AutoML and ExplainableAI for JMP (+Python) for Windows and Mac Download latest version
Predictor explainer automates the screening of process variables using feature engineering and machine learning (known as AutoML). Parallel coordinate plots and trends will be automatically shown to interpret the results. If PyJMP is installed (optional), SHAP plots and UMAP will be automatically calculated as well.
For further details and applications of ML applied to industrial processes, you can have a look at our open-access review:
Book chapter - Industrial Data Science for Batch Manufacturing Processes
Batch demo - Predictor Explainer at JMP Discovery Summit 2023
Download and click the . jmpaddin file to install Predictor Explainer:
If you have a previous version installed, it will be automatically removed.
Optional: PyJMP (Python for JMP (Windows only)) can also be installed (SHAP plots will be generated in JMP). If you have Mac or another Python installation, you can still run the notebook included in the folder by clicking on the button "code|examples". Temporary CSVs will be generated inside the 'temp' folder.
In this example will be illustrate how to screen tags (sensor data) to quickly identify correlated factors to the yield in a distillation column.
Go to the Add-ins menu and open Predictor Explainer.
Predictor explainer contains example files. To open the folder, click in the button on the bottom left corner (see A in image below). The following steps are using the file 'distillation_column_na.jmp'.
(A) The add-in folder contains other example files and a python code
(B) If pyJMP is installed, additional options for SHAP and UMAP will be shown.
The main results are:
A hidden and temporal table with all the pre-selected predictors, target and time variables can be accessed via JMP home. The original analysis table won't be modified.
If PyJMP is installed and the option to show SHAP plots is activated, an interactive violin plot will appear after the analysis.
Additional hidden tables containing SHAP, UMAP and clustering results will be accessible via the home menu.
Predictor Explainer can also be used to screen sensors measuring batch processes. The file named 'Fermentation_Batch_Data.jmp' illustrates the challenge of combining unique values coming from a lab analysis with process data.
Predictor explainer will first create a table with summary statistics (also called fingerprints or landmarks).
If there is information about product (grade) or phase (stage) of the batch, these will be also used to generate more granular summary statistics. Using the noise contribution as a filter will eliminate all calculated features below it. When PhaseID is introduced as numeric and ordinal column (1, 2, 3…), an automatic aligning of the batch will be performed and shown.
In the example, Predictor Explainer identified the strongest sensor (tag) in terms of correlation (supervised learning).
If no output is given (no Y, only X's), Predictor screening identifies the sensor with highest variability and creates a global anomaly score (unsupervised learning).
Predictor explainer will calculate the rate of change of all the sensors if row differences is activated.
If you want to distribute or modify a new version of the JMP addin, there are two important things to consider.
Changing the Jupyter notebook will not require changing the JMP add-in itself, as JMP calls the Python code independently via PyJMP. JMP will generate temporary files to communicate with Python and then read the results. The Jupyter Notebooks contains all the steps, to access it open the folder clicking the button on bottom left corner. There is one executable called open_notebook.bat which will allow you to modify the jupyter that is executed.
When modifying the JML add-in source code and saving it (exporting application), make sure to keep the same Unique ID:
As well as add the Notebook and other files you want to distribute with the add-in.
In case you have suggestions, errors, or success stories (!) contact us