
