11

Thinking Pi

TUTORIAL

iris[‘target’] contains the output flower name

(encoded as a 0,1, or 2) for each flower row of the

input data matrix. Run ‘iris[‘target’]’ and you can

see the first 50 are all 0s, so the first 50 rows from

the input data matrix are all the same flower type.

Each number corresponds to the flower names in

iris[‘target_names’]. 0 is I. setosa, 1 is I. versicolor,

and 2 is I. virginica.

You can get more information on the dataset by

printing out the DESCR of the object:

print(iris[‘DESCR’])

OK, so we have our data! The machine learning

algorithm that will learn the mapping between the

flower features to the flower name will be a decision

tree. Decision trees are models that try to build a

tree of questions that split the data into the separate

classes (flower types).

We’ll start by importing the algorithm from

scikit-learn.

from sklearn.tree import DecisionTreeClassifier

We want to train our model on some of the data,

and save a portion of our data for testing. It would not

be wise to test a student by just getting them to do

an exam they have been practising with and already

have the answers for, right? They could just memorise

the answers without learning the pattern. So, you give

them a different exam that they don’t already have the

answers to, and compare the answers they ‘predict’

with the real ‘target’ answers.

So, we split the data into training and test data, and

also shuffle the data rows so that there is a good mix

of each flower in both train and test data.

The train_test_split function in scikit-learn does

all of this for us and, by default, splits the data so that

25% is put into test, and 75% for training.

from sklearn.model_selection import train_test_

split

X = iris[‘data’]

y = iris[‘target’]

X_train, X_test, y_train, y_test = train_test_

split(X, y)

Now we create a new decision tree model and

train/fit it to our training data. This is the training

phase, so it gets to look at the inputs (flower features

stored in X_train) alongside the outputs (flower

names stored in y_train).

Left

Example of simple

decision tree model

after learning

iris problem

We split the data into training and test data,

and also shuffle the data rows so that there

is a good mix of each flower in both train and

test data

”

”

Petal Length > 2.4

Setosa Petal Length < 1.4

Versicolor Virginica

YES

YES

NO

NO