Orange Data Mining: Components and the Widget Catalogue

Hello students, welcome to Kwickprep. Can you load data, draw charts and train a model just by dragging boxes and joining lines? With Orange, you can. Today we explore the Orange canvas, how widgets connect, the main widget groups, and we build a first workflow.

Let us first meet the tool. Orange is a free and open source tool for data mining, which means finding useful patterns in data. It was developed at the University of Ljubljana in Slovenia. It uses visual programming, so you drag boxes and join them instead of writing code. It runs on Windows, macOS and Linux, so any school lab can use it.

When Orange opens, you see a few main parts. The widget panel on the left lists all the widgets, grouped by category. The canvas in the centre is the big empty space where you build your workflow. A widget is a small round icon that does one task, like loading a file. A link is a line between two widgets that carries data. The help area explains the widget you select.

Now, how do widgets connect? Each widget has an output, shown as a curved edge on its right side. Many widgets also have an input on their left side. To connect, drag a line from the output of one widget to the input of another. Data then flows along the link, usually drawn from left to right. Double click any widget to open its window and change its settings.

What travels on a link is called a signal, through a channel. The Data channel carries a table of rows and columns, like from File to Data Table. The Learner channel carries a learning method that is not trained yet, like Tree into Test and Score. The Model channel carries a trained model, used by the Predictions widget. Evaluation Results carry scores, from Test and Score to Confusion Matrix.

Pause and predict. A student drags a line from Data Table back into File. Nothing connects. Why? File only produces data and does not take a data table as input. Always link output to input, so File goes to Data Table, not the other way.

The default widget catalogue groups widgets into categories, and each category has its own colour. Data widgets load and view data, like File. Transform widgets reshape and clean data, like Preprocess, and older versions keep these inside Data. Visualize widgets draw charts, like Scatter Plot. Model widgets hold learning methods, like Tree. Evaluate widgets test how good a model is. Unsupervised widgets find groups without labels, like k-Means.

Let us look inside each group, starting with Data. File loads a data file from your computer, like a CSV file or the built-in iris sample. Datasets downloads ready sample data sets, like heart disease data. Data Table shows the rows so you can check the values. Select Columns chooses which columns are features and which one is the target to predict. Data Sampler splits data, for example seventy percent for training.

Visualize widgets help you see the data. Scatter Plot places points using two features, like petal length and width. Box Plot shows the spread and middle value, called the median, like marks for each section. Distributions shows how often values occur, like students in each age group. Bar Plot compares values across categories, like sales in each city.

Model widgets contain learning methods. Tree builds a decision tree of yes or no questions, and it is easy to explain. kNN means nearest neighbours, and it labels a point by looking at the closest examples. Logistic Regression predicts the probability of a class, useful for two classes like pass or fail. Naive Bayes uses probability and works well for text, like spam. Random Forest builds many trees that vote, which often improves accuracy.

Now the last two groups. Test and Score, in Evaluate, measures accuracy using methods like cross validation. Predictions shows what a model predicts for each row. Confusion Matrix counts right and wrong predictions for each class. In Unsupervised, k-Means splits data into a chosen number of clusters, which are groups of similar rows. Hierarchical Clustering builds a tree showing how clusters join together.

Let us build a first workflow with the iris flower data, which ships with Orange. First, place a File widget, which loads iris by default. Link it to Data Table to check the one hundred fifty rows. Link File to Scatter Plot to see how petal size separates the three flower types. Add Tree and kNN, and link both of them, along with File, into Test and Score. Test and Score compares the accuracy of the two learners. Finally, link Test and Score to Confusion Matrix to see which flowers get mixed up.

In Test and Score, look at the column called C A, short for classification accuracy. Tree scores roughly zero point nine four, which means about ninety four flowers in a hundred are classified correctly. kNN scores about the same, near zero point nine five, though your exact numbers may differ slightly. The Confusion Matrix shows that most mistakes happen between versicolor and virginica, two similar flower types.

Let us revise what we learned today. Orange is a free visual tool where you drag widgets onto a canvas. Links carry data from the output on the right to the input on the left. Widgets are grouped as Data, Transform, Visualize, Model, Evaluate and Unsupervised. Key widgets include File, Tree, kNN and Test and Score. Your first workflow joins File, Data Table, Scatter Plot, learners, Test and Score and Confusion Matrix. Open Orange and build it yourself today.