[Unsupported Extension Notice] - The Interpretation Extension is not officially supported by Altair RapidMiner. While we're always improving and updating our offerings, we can't guarantee any help or fixes for this extension.

Changes in 0.8.0

•  Fixed a bug which made ConformalPrediction models unstorable.

 

Changes 0.7.1

• Added a proper user error for missing values in Estimate Uncertainty
• Fixed a bug where regression based conformal predictions was using wrong normalizations in LIME
• Added a progress bar to estimate uncertainty

 

Changes in 0.7.0

• Added new Operator Interpret Groups that currently calculates the average differences between categories
• Added a new Operator Estimate Uncertainty, which uses Conformal Prediction to estimate the uncertainty of your prediction (Classification or Regression)

Changes in 0.6.0

• Generate Interpretation now supports IOTablePrediction Models. Those are (as of RM 9.11):
  • - Function Fitting
  • - XGBoost
  • - Anomaly Models from Anomaly Extension (version 4.0+)
• Fixed a bug where Generate Interpretation fails if you provide a python based classification model which does not provide a confidence. You now get a proper error message.

 Changes in  0.5.0

• LIME is now not re-scoring the artifical table if you you don't redraw in each iteration. This can speed up LIME for complex and python models by a lot.
• Generate Interpretation now adds an additional column "Interpreted Column" which explains which column was interpreted. This can be very useful for classification problems.
• Added an additional error message when your model does not know it's training label. This can happen in rare cases when you use a Python Learner.

 

Changes in  0.4.1

•  Added Accumulated Local Effects to Generate Univariate Interpretation
•  Generate Univariate Interpretation's attributes are now called PDP(attributeName) and ALE(attributeName).
• The number of attribtues which are shown in the interpretations column of Generate Interpretation is now configurable.
• Fixed a bug in Generate Interpretation which caused that only the top values and not top absolute values were shown.

 

Changes in 0.4.0

• Added a new Operator Generate Univariate Interpretation which currently provides PDP as an algorithm.
• Fixed a bug which made Shapely fail on data sets with 2 attributes

 

Changes in 0.3.0

• Added a new operator Generate Neighborhood which returns the LIME-weighted Neighborhood
• Fixed a major bug in LIME, which made the weights incorrect.

Changes in 0.2.0

* Grouped models are now supported

* Fixed an issue which made the result not reproduceable, even though a static random seed was used

* Fixed a bug which influenced the calculation of LIME on nominals. Nominals were treated like numericals.

 

Changes in 0.1.1

* Added an icon

* Fixed an error that there was no proper UserError if types of training and testing were different

* Fixed an error that real and integers were considered of different types.