Knime L3 Study Material

[L3-PC] KNIME Server:
Productionizing and Collaboration

KNIME GmbH
1
Agenda
§ Session 1
§ KNIME Software Overview
§ Working with KNIME Server
§ Connect to KNIME Server
§ Server-Side Workflow Execution
§ Remote Workflow Editor
§ Permissions & Versioning
§ Session 2
§ Introduction to Components
§ Component Configuration
§ Composite Views
§ WebPortal Applications
§ Session 3
§ KNIME Server REST API
§ Integrated Deployment
§ KNIME Server Administration
© 2021 KNIME AG. All rights reserved. 2

KNIME Software - Overview
KNIME Software – One Ecosystem
Create Productionize
Blend & Transform Model & Visualize Deploy & Manage Consume & Interact
KNIME KNIME
Analytics Platform Server
KNIME KNIME Community Partner Team End User API Managed

Extensions Integrations Extensions Extensions Collaboration Applications Services Execution

KNIME Analytics Platform: loved by individuals
3000+ Nodes for all Steps of “End-To-End” Data Science

Data Science for the Business:
Individual:Creation
Create & Production
Great
Model or Report

Data Science Practice: Multiple Stakeholders’ Needs
Data Engineers Smart Business Users (more than Excel)
Data Science “coders” (Python, etc.)
Application Users – Interaction required
Data Science Specialists
Data Science Visual workflow / Application Users – Made to spec
generalists Report Consumers
Model / ML Operations Operations Consumption

Feeding production systems (Applications, Systems, Edge, etc.)
IT Operations
Centralized resources / strategies
Standards and preferred platforms used, Financial / Risk Oversite
infrastructure options Costs allocation
Exit strategies Compliance officer
IT Security Data/model Governance,
Data, applications traceability, GDPR

KNIME Software - one Ecosystem
Create Productionize
Gather & Wrangle Model & Visualize Deploy & Manage Consume & Optimize
Open Source Commercial

KNIME Analytics Platform KNIME Server
Data
KNIME KNIME Community Partner KNIME
Science as
Extensions Integrations Extensions Extensions WebPortal
a Service

Data Science Practice: Teams !

IT Operations

Reuse, Share and Document
Workflows & Components build all types of applications – automated and interactive
Features:
§ Self Documenting
§ No limits: All nodes
Workflow § DB, Spark, DL, Python
etc.
§ Task packaging
§ Mix and Match
§ Sharable / Reusable /
Instantiated
Component
https://www.knime.com/blog/knime-analytics-platform-40-components-are-for-sharing
© 2021 KNIME AG. All rights reserved.

Capture IP, Leverage Specialists, Collaborate
Sharable / Reusable / Instantiated Workflows, Components and Collaborative Development
Database Specialists Data Engineers Python Specialists
Data Science Specialists Visualization Specialists

Data Governance
Data Science Generalists
As a web-based
application
within a workflow
(manual/automated)

Recoverable, Backward Compatible, No Migration
Workflows, Components and Collaboration Features:

§ Instantiated / Updated
§ Secure (create, use, etc.)
§ Versioned and snapshots
§ Version comparison
§ Locked / Encrypted
§ Reproducible
§ Guaranteed backward
compatible
https://www.knime.com/blog/knime-meets-knime-will-they-blend

IT Operations

Data Science for the Business: Creation & Production

Empower Business Users Appropriately
Delivering reports and output to business users appropriately
Features:
§ Visualizations
§ Plotly, JavaScript, etc.
§ Reports Creation BIRT
§ Integration with
§ Excel
§ Functionality exploitation
not just CSVs
§ PowerBI
§ Tableau
§ Qlik
https://www.knime.com/community/continental-nodes-for-knime-xls-formatter § Spotfire
§ …
© 2021 KNIME AG. All rights reserved. § 15
Empower Business Users Appropriately
Guided Analytics for Building Applications

Appropriate levels of Automation & Human Interaction at any stage of the Data
Science Life Cycle depending on task and audience
Features:
§ Workflows and webportal nodes
build interactive applications
& dashboards
§ KNIME WebPortal manages access
https://www.knime.com/blog/principles-of-guided-analytics


IT Operations

Data Science for the Business: Creation & Production

Flexible Delivery Options: Automate
Automated workflow execution
Features:
§ Scheduled
§ Triggered
§ Called (Rest / SAAS)
§ Call Actions based on status
§ Scale and Pin Execution
§ View, edit, execute workflows
remotely
https://docs.knime.com

Continuous Integration / Continuous Deployment
Integrated Deployment
Features:
Creation Workflow § Eliminates the gap
between
Creation and Production
§ mark what‘s necessary
for production
§ not just the model
§ no manual intervention
Production Workflow § No rewrite, copy paste
§ Capture all nodes &
settings
§ Automatically created
production workflow
https://www.knime.com/integrated-deployment § Always in sync

IT Operations

Governance & Compliance
Explainability / Interpretability of models
Features:
§ Many Techniques available
§ LIME
§ SHAP
§ Shapley
§ Partial Dependence / ICE
§ Binary Classification Inspector
https://hub.knime.com/knime/extensions/org.knime.features.mli/latest

Governance & Compliance
Data / Model Lineage
Archive
Document
Explore &
Analyze


IT Operations

Manage Infrastructure & Users
Central Management and Monitoring capabilities
Features:
§ Client Customizations
§ Custom update sites
§ Manage preferences via profiles
§ Node repository & libraries
§ Monitor server activity
§ Running and scheduled jobs
§ Adjust permissions
§ Manage ongoing services

Single Sign-on, Integrate with multiple Security protocols
OAUTH, LDAP, AD Integration
Features:
KNIME
Server § Single sign-on (SSO) to KNIME
Server
§ Integrate with multiple identity
providers
§ Flexible configuration capabilities
Client
Identity
to map users and groups
Provider
§ Manage all aspects of KNIME
usage
https://docs.knime.com

KNIME Hub: Sharing Resources
https://hub.knime.com/

Log in to KNIME Hub and publish your Workflows
KNIME Forum
Account Credentials

Edit the Workflow
Drag & Drop

Working with KNIME Server
Connecting to KNIME Server
31
Set Up a New Mount Point
Server connections are shown as “mount

points” in the KNIME Explorer. To add a
new mount point simply:
1. Click the Configure button in the KNIME
Explorer.
2. Click New…
3. Configure a mount point with your details

Server Mount Point as a Shared Repository (1/2)
The Server provides an area in the

Explorer for sharing work with your
colleagues. Use workflow groups to Click to add new
organize your workflows, components, mountpoint
and data files. Data
Workflow Groups
To move resources simply drag Workflows

Jobs
and drop or copy and paste
Components

Server Mount Point as a Shared Repository (2/2)
Another way to deploy resources on KNIME Server..

Inspecting a Workflow from KNIME Server
§ By double-clicking a workflow on KNIME Server, the client downloads it (to a

temporary location) and subsequently opens it automatically
§ The yellow bar at the top of the editor indicates that this is a temporarily
downloaded server workflow

Automation - Remote Execution
36
Executing a Workflow on the Server – Remote Execution
Check to reset workflow before

execution. All nodes are reset
(including File and Database
Reader nodes, etc.).
If selected, the executed job is

deleted immediately after
execution, and is not saved.
Enter one or multiple email

addresses (separated by
commas) to which a notification is
sent after the workflow execution
has been finished.
The name of the workflow job as

it is displayed in the server view.
By default this is the name of the
workflow. The execution date is
always appended to the name.

If the workflow contains a report,
you can select to save it on
KNIME Server.
By default the name of the report

is the name of the workflow. You
can define a custom report file
name, if you wish so.
You can overwrite the report with

every execution or append a
timestamp.
Here you can define the location,

where you want to store the report
on KNIME Server.
You can select in which formats

the report should be saved.

Depending on whether your job

executed successful or not, you can
configure to run subsequent
workflows

Here you can specify the first and
optionally the last execution date of
the scheduled job (in case it’s a Repeating jobs can be repeated
repeating job). after a certain number of minutes,
hours, or days. The latter takes into
account daylight saving, i.e. the
start hour will be the same in winter
and summer (e.g. 12:00).
By default, repeating jobs are

executed every day. Here you can
filter whether they should run only
on certain days of the week, days
of the month, or only in certain
months. “Last” means the last day
of the month.
If the previous job is still running

when the next execution is
supposed to start, you can opt to
skip this execution.
Scheduled jobs can be disabled

temporarily.

Configuration nodes from the

workflows can be set via execution
dialog to parametrize workflow
execution

Workflow Jobs
Remotely executed workflows are run as Jobs:

§ A workflow job is a copy of the workflow with specific settings and data
§ Jobs are tied to the version from when the job was created
§ Orphaned jobs are colored red
§ Jobs have messages (e.g. successful or failure)
§ Jobs can be saved as a workflow for data provenance and debugging
(right-click à Save as)

Remote Workflow Editor
43
Remote Workflow Editor – 1/3
§ Remote Control of Job running on KNIME Server

§ Capabilities:
§ Live update of workflow job execution
(executing node and progress)
§ Execute and cancel execution supported
§ Add/delete nodes
§ Change node settings
§ Inspect data tables / flow variables
§ JavaScript nodes can show data/views

Remote Workflow Editor – 2/3
What’s my Change node JavaScript

workflow doing configurations in View support
now? the Job

Remote Workflow Editor– 3/3
§ To use the Remote Workflow Editor please install the KNIME

extension Remote Workflow Editor

Permissions
Permissions
§ Permissions can be set for all types of items: workflows, workflow groups,
components, and data files
§ Permissions are assigned to either individual people or user groups
§ The user who uploads an item, automatically becomes its owner
§ Users with admin rights have no restrictions on permissions
§ The owner, plus everyone with admin rights, can assign and change
permissions for an item
§ It is also possible to set permissions on schedules, such that a schedule can be
maintained/changed by a team member while the owner is e.g. on vacation

Permission Types
Type Workflow Workflow Groups Files/Components
Read Download a workflow job - See the content of a workflow group § File: download data and
including data execute workflows that use the
data
§ Component: use and download
Write Overwrite, create snapshots, Create and upload new items in a Overwrite a file or component
and delete workflows workflow group
Execute Execute a workflow by creating

a workflow job

Setting Permissions
Everybody Else

Versioning
Versioning
§ Possibility of creating a history of items on Server

§ Create snapshots of workflows, data files, and
components. These are stored with a timestamp
and a comment

KNIME Workflow Difference (1/2)
§ Automates identification and comparison of nodes in a workflow, metanodes,

and two different workflows
§ Identifies insertions, deletions, substitutions, and parameter change

KNIME Workflow Difference (2/2)
Highlight differences:
§ Nodes included/excluded
§ Node configurations

Review and Exercises
55
Config Details – Access KNIME Server
§ KNIME Server Address: http://3.123.122.13

§ KNIME WebPortal Address: http://3.123.122.13/knime
Login credential: firstname.lastname (*)

Password: knime
*for double names and double surnames the whitespace has been removed

Working with KNIME Server, Activity 1
§ Configure a mount point for KNIME Server with the details provided in the
Config Details – Access KNIME Server slide at the end of the slide deck
§ Download Server Training Material in your LOCAL workspace (hint: drag
and drop or copy and paste the entire folder)

Working with KNIME Server, Activity 2
Path to the workflow:

Exercises à 01_remote_workflow_execution à 01_data_mining
§ Create a local copy of the workflow and rename it as

“01_data_mining_Use_Case_YourInitials”
§ Deploy the workflow on KNIME Server under the directory

Solutions à Uploaded Workflows
§ Schedule a workflow execution (5 mins later)

Working with KNIME Server, Activity 3 (1/2)
Path to the workflow:

Exercises à 01_remote_workflow_execution à 01_data_mining
§ Make some changes to the workflow (e.g. generate 5 clusters instead of 4),
save it as “01_data_mining_YourInitials” in your Local Workspace and
afterwards deploy it to KNIME Server under the directory
Solutions à Uploaded Workflows
§ Make sure to create a snapshot when updating your workflow on KNIME Server
§ Overwrite the existing item
§ Check the option Create Snapshot before Overwriting and provide a name to the snapshot
§ The snapshot is listed in the Server History view (View à Other à Server History)

Working with KNIME Server, Activity 3 (2/2)
§ Download the snapshot to LOCAL Workspace

§ Use the WorkflowDiff feature to compare the downloaded workflow with the
latest workflow deployed on KNIME Server:
§ use Ctrl (or Cmd) button to select both workflows from KNIME Explorer;
§ right click on one of the two workflows
§ click Compare
§ select the two nodes that you wish to compare
§ click Show configuration differences of highlighted nodes

Components
What are they good for?
§ Components encapsulate functionalities that can be reused as your personal

customized KNIME nodes, to perform tasks that you often repeat.
§ They can also be shared with others via KNIME Hub and KNIME Server.

KNIME Verified Components
§ Verified Components reuse bundled functionalities, verified by KNIME experts

§ Released and updated on the KNIME Hub

Metanodes vs. Components
Metanodes Components
Configuration Not configurable Via Configuration nodes (local workflow) and

Widget nodes (KNIME WebPortal)
Variable scope Global Configurable: Local or global
WebPortal Usage Executed in the background JavaScript Views and Widgets inside the
component are shown on a WebPortal page
Execution mode Normal execution Allows Simple Streaming execution
Recommended uses Workflow cleaning Enabling custom interactions, producing
interactive views, sharing functionalities

Component Setup
§ Add input and output ports to components

§ Remove ports to adapt to changes after
creation of the component

Flow Variable Scope of Components
§ Flow Variables are -by default - only available locally inside the component
§ Configure the component input/output to pass Flow Variables
from/to outside the component

Component Description
§ Double click a component to configure it

§ Providing meaningful description is best
practice

Components: Creating custom
nodes from KNIME nodes
Creating a Configuration Dialog
§ Configuration nodes allow creation of a Component’s configuration dialog

§ Configuration nodes enable different types of user inputs such as string input,
integer input, selecting one value from a list and many more

Creating a Configuration Dialog
§ Use Configuration nodes to create Flow Variables

§ Every Configuration node has a label, description and parameter/variable name
§ Depending on the node, additional options and visual properties can be set
§ Flow variables created in the Configuration nodes can then be used to overwrite
the settings of subsequent nodes

Configuration of a Component
§ Double click a component to configure it

Shared Components
What is a shared component?
§ Components can be saved in your

KNIME workspace, on KNIME
Server or KNIME Hub for later
reuse
§ To do this, simply right-click any

component and select “Share…”
§ Shared components are read-only
instances of a component
§ Public Shared Components are
available on the EXAMPLES
Server and on the KNIME Hub

How can you edit a shared component?
§ Components can be edited using

the Component Editor, similar to
workflows
§ To edit a component using the
Component Editor, double-click the
component in its location in the
KNIME Explorer
§ To ensure components are

executable when opened in the
Component Editor, choose the
option to “Include input data with
component” when sharing it

How can you use a shared component?
§ To use a Shared Component, drag

and drop it to your workflow editor
§ Instances of Shared Components

can be updated either manually or
when the workflow is opened
§ A Shared Component can also be

unlinked from its original location,
which makes it editable in the
workflow directly
§ Update Shared Components by

overwriting them

Component Composite View
Creating a Composite View
§ JavaScript visualization and Widget nodes allow creation of a composite view

§ Widget nodes allow user interaction in a composite view, e.g.
§ Adjustment of parameters via various input nodes
§ Selection of predefined values
§ Consumption of outputs

§ Multiple JavaScript View nodes can be combined in Components

§ Selections are transmitted to all other views
§ User can give input via Widget nodes

§ You can open a composite view directly from

the node’s context menu.
§ It will open a browser window that displays
all containing JavaScript and Widget nodes.
§ The same view will also be shown via
KNIME WebPortal

Configuration of Widget Nodes – Input Nodes
§ Widget input nodes have very similar configuration dialogs

Layouting
§ A layout can be defined for any Component that contains at least one widget
node or JavaScript-enabled view
§ The layout editor can be accessed from the top toolbar, when inside the
component

Layouting – Nested Components
§ It is possible to include components inside other components

§ It allows to build up a library of components that contain useful linked views and
then easily assemble them to create complex views for visualizing and
interacting with datasets

Defining the Layout
§ Four tabs available:

§ Node Usage
§ Visual Layout
§ Basic Layout
§ Advanced Layout

Layouting – Order Items to Show on the WebPortal
§ The “Append the IDs to node names” button on the top bar shows the ID of each
node
§ This is useful to reorder the items in the layout structure for the WebPortal

Layouting I – Node Usage
§ Enabling the view of Widget

nodes on the WebPortal or on
the Component view
§ Enabling the view of the input
given by the Configuration node
on the Component Dialog

Layouting II – Visual Layout
§ Makes it easier to specify

how individual JavaScript
views are laid out in a
WebPortal page or
Component composite view

Layouting III – Order Items to Show on the WebPortal

Layouting IV – Order Items to Show on the WebPortal

KNIME WebPortal
90

IT Operations

Close the Gap: Guided Analytics and Automation
Extending data
Incorporate domain Amplifies the best
science to the
experts’ knowledge data science
Business Analysts
KNIME
Guided
Analytics &
Automation
R&D MFG IT Marketing Sales more …

KNIME WebPortal
§ All workflows on KNIME

Server available as web apps
§ Step-by-step execution of
workflows from any browser
§ Simple, clean interface for
end users: Guided Analytics
§ Customize layout to match
corporate design

Guided Analytics: Interactive Data Science
Interaction Points

Inside an Interaction Point

Access to KNIME WebPortal from the Client
§ Possibility to start the execution of a workflow on KNIME WebPortal directly from

KNIME Analytics Platform: right-click on the workflow available in the KNIME
Server Mountpoint and click Open in WebPortal

Detail Pane
§ If a workflow is selected in the left section, its details page is shown in the
section on the right
§ Server sends a notification email to the address as soon as the workflow

execution finishes

WebPortal Applications:
Guided Analytics for ML Automation
Guided Analytics for ML/AI Automation
§ Interaction & Automation

§ Data Scientist’s Choice: right mix for target audience

WebPortal Applications:
Customer Segmentation
103
Classic CRM Analytics
§ CRM System § Churn Prediction

§ Data about your customer § Upselling Likelihood
§ Demographics § Product Propensity /NBO
§ Behavior § Campaign Management
§ Revenues § Customer Segmentation
§ …
Model

Customer Segmentation
§ Customer Segmentation is a standard technique that consists of dividing a

customer base into groups of individuals that are similar in specific ways
relevant to marketing, such as age, gender, interests, and spending habits
§ Companies need to treat their customers differently depending on the segment
to which they belong
§ By providing personalized offers and communication to each segment,
companies can:
§ Increase customer lifetime value
§ Reduce irrelevant customer interactions
§ Differentiate their brand relative to their competitors
§ Generate higher profits
§ We will build a workflow that implements a customer segmentation technique and then
construct a web user interface to inject business experts’ knowledge into the final results

Basic Workflow for Customer Segmentation
§ Predefined data sources

§ Static configuration

WebPortal Workflow for Customer Segmentation
§ Predefined interaction points

§ Allows expert to give inputs and consume outputs via a browser

Step by Step Walkthrough

Define Cluster Parameters
KNIME AP KNIME WebPortal

Display Cluster Result

Take Notes

Display Labeled Clusters

Stepping through Pages – Wizard Execution
§ Each Component node represents one

WebPortal page
§ To preserve the order of the sequence

of each web page, these need to be
connected to each other
§ It is also possible to use Component

nodes inside loops to iterate over an
item set or recursively refine a model

KNIME WebPortal: URL Parameter (1/2)
§ It is possible to link directly to specific workflows in the WebPortal. URLs are

generally set up like this:
§ http://<server-address>/knime/webportal/<ItemPath>?exec&<WorkflowParameters>
§ <ItemPath> = The path to a workflow, workflow group, or workflow job (a workflow job is
referenced with its ID like: WorkflowGroup/Workflow?exec=job_id)
§ <WorkflowParameters> can appear in any order, but have to be after the <ItemPath>?exec.
Parameters are always appended with a leading ‘&’
§ List of available workflow parameters:

§ &pm:<name>=<value> - Set widget parameters: sets the named widget parameter to the
specified value.

KNIME WebPortal: URL Parameter (2/2)
§ &emails=sample@mail.com – Enable email notification: enables email notification and sets the
specified comma-separated list of email addresses
§ &formats=<formats> – Set report formats: sets the report formats included as attachments in the
notification email specified by a comma-separated list. Available formats are: pdf (enabled by
default), html, doc, docx xls, ppt, pptx, ps, odt, ods, and odp.
A complete URL might look like:

http://localhost:8080/knime/webportal/demo/file%20to%20csv?exec&pm:title=foo&emai
ls=sample@mail.com


Login credential: firstname.lastname (*)

Password: knime
*for double names and double surnames the whitespace has been removed

Using Components Exercise
Path to the workflow: Exercises à 02_working_with_components à

00_import_components
§ Drag&drop the Reading and Pre-Processing Data_v1 Component, and the

Customer Segmentation_v1 Component available on KNIME Server in the
workflow editor
§ Connect the two components accordingly and execute the workflow

Creating Components Exercise (1/3)

01_define_cluster_parameters
§ Filter only numeric columns (hint: use the Column Filter node)
§ Use the Integer Widget node to define the number of clusters Configuration:
§ min number of clusters 2
§ max number of clusters 10
§ default value: 4
§ Use the Column Filter Widget node to define the column to be included for
clustering

§ Use the Text Output Widget node to write the webpage description
Text for the WebPage (hint: use html as text format):
<h2>Define Cluster Parameters</h2>
<p>Set parameters to be taken into account in the following clustering.</p>
<p>Click 'Next' to start the clustering process.</p>
<P>If you do not know what a clustering process is, check <a
href="https://en.wikipedia.org/wiki/Cluster_analysis">Cluster Analysis</a> and specifically the <a
href="https://en.wikipedia.org/wiki/K-means_clustering">k-Means algorithm</a>.
§ Encapsulate the 4 created nodes in a component and configure 2 outports: one
for the Integer Widget node and one for the Column Filter Widget node


01_define_cluster_parameters
§ Define the layout of the items with in order to have the items ordered as shown
in the figure
Text Output
Widget
Column
Filter Widget

WebPortal Exercise
§ Access to the WebPortal (details available in KNIME Server Configuration slide)

§ Execute the workflow located at
Use Cases à Customer_Segmentation_Use_Case
§ Discard the job of the workflow once it has been executed (from KNIME
Analytics Platform!)

KNIME Server REST API
§ Integrate KNIME Server functionality with IT infrastructure

§ REST is a design pattern used for building networked applications
§ REST = Representational State Transfer
§ Communication based on HTTP
§ Usually clear text
§ Execute workflows, check server status, set permissions, and more
§ Entry point for the REST Interface is https://server-address/knime/rest/
See Blog Posts for detailed tutorials:

https://www.knime.org/blog/giving-the-knime-server-a-rest
https://www.knime.org/blog/the-knime-server-rest-api
https://tech.knime.org/wiki/using-knime-server-rest-api-for-file-uploads-and-downloads

§ All server functionality available via REST API

§ Programmatically control KNIME Server
§ Upload/download/delete resources
§ Staging from Dev to Prod
§ Upload licenses
§ Empty trash/restore items
§ Execute workflows
§ Schedule jobs
§ Set permissions
§ Create users & groups, etc.
§ Documentation available at:
/knime/rest/doc/index.html

§ Enables external integration

§ Build applications around KNIME Server Deploy KNIME workflows
as web services
§ e.g. for Microservices and real time scoring
Output in server
Input data response

SwaggerUI definitions of individual workflows (KNIME Server Medium and Large)
§ One of the key functionality of KNIME

Server is to allow individual workflows
to be exposed as REST endpoints
§ SwaggerUI interface allows to
document and test your web services
§ “Show API definition” is available in the
context menu item on the KNIME
Explorer. Selecting this option opens
the SwaggerUI page for that service in
your web browser.

SwaggerUI definitions of individual workflows

REST API – Use Case: Workflows Calling Workflows
§ Execute the workflow via Call Remote

Workflow node
§ Analyses input parameters
§ Prepare input data accordingly
§ Executes job and gets back results

New Node: Call Workflow (Table Based)
§ Makes it easier to call other

workflows using an entire
KNIME table
§ A caller workflow can send a
table and flow variables to a
callee workflow and receive a
table from the callee via the
Container Input (Table) node

New Node: KNIME Server Connection
§ Allows to connect to KNIME Server

§ After a connection has been
established, any of the remote file
handling nodes can be used with the
connected server
§ The server connection can also be used
together with the Call Workflow (Table
Based) node in order to run workflows
that are shared via a KNIME Server

REST API – Use Case: Integration with External Applications
§ The workflow can also be executed by external tools such as Postman or Curl
for debugging purposes
§ KNIME Server as backend for third party analytical applications

Data Science: Development != Deployment

Moving Data Science into Production
Issues:
• Development =!
Deployment
• Needs Copy/Paste, Rewrite
• Transport of models is non-
trivial
Þ Inefficient & Error prone

Side Note: It’s not just a Visual Workflow Issue
Creating Data Science Data Science in Production
# read data
raw_target_data = read_xls_data() productionize # Predictions, running as a flask service
# remove duplicates, handle missing values:
target_data = basic_data_cleanup_with_pandas(raw_target_data) # load saved components
raw_feature_data = fetch_db_data_using_psycopg2() feature_scaler,trained_RF = load_models_with_joblib()
# remove duplicates, handle missing values:
feature_data = basic_data_cleanup_with_pandas(raw_feature_data) # read and prepare data
raw_prediction_data = get_dataframe_from_request()
# basic feature engineering with sklearn prediction_data = basic_data_cleanup_with_pandas(raw_prediction_data)
feature_scaler = sklearn.preprocessing.StandardScaler().fit(feature_data) scaled_data = feature.scaler.transform(prediction_data)
standardized_features = feature_scaler.transform(feature_data)
filtered_feature_data = variance_feature_filter_with_sklearn(standardized_features,target_data) # generate predictions
predictions = trained_RF.predict(scaled_data)
# build model with sklearn prediction_probs = trained_RF.predict_proba(scaled_data)
training_feature_data,testing_feature_data,training_target_data,testing_target_data = predictions = join_tables_with_pandas(predictions,prediction_probs)
split_data_with_sklearn(filtered_feature_data,target_data)
RF_params = RF_hyperparameter_search_with_sklearn(training_feature_data,training_target_data) # return results from the service:
trained_RF = build_RF_using_params(training_feature_data,training_target_data,RF_params) return_dataframe_to_service(predictions)
# validate model
generate_validation_report(trained_RF,testing_feature_data,testing_target_data)
#--------------------
# save models
save_models_with_joblib((feature_scaler,trained_RF))


Capture and Combine Parts of a Workflow

Write Production Workflow from Captured Parts

KNIME Server Administration
Architecture
Central Components of KNIME Server
Executor(s)
KNIME Server Large Message
Queue
Workflow
Repository Request
Request
Tomcat
Request
…
Web Container
…
Client

KNIME Server – Scaling Options
§ KNIME Server supports ‘Scale Up’
§ KNIME Executors allows ‘Scale Out’

Scaling Out - KNIME Executors - KNIME Server Large
§ Run multiple executors on independent hardware

§ The distributed Executors rely on the KNIME REST interface and on a message
queueing system called RabbitMQ

KNIME Cloud Offerings – AWS and Azure
Features:
§ KNIME Analytics Platform
§ KNIME Server Small & Medium
§ KNIME Server Large BYOL
§ Supports Server Large with multiple
Executors
§ Has an embedded Executor so can be
stand-alone
§ KNIME Executors
§ Multiple Executors that can be used by
KNIME Server Large
§ Pay as you go (PAYG) offering supports
elastic scaling
https://www.knime.com/knime-software-on-amazon-web-services
§ Bring your own license (BYOL) offering
https://www.knime.com/knime-software-on-microsoft-azure uses cores from your Server license

Flexible Cloud deployments to meet computing needs
Mixed Cloud Usage

Virtual Private Cloud
Executors BYOL
Features:
§ Supplement traditionally licensed
Executor Executor Executors with Pay-as-you-Go
(PAYG) model
ı
Executor Executor
§ Meet periodic demand peaks
KNIME § Fulfill need for speciality hardware
Server Executors PAYG
(e.g. GPU‘s)
Executor Executor § Meet budgeting needs
ı
Executor Executor

Properties
148
Set Properties

Job Pools
§ Allows frequently executed workflows to stay in memory

§ Advantage: Eliminates the overhead of loading the workflow in an Executor for
each execution
§ Useful when job loading time is large compared to execution time
§ A job pool can be enabled by setting a property on the workflow that should be
pooled

Workflow Pinning
Features:
RAM
Executor
CPU
§ Match workflow needs to
Executor capabilities
KNIME
Executor
GPU CPU § Partition compute resources
Server
by capability, department,
RAM GPU
usage, …
Executor
§ Workflow needs determined
by workflow publisher

Workflow Pinning
§ Workflow pinning is useful when using multiple KNIME Executors

§ For a workflow that needs certain system requirements (e.g. specific hardware
(like GPU), extensions, or system environments (like Linux) it is possible to
define such Executor requirements
§ Only Executors that fulfill the requirements will accept and execute the workflow
job
§ The system admin of the Executors must specify a property for each executor
separately
§ The requirements can be defined by setting a property on a workflow, which is a
simple comma-separated list of user defined values.

Executor Groups
Supporting Executors in the Enterprise
Features:
CPU RAM Marketing
Executor § Logical groupings of
Group 1
Executors
KNIME Executor
CPU Database Finance
Finance § Match users/groups to
Server Group 2
Executor Groups
Executor
CPU GPU Engineering § Partition compute resources
Group 3 by groups, department, …
§ Partitioning managed by
Server administrators

KNIME Server Management Services
About Managing Preferences
§ Management (Client Preferences) makes IT operations easier by centrally

managing KNIME Analytics Platform preferences
§ Solution: Management (Preferences) part of the Management Tools available

with KNIME Server Medium and KNIME Server Large
§ Features:
§ Easier IT Operations.
§ Manage Analytics Platform preferences centrally
§ Include dependencies – e.g. driver files.
§ Deliver updates to configurations automatically
§ Simplifies Executor setup in distributed Executor environments

Management (Client Preferences)
§ Different departments/teams
have different requirements
§ Multiple OS deployments Marketing Finance R&D

§ Windows 7
§ Windows 10
§ Linux
§ macOS
Windows 10 Windows 7 Linux and macOS

Hive Oracle Python
Spark MS Access R

§ Client-profiles
§ Python-Linux
KNIME Server § Python-macOS
§ R-Linux
client-profiles
§ Databases-Win7
Databases- Python-macOS § Big Data-Win10
Big Data-Win10
Win7 (etc) § Executor
Profiles can include:

Preferences, drivers,
and more
Marketing Finance R&D

Client-profiles
§ Default
§ default.epf
KNIME Server § Python-Linux
§ python.epf
client-profiles § py-linux.sh
§ Python-macOS
Databases- Python-macOS
Big Data-Win10 § python-mac.epf
Win7 (etc)
§ py-mac.sh
§ R-macOS
§ R-mac.epf
§ Databases
§ db.epf
§ oracle.jdbc
§ msaccess.jdbc
§ Big Data
§ bigdata.epf
Marketing Finance R&D

1. Knime.ini
By adding lines to the knime.ini (file available in the same directory as the KNIME Analytics Platform
executable)
On application startup of KNIME Analytics Platform, KNIME Server is queried for the specified
preference profiles. Preferences are applied before finishing startup

2. Manually set a single preference
Allows users to browse available profiles

and add them on demand.

KNIME Server Maintenance
Admin Workflows
§ KNIME Server comes with a set of admin workflows

§ Admin workflows make use of KNIME Server REST API
§ Allows automated KNIME Server maintenance

Explore KNIME Server REST API (1/2)
§ Access KNIME Server REST API documentation pages:

http://3.123.122.13/knime/rest/doc/index.html
§ Explore different endpoints

Explore KNIME Server REST API (2/2)
Path to the workflow: Examples à REST à Predict Results Using REST API
§ Right click menu à Show API Definition
§ Explore Execution Endpoint: GET Request
§ Try out and execute from browser

Integrated Deployment (1/2)
§ Capture workflow parts using Capture Workflow Start & Capture Workflow End
nodes for
§ Data Preparation
§ Scoring
§ Combine captures using Workflow Combiner node
§ Write combined deployment workflow relative to the modeling workflow

Integrated Deployment (2/2)
§ Deploy this workflow to your user folder on KNIME Server

§ Execute workflow on KNIME Server
§ Inspect resulting deployment workflow


Login credential: firstname.lastname

Password: knime

Additional Info
§ We will keep KNIME Server up and running for an additional week to let you
play around a little bit more with it
§ Interested in a trial license? Just send me an email at

Q&A
Thank You!
180

Knime L3 Study Material

Uploaded by

Copyright:

Available Formats

Knime L3 Study Material

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Knime L3 Study Material

Uploaded by

Copyright:

Available Formats

[L3-PC] KNIME Server:

Productionizing and Collaboration

© 2021 KNIME AG. All rights reserved. 2

KNIME KNIME Community Partner Team End User API Managed

© 2021 KNIME AG. All rights reserved. 4

3000+ Nodes for all Steps of “End-To-End” Data Science

© 2021 KNIME AG. All rights reserved. 5

© 2021 KNIME AG. All rights reserved. 6

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 7

Open Source Commercial

© 2021 KNIME AG. All rights reserved. 8

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 9

© 2021 KNIME AG. All rights reserved.

Database Specialists Data Engineers Python Specialists

Data Science Specialists Visualization Specialists

© 2021 KNIME AG. All rights reserved. 11

Workflows, Components and Collaboration Features:

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 13

© 2021 KNIME AG. All rights reserved. 14

Delivering reports and output to business users appropriately

Guided Analytics for Building Applications

© 2021 KNIME AG. All rights reserved. 16

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 17

© 2021 KNIME AG. All rights reserved. 18

Automated workflow execution

© 2021 KNIME AG. All rights reserved. 19

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 21

Explainability / Interpretability of models

© 2021 KNIME AG. All rights reserved. 22

Data / Model Lineage

© 2021 KNIME AG. All rights reserved. 23

Model / ML Operations Operations Consumption

© 2021 KNIME AG. All rights reserved. 24

Central Management and Monitoring capabilities

© 2021 KNIME AG. All rights reserved. 25

OAUTH, LDAP, AD Integration

© 2021 KNIME AG. All rights reserved. 26

© 2021 KNIME AG. All rights reserved. 27

© 2021 KNIME AG. All rights reserved. 28

Drag & Drop

© 2021 KNIME AG. All rights reserved. 29

Server connections are shown as “mount

© 2021 KNIME AG. All rights reserved. 32

The Server provides an area in the

To move resources simply drag Workflows

© 2021 KNIME AG. All rights reserved. 33

Another way to deploy resources on KNIME Server..

© 2021 KNIME AG. All rights reserved. 34

§ By double-clicking a workflow on KNIME Server, the client downloads it (to a

© 2021 KNIME AG. All rights reserved. 35

Check to reset workflow before

If selected, the executed job is

Enter one or multiple email

The name of the workflow job as

© 2021 KNIME AG. All rights reserved. 37