11

Thinking Pi

FORGE

model = DecisionTreeClassifier()

model.fit(X_train, y_train)

That’s it! Now you get predictions on the test data:

model.predict(X_test)

You can see how many of these it predicts correctly

by manually comparing them to the answers stored

in y_test.

You should find that they’re mostly correct,

but we can check the model’s performance in

an easier way. Scikit-learn models have a .score

function that will tell you what ratio of predictions

it gets right for some input data, alongside

their answers:

model.score(X_test, y_test)

We got a result of 0.9473684, which means we

were 94% correct. It’s not perfect, but pretty good for

a few lines of code.

Now, because we are on a Raspberry Pi using

Python, we can control anything we want with the

output. Some LEDs perhaps? We can import the

gpiozero library, which will allow us to easily control

hardware using Python.

from gpiozero import LED

# connect an LED to pin 17 on the Raspberry Pi

led = LED(17)

We can create a function that will turn on the LEDs

if we predict a flower to be of type setosa (output

label of 0).

def led_on_if_setosa(input_data):

# input data is a list with the following

values:

#[sepal length (cm), sepal width (cm), petal

length (cm), petal width (cm)]

prediction = model.predict([input_data])[0]

if prediction == 0:

led.on()

else:

led.off()

Let’s see if we a found a setosa flower:

led_on_if_setosa([ 4.7, 3.2, 1.3, 0.2])

Hopefully your LED has turned on.

This is the basic structure for code to bring machine

learning to physical computing projects. You need

some training data, a learning algorithm, and a way

of performing actions depending on predictions. If an

automatic flower identifying kit isn’t what you’re after,

then you can send almost any type of data into this

system (provided it’s not too noisy). Temperature and

other environmental sensors can work really well, but

it depends on what you want to control.

Above

Iris versicolor, also

known as the Blue

Flag or Purple Iris,

is the official flower

of Quebec

Credit

Danielle Langlois

CC-BY-SA

IRIS DATASET

The iris dataset is one of the most commonly

used in machine learning. It was first presented

by Ronald Fisher in the paper, The use of multiple

measurements in taxonomic problems, released

in 1936. The four attributes (Sepal Length, Sepal

Width, Petal Length, and Petal Width) combine to

determine the species, though no one can do it

by itself.

It’s a great dataset to get started with. If you

want to try your new-found machine learning skills

with more inputs, there are some other datasets at

hsmag.cc/HGGFOa that you can download

and use (though some of them may need a little

manipulation before they’re in a suitable format).