Loading the Iris Dataset with scikit-learn

Every machine learning project begins with data, and the first practical skill is getting that data into your program. This article follows one small, complete step: loading a classical dataset called the Iris dataset so that Python can work with it. Before reading the explanation, run the code below and watch what each line does. The explanation will make more sense once you have seen the real execution.

The surprise in that reveal is the key idea of this article. The call does not hand you the measurements directly. It hands you one container, and the measurements and the correct answers are stored inside that container in separate places. Keep that in mind as the rest of the article traces where each half lives.

You do not have to find, download, and clean up the Iris data yourself in this lesson. The library ships with the data, so retrieving it takes a single function call. The goal here is to understand that retrieval step and the structure of what it returns. Exploring the contents of the dataset in depth comes later.

Because the Iris dataset appears constantly in scikit-learn tutorials, it is a common first dataset for learning how supervised data is organized. Its importance comes from its shape and its simplicity, not from any special behavior of the code that loads it.

There are two separate actions in the loading code, and it helps to keep them apart. The first is the import statement, from sklearn.datasets import load_iris. It tells Python where to find the name load_iris. Importing does not run the function. It only makes the name available.

The second action is the call, load_iris(). The parentheses are what make the call happen. The function then returns a value, and the assignment statement iris_dataset = load_iris() stores that returned value in the variable iris_dataset. Reading the line from right to left matches the order of events: the function runs first, then the result is stored under a name.

The object returned by load_iris is a Bunch object. It is very similar to a dictionary: it contains keys and values. A key is a name you use to look something up, and the value is what that name points to. In this dataset, the two keys you need are data, which holds the measurements, and target, which holds the correct answers.

The diagram shows the split that happens next. The data key supplies the X array, and the target key supplies the y array. You access a key by writing the object name, a dot, and the key name, so iris_dataset.data gives you X and iris_dataset.target gives you y. Both come from the same returned object, which is why they describe the same samples.

The first array is called X. It holds the data itself, the measurements that describe each example. X is two-dimensional: it has one row per data point and one column per feature. For the Iris dataset, each row is one flower, and the columns are four measurements taken from that flower. That gives a total of 150 rows, one for each flower, and 4 columns, one for each measurement.

The diagram uses a made-up array with four rows and two columns, so that every number is easy to follow by hand. Each row is one sample, and the two values in that row are its two features. The same reading applies to the real Iris array, where each row carries four feature values instead of two. Reading the index tells you the sample: X[2] is the third row, which is the third sample.

The second array is called y. It holds the correct or desired outputs that go with each piece of data. y is one-dimensional: it has one entry per sample and no column structure. Each entry is one integer class label from 0 to 2, so each flower's correct answer is a single whole number chosen from three possible classes.

Position 2 in y is the answer for row 2 in X. In the made-up example, X[2] is [5, 6] and y[2] is 2, so together they describe one sample and its correct label. The two arrays must be read together, because neither one means much alone. The label 2 is meaningless without the row it belongs to, and the row [5, 6] is only a training example once its answer is attached.

When a result is not what you expected, trace the line where the value is created or stored. A missing set of parentheses is a common cause: the code runs without an error, yet the variable holds the wrong kind of object. Compare what the line actually does with what you expected it to do, and the mismatch usually points straight to the line at fault.