This workflow generates a fully automated web based application to select, train, test, and optimize a number of machine learning models. The workflow was designed for business analysts to easily create predictive analytics solutions by applying their domain knowledge. Each of the components will generate a web page with which the business analyst can interact.
This is a companion discussion topic for the original entry at https://kni.me/w/eAGfGtEAIr-1iYR-
Please note: In order to execute this workflow you also have to download all workflows and files which are part of the same workflow group. You can do this via our Example Server from KNIME Analytics Platform at
/Examples/50_Applications/36_Guided_Analytics_for_ML_Automation/. Stay tuned, soon you can download entire workflow groups also from the hub
Is it possible to run this on a web server like Tomcat? Or do we have to have a KNIME Server license first?
you can run the guided analytics applications for free from the KNIME Analytics Platform. You open a view of the single Wrapped Metanode and a browser window will pop up. You interact with the view and then in the bottom right corner you can click “Apply” and “Close”. The interactions will be saved and passed on to the next nodes and you will then open the view with the next Wrapped Metanode going through the application step by step like this.
However if you want run the guided analytics application as web application remotely accessible by any web browser, you can without even having the KNIME Analytics Platform installed!
To do this you need the KNIME WebPortal, which comes with KNIME Server for which you would need a license. From the hardware point of view you can either use a Tomcat server on premises or on a cloud service like AWS EC2 Instance. KNIME Server is also available on Azure and Google Cloud.
So to answer your question you would need to have a license first, sorry
If you are interested in a KNIME Server Trial contact me at email@example.com and I will see what I can do for you.
we recently released a fresh revamp of this workflow with the following changes:
The workflow now runs smoothly with default settings on a small data set within minutes for you to test it, no need to set up all the nodes settings one by one while still learning how it works.
From the KNIME Analytics Platform you can play with three default data set from a drop-down bar. Of course you can also manually add your file path manually by opening the settings of the node “Upload Body” node. If the workflow is executed remotely from KNIME WebPortal an upload button will be provided.
- The flowchart header is now provided as a template component that is easily reusable in another workflow. No need to code, just double click on the “Header” component and adjust the look of the flowchart.
- We rebuilt the inside of the components by dividing the dashboard in nested components. It is now much easier to explore sets of nodes generating different parts of the user interface.
CSS nodes added to the workflow.
Guided and automated parameter optimization customization available for all models.
- We added useful links to the hub pages. There are now referenced videos and blog posts for you to understand the full strategy behind this blueprint.
To download and use this workflow please make sure to grab the entire Workflow Group from the Example Server at:
Please provide feedback on any features or anything you would like to change in this blueprint.
we recently uploaded a new version of the workflow which fixes some small issues users had when running the default settings.
Let us know if you encounter any other issues.
Recently I’ve been running this workflow on my machine and ran into a problem.
I can’t seem to filter my columns. No matter how I set the slider in the “Range Slider Filter Definition” node (2:663:0:239:0:214), the following “Filter Apply” node (2:663:0:239:0:215) always filters the columns with the value of “0” for my chosen “Range Column”, ignoring what I select.
Using the “airline.csv” dataset that you provide, if I choose my “Range Column” to be “Overall Column Relevance” in the options of the “Range Slider Filter Definition” node, the following “Filter Apply” node will always output a table with rows relative to the “Year”, “Cancelled” and “CancelledCode” features (which all have a value of “0” for the “Overall Column Relevance”) no matter how I set the slider values in the former node.
Hello @aecarvalho -
Welcome to the forum and thanks for the feedback. Coincidentally, I just filed a ticket about this exact problem - the Range Slider Filter Definition node seems to have a bug where the minimum and maximum defaults are not being passed via the filter definition port.
Sorry for the trouble - we’re looking into it!
First of all thanks for the swift response.
I just found out that that bug seems to be solved on the previous version 3.7.2.
I decided to share just in case it might help you guys solve the issue.
we fixed the issue also on 4.0.1
You can now filter just like before.
@SimonS also added XGBoost to the list of trainable models.
We added today new features and bug fixing to this workflow!
- String to Number node in Selection Target Component does not fail anymore
- Workflows updated to KNIME Analytics Platform 4.1
- Minor changes to default settings on Parameter Optimization for Gradient Boosted Trees and Deep Learning.
- Added new view to final dashboard: Binary Classification Inspector node! (https://kni.me/n/3-JGPq9anCe8LGG6). The view is activated if binary classifiers are trained. You can now evaluate the model confusion matrix by interactively optimize the threshold of the models. Video on this tweet:
is there a way to get the 4.01 version?
Today we pushed to the KNIME Hub another update of Guided Automation.
New look with colorful Component with added Metadata:
- Yellow for data manipulation
- Blue for views
- Light Green for model training
- Dark green for model validation
We also added Parallelization nodes ( Parallel Chunk Start and Parallel Chunk End ) where ever it was possible. This speeds up the computation executing workflow benches in parallel when ever possible.
This option can be removed via the Execution Settings view:
This is especially efficient when using KNIME Server and multiple KNIME Executors:
In the next few hours we are going to update this workflow with updated graphics from the new WebPortal design! Enjoy! (and thank you @daria-tombolelli for taking care of the update)!