In this course, we will learn all the core techniques needed to make effective use of H2O. Even if you have no prior experience of machine learning, even if your math is weak, by the end of this course you will be able to make machine learning models using a variety of algorithms. We will be using linear models, random forest, GBMs and of course deep learning, as well as some unsupervised learning algorithms. You will also be able to evaluate your models and choose the best model to suit not just your data but the other business restraints you may be under.
Decision Trees
Loading...
来自 H2O 的课程
Practical Machine Learning on H2O
27 个评分
从本节课中
Trees And Overfitting
与讲师见面
Darren Cook
So, decision trees are structured form of human thinking.
At least, when humans are thinking logically.
Here's an example of decision tree.
It's a good trend from last week,
we all use dataset.
Our first question is how long are the petals?
Because it turns out, all those setosa have short petals.
So this one rule,
we can get 100 percent accuracy at least based on the training data,
based on all the others that we've seen for that one class.
When they're longer than that length,
we come to another question.
And this question helps us split between the remaining two classes.
And in this case, we're looking with the petal width.
This gives us about 90 percent accuracy split.
To split them perfectly,
we would need more questions,
more depth to the tree.
But then we would also start where it would fit.
So as you saw that,
decision trees are natural classifiers.
But they can be used for regression as well.
Simply by averaging the values of each of the leaves.
As an example, it's not official example.
It's a bit silly to predict someone's age.
You can say we're asking do they have a walking stick and if they don't,
we ask how tall are they.
And that give us one of those three ages.
With more realistic examples,
regression decision trees are actually quite a reasonable tool.
But this example was probably based on maybe just a dozen records.
And when we don't give them enough data,
decision trees can easily over-fit.
You can see they're just partitioning based on whatever data they can
see at each level in the tree.
Real decision trees are often much deeper than
these artificial examples I just showed you.
So, although explainability is often given as a good reason
to use decision trees and the other tree algorithms we'll look at this week.
In practice, realistic examples tend to be large
complex and so the tree diagrams
themselves tend to be large complex and not that helpful for explainability.
But anyway, let's go on to look at some enhancements of the decision tree idea.
