Building AI with Visual Workflows vs. Python code. Part 1: "A Picture is worth a thousand words"
I was humbled by the response to my recent article, "Is Corporate Australia Missing the AI Revolution. A postgrad's challenge." A few of you reached out asking to explore both sides of the argument. It was clear that many of you were not aware of any visual tools, as your IT departments were only pitching the need for Data Scientists without an alternative.
This post is the first in a series that will tackle a key debate: visual, low-code platforms like KNIME and SAS Viya vs. traditional, text-based programming in languages like Python.
The goal of this first post isn't to flood you with information, but to show you the visual difference between the two approaches. For many, this will be an eye-opener.
For this post, I’ll use a predictive model I created in KNIME to forecast factory sales for the next 12 months. The entire end-to-end process was a single, visual workflow—from ingesting and cleaning data to training and scoring eight different forecasting models to identify the most accurate algorithm.
As you can see from the image below, the entire pipeline is a transparent, intuitive canvas. Once we narrow down the 2 best algorithims for forecasting future sales, we use these to forecast future values, with all results easily passed to Power BI for a final visualization.
Below is a description of the 9 stages of the model (highlighted by the red numbers in the image below)
The 9 Stages of a KNIME Predictive Model
1- Extract, Transform, and Load Data: Imports source data, performs transformations, and fills in missing values.
2 - Descriptive Data Analytics: Generates various plots (Box Plot, Decomposition, Lag Plot, ACF/PACF) to understand the source data.
3 - Data Partitioning & Signal Decomposition: Splits the data into a training and a test set for model development and evaluation.
4 - In-Sample Forecasting Models: Trains and scores eight different forecasting algorithms against the real values to assess accuracy.
5 - Training Results Comparison: Appends the score statistics from the training models to an Excel file for easy comparison.
Recommended by LinkedIn
6 - Testing Results Comparison: Appends the score statistics from the test models to an Excel file for evaluation on unseen data.
7 - Power BI Connection: Sends the training and testing statistics to Power BI to generate comparison charts.
8 - Forecasting Future Values - ARIMA: Uses the most accurate algorithm (ARIMA) to forecast future sales for the next 12 months.
9 - Forecasting Future Values - Gradient Boosting: Uses a second highly accurate algorithm (Gradient Boosting) to also forecast future sales for the next 12 months.
Now, let's look at the alternative: a text-based Python program designed to replicate this exact same process. While conceptually similar, the Python code would be hundreds (or potentially thousands) of lines long if we wanted the ability to show intermediate outputs at every single stage—a feature that is effortless in the visual KNIME environment.
This is the key point: The KNIME workflow is a language that a business analyst with machine learning training can develop, read, interpret, and maintain. The Python code, while powerful, is cryptic to a non-coder and incredibly difficult to maintain once the original coder leaves the company.
This visual snapshot shows why KNIME is so much better for democratizing data science, empowering business experts to build and manage sophisticated models themselves.
What are your thoughts on this fundamental difference?
Paul Witschey is a data-driven senior business improvement professional who partners with organisations to transform their business strategy, operating model, and processes to achieve measurable improvements in operational efficiency and effectiveness.
He has over 25 years of experience working with blue-chip companies in 5 continents, including 12+ years of management consulting experience with three management consulting firms. He holds a Bachelor of Business Administration from the McCombs School of Business at the University of Texas at Austin, a Master in Business Administration and a Master of Business Informatics from the Rotterdam School of Management at Erasmus University in the Netherlands, and a Master of Business (Data) Analytics from the Sydney Business School at the University of Wollongong in Australia.
#AI, #DataAnalytics, #Productivity, #DigitalTransformation, #BusinessIntelligence, #Innovation, #Knime, #SASViya, #FutureOfWork, #UOW, #CorporateAustralia, #DataScience, #Analytics #LowCode, #Python #BusinessIntelligence #AIRevolution
