Performance Spectrum Miner

The latest release is here.

Overview

The Performance Spectrum Miner (PSM) is a visual analytics tool for process performance analysis. It takes an event log in a CSV or XES format, visualizes the flow of all cases over time, and gives detailed insights into performance characteristics.

The PSM visualization

• shows how the performance of a process varies over time regarding throughput, volume, steadiness levels, peaks, and drops,
• allows to analyze detailed performance characteristics of each step such as variability in waiting, prioritization of cases, delays and synchronization behavior effecting multiple cases together,
• reveals various performance patterns such as queueing disciplines, batching, prioritization and overtaking, slow movers, temporary bottlenecks, changes in process, and many more, and thereby
• gives insights into different performance variants of the process within each step and across steps, and how these change over time.

The PSM project provides two implementations of the Performance Spectrum Miner as a plugin to the Process Mining Framework ProM and as a stand-alone application.

The PSM project is the result of the joint research project on Process Mining in Logistics between Eindhoven University of Technology and Vanderlande Industries, and developed by Vadim Denisov, Elena Belkina, and Dirk Fahland.

Publications

• Predictive Performance Monitoring of Material Handling Systems Using the Performance Spectrum (ICPM2019)

  • The paper describes a method to extract features for training a performance predicition model from the performance spectrum. The User manual provides details on how to extract performance spectrum-based features. Additional paper-specific documentation and the datasets used in the experiments are documented here.

• Unbiased, Fine-Grained Description of Processes Performance from Event Data (BPM2018)

• The Performance Spectrum Miner: Visual Analytics for Fine-Grained Performance Analysis of Processes (BPM2018, Demo)

• BPIC'2018: Mining Concept Drift in Performance Spectra of Processes (BPIC2018)

How to Install

System requirements

• An OS supporting Java 8, e.g., Microsoft Windows 7 or higher, MacOS, Linux
• 2 GB RAM minimum, 16 GB RAM recommended
• 100MB hard disk space for ProM, 2 GB hard disk space for caches recommended

Prerequisite: Java 8

The PSM is implemented and tested with Java 8 and is not binary compatible with other Java versions.

1. Install the most recent JRE/JDK 1.8 64bit
2. Make sure that a correct installation of Java is configured: execute java -version in the command line. You should get a response like this:

java version "1.8.0_171"

Java(TM) SE Runtime Environment (build 1.8.0_171-b11)

Java HotSpot(TM) 64-Bit Server VM (build 25.171-b11, mixed mode)

*If you do not want to change your current Java installation to Java 8, you can download Java 8 and explicitly call it while starting the PSM or ProM (in 'ProM.bat'), for example://

"C:\Program Files\Java\jre1.8.0_171\bin\java.exe" -jar perf_spec-assembly-1.0.2.jar

Using ProM release for the ICPM 2019

Not available yet...

Installation of the PSM as a ProM plugin (with a nightly ProM build)

1. Download ProM nightly build. The PSM has been tested with versions 14th August 2018 and later.
2. Run ProM Package Manager (execute PackageManager.bat)
3. Go to tab 'Not installed', find and install plugin PerformanceSpectrum.
4. Exit ProM Package Manager
5. Recommended for large datasets: open file ProM.bat in any text editor and change parameter –Xmx from 4 to a value equal to your laptop's RAM size minus 2
6. Execute ProM.bat to run the PSM

Update of the PSM ProM plugin

1. Close ProM (if opened) and run ProM Package Manager (execute PackageManager.bat)
2. Go to tab 'Out of date', find and update plugin PerformanceSpectrum.
3. Exit ProM Package Manager

Installation of a stand-alone version of the PSM

1. Download and unzip perf_spec-assembly-1.1.1.jar
2. Execute java -jar perf_spec-assembly-1.1.1.jar to run the PSM

Getting Started

Analyzing the Performance Spectrum of a process with the PSM has three steps:

1. Transformation of an event log, either in XES or CSV format, to a Performance Spectrum disk (file) representation. Different performance classifiers and aggregation functions can be used.
2. Opening the transformed data for analysis with the PSM
3. Exploring the Performance Spectrum

Additionally the PSM allows encoding and exporting performance spectrum-based features into a training and test sets.

Transforming an event log in the XES format for Performance Spectrum Analysis

... in ProM

1. Load the event log into ProM via the Import... button.
2. Go to the Action Tab and select the Performance Spectrum Miner plugin from the action list.
3. Choose parameters for generating the performance spectrum data.
   • A configuration dialog will show providing default values for the transformation.
   • See the User Manual for details.
   • The transformed data will be stored on disk in the Intermediate storage directory together with a meta-data file (session.psm). You can load this transformed data also later into ProM by loading the session.psm meta-data file.
   • Choose Process & open
   • The transformation may require some time and main memory depending on the Bin size chosen. Transformation for larger bin sizes are faster and require less memory.
4. In ProM the transformed data will then also be opened automatically

... in PSM standalone

1. Load the event log (XES format) via the Open... button.
2. Choose parameters for generating the performance spectrum data.
   • A configuration dialog will show providing default values for the transformation.
   • See the User Manual for details.
   • The transformed data will be stored on disk in the Intermediate storage directory together with a meta-data file (session.psm). You can load this transformed data also later via the Open... button.
   • Choose Process & open
   • The transformation may require some time and main memory depending on the Bin size chosen. Transformation for larger bin sizes are faster and require less memory.

Transforming an event log in the CSV format for Performance Spectrum Analysis

Often event data are available in the CSV format as a database or a distributed file storage dump, stored in one or many CSV files. Converting such dumps to XES format can be difficult for large event logs. The PSM supports a direct import of one or many CSV files. To prepare CSV file(s) for import, put the file(s) into a directory and provide a description as a text ini file with extension .csvdir. This file must include the following fields (see example):

| Field |Sample value | Comment | |:------------- |:-------------|:-----| | dateFormat | dd-MM-yyyy HH:mm:ss.SSS | Datetime format in Java DateTimeFormatter format | | zoneId | Europe/Amsterdam | Time zone ID in Java ZoneId format | | startTime | 01-09-2018 00:00:00.000 | Since then the performance spectrum should be computed, in the format described above | | endTime | 08-09-2018 00:00:00.000 | Until then the performance spectrum should be computed, in the format described above | | caseIdColumn | CaseID| Column name for case ID | | activityColumn | Activity | Column name for activity | | timestampColumn | Timestamp | Column name for timestamp |

... in ProM

Import your .csvdir file and use it in the PSM plugin, exactly as XES files.

... in PSM standalone

Import your .csvdir file and use it exactly as XES files.

Transforming a dataset of segments for Performance Spectrum Analysis

Sometimes event data are availabe not as an event log but as segments. Find here how to import such data into the PSM.

Opening the transformed data for analysis with the PSM

1. By choosing Process & open during data transformation, the transformed data will be opened automatically. Alternatively, you can also load a previously transformed data set by opening the .psm meta-data file (via Import in ProM, and choosing Performance Spectrum Miner View, or via Open in the stand-alone version).
2. Cho