No-code AI with Orange Data Mining

Hello students, welcome to Kwickprep. Can you train a machine learning model without typing a single line of code? Yes, with a free tool called Orange. Today we will learn what data mining is, how to start Orange, its main widgets, and workflows for data science, computer vision and NLP.

Let us begin with data mining. Data mining is the process of finding useful, hidden patterns in large amounts of data. It is like mining gold, where we dig through a lot of rock to find a little gold. Common tasks are classification, clustering, finding links between items, and prediction. For example, a supermarket may discover that bread and butter are often bought together, and place them near each other.

Orange is a tool for data mining and machine learning. It is free and open source, which means anyone can use it and see its code. You build a workflow by dragging and dropping blocks, with no coding needed. It was made at the University of Ljubljana in Slovenia. And it runs on Windows, macOS and Linux.

Here is how to get started. Download the installer from the official website, orangedatamining dot com. Install it like any other software, then open Orange. A new, empty canvas appears, ready for your first workflow. You can also open a ready example workflow from the welcome screen to explore.

Orange has four main components. The canvas is the large workspace where you build your workflow. A widget is a block that does one task, such as loading data or drawing a chart. A link is a line that carries data from one widget to the next, from its right side to the next widget's left side. The toolbox on the left holds the catalogue of all widgets, grouped by category.

Orange's default catalogue has six main categories. Data widgets load and view data, like File and Data Table. Transform widgets change data, like Select Columns and Preprocess. Visualize widgets draw charts, like Scatter Plot and Box Plot. Model widgets learn from data, like Tree, kNN and Logistic Regression. Evaluate widgets test models, like Test and Score and Confusion Matrix. Unsupervised widgets find groups without labels, like k-Means and Hierarchical Clustering.

Let us build a first workflow. Drag the File widget onto the canvas and choose the built-in iris flower dataset. Draw a link from File to a Data Table widget. Double-click Data Table to open it and see the rows of flower measurements. Then link File to a Scatter Plot, and you can see the three flower types as coloured dots.

Now a full data science workflow for classification. The File widget loads a patient dataset, where the target is whether a person has heart disease. Preprocess fills in missing values and scales the numbers. We connect two model widgets, Tree and kNN, to compare them. Test and Score checks each model on data it has not seen and shows accuracy. Confusion Matrix shows where each model was right and wrong. Finally, Predictions applies the better model to new patients.

Pause and predict. Test and Score shows the Tree model with accuracy zero point seven eight. The kNN model shows zero point eight four. Which model should we connect to Predictions? The kNN model, because it was correct more often on unseen data. Always choose using test results, never training results.

For images and text, we add extra widget packs called add-ons. Open the Options menu and click Add-ons. Tick Image Analytics for computer vision work. Tick Text Mining for natural language processing, or NLP, work. Then restart Orange, and the new categories appear in the toolbox.

Here is a computer vision workflow that sorts photos of cats and dogs. Import Images loads a folder, where sub-folders named cat and dog give the labels. Image Embedding turns each picture into a list of numbers that describe it, using a pre-trained deep network. Logistic Regression learns to separate cats from dogs using those numbers. Test and Score checks the accuracy, and Predictions labels new photos.

Finally, an NLP workflow on product reviews. A corpus is a collection of texts, and the Corpus widget loads the reviews. Preprocess Text makes words lowercase, splits them into tokens and removes common words like the and is. Bag of Words counts how often each word appears in each review. Word Cloud shows the most frequent words in large letters. Sentiment Analysis scores each review as positive or negative.

A few tips for your practical work. Data always flows from left to right along the links. Open a Data Table after any step to check what the data looks like. Save your workflow with File and Save, and it is stored as an OWS file. When you write answers, name the exact widgets you used, in order.

Let us revise what we learned today. Data mining finds hidden, useful patterns in large data. Orange is a free, drag and drop tool that needs no code. Its components are the canvas, widgets, links and the toolbox. Its default widgets fall into six categories, from Data to Unsupervised. And the Image Analytics and Text Mining add-ons let you build computer vision and NLP workflows.