Getting Started with Orange 20: Multivariate Projection - Freeviz Video Lecture | Orange Data for Beginners - Data & Analytics

We love our data,
but most of the time our data is big, with many variables,
and making sense of it is a difficult task.
Well not so difficult in Orange.
Let us load the iris data that we have used in some of our previous videos
and connect it to Scatter Plot.
Scatter plots are great!
They can show us the relationship between two variables
But our data rarely has only two variables.
Even iris data has four.
How could I see the relationship between more than two features?
With FreeViz.
FreeViz is a multi variable projection method
for uncovering feature interactions
in a data with class variable.
If we pass iris data to FreeViz
the widget would initially place the variables on an invisible circle,
making them equally important.
Press Optimize,
and FreeViz will rearrange the projection,
and find the one that best separates points of different classes.
By doing this, optimization exposes the most relevant features of the data.
Variables associated with longer axes
are more important for a specific class value.
And those variables that lie closer together
are more correlated, too.
Iris setosa flowers, which are marked with blue circles,
have high values of sepal width,
and large petals are distinctive for iris virginica.
On the other hand sepal length does not play any role.
Let us see this on a different example.
The zoo data set contains 100 animals classified
into seven groups including mammals,
fish, bird and so on.
We will use FreeViz projection
to investigate relationship between all 15 variables
and see whether there's some structure in the data.
In the File widget simply change iris to zoo
with Browse documentation data sets.
In FreeViz, we instantly see the unoptimized projection.
Let's use Optimize again
to reveal the relations between features and animal classes.
We can,even use 'Show class density'
to observe whether the data projection in FreeViz
really managed to distinguish between different class values.
It looks like it does.
Having hair and giving milk
is a distinguishing property of mammals
while being aquatic is a property of fish.
On the other side of the projection
are those animals that lay eggs and have feathers.
This means mammals do not have any of those two properties
since they are placed opposite of the two features.
Finally, at the center of the plot,
we see features that are not very informative,
such as animal domesticity.
We can exclude these features from the graph
by increasing the blackout radius.
Visualizations make data interpretation so much easier.
In this video, we learned how to uncover interesting relations
between classes and features
in a class label data set
and how to interpret the FreeViz projection.