Decision Trees

Question 1

Gehören Entscheidungsbäume zum überwachten oder unüberwachten Lernen?

Answer 1

zum überwachten Lernen

Question 2

Was sind die Vorteile von Entscheidungsbäumen?

Answer 2

Question 3

Was sind die Nachteile von Entscheidungsbäumen?

Answer 3

Question 4

Was bedeutet eine geringere Entropie?

Answer 4

weniger Unsicherheit

Question 5

Wie berechnet man die Entropie eines Datensets?

Answer 5

Question 6

Wie bestimmt man die Information (entropy) to classify a tuple in D?

Answer 6

Question 7

Wie bestimmt man die Average information needed, after using attribute A to split D into k partitions?

Answer 7

Question 8

Wie berechnet man den Information Gain?
(Information gained by branching on attribute age)

Answer 8

Question 9

Wie bestimmt man die Gini Impurity?

Answer 9

Question 10

Wie bestimmt man die Qualität eines Splits anhand der Gini Impurity?

Answer 10

Question 11

Wie berechnet man den GainRatio?

Answer 11

Question 12

Wozu braucht man den GainRatio?

Answer 12

Weil die Gini Impurity biased gegenüber Splits mit mehreren unterschliedlichen Klassen ist, über die, die Daten aufgeteilt werden

Question 13

Wie würde das Gain Kriterium aussehen, wenn man fehlende Datan an einem Attribut hat?

Answer 13

Question 14

Was ist Pruning?

Answer 14

Das entfernen von Subbäumen und austauschen durch Blätter nach der Buildingphase, um Overfitting zu vermeiden

Question 15

Was ist das Problem beim Prepruning?

Answer 15

Es ist schwierig einen threshold vorher zu definieren

Question 16

Wie bestimmt man den Prediction Error, um über Pruning zu entscheiden?

Answer 16

Question 17

Nenn Limitationen von Entscheidungsbäumen

Answer 17

• Construction / search
  process depends only
  on local information
• Greedy: current best split does not consider future splits

Question 18

Wie kann man mit Entscheidungsbäumen regressieren?

Answer 18

Question 19

Sind Gini coefficient und Gini impurity das gleiche?

Answer 19

Nein

Question 20

Entropy is small if the distribution is heavily skewed.
Stimmt das?

Answer 20

Ja

Question 21

Gini-impurity and information gain are biased towards attributes with a small number of values.
Stimmt das?

Answer 21

Nein

Question 22

With the rule extraction algorithm, it is possible to find rules that perfectly distinguish between different classes for any dataset.
Stimmt das?

Answer 22

Nein

Question 23

Which of the following statements are correct?
1. Decision trees can only handle discrete or continuous attributes, but no mixed attributes.
2. Decision trees are especially good at dealing with connected attributes / doing diagonal classification.
3. Greedy decision tree algorithms most often generate globally optimal trees.
4. If any k out of n class conditions are met, the tree needs (𝑛 über 𝑘) regions for this class

Answer 23

4.

Question 24

Which of the following statements about information gain are correct?
1. To decide on how to split the data, the information gain has to be computed only for one attribute.
2. Information gain has to be maximized.
3. Information gain Gain(A) is the difference of the entropy without splitting and the average entropy of the partitions created by using attribute A to split the data.
4. Splitting based on information gain favors large inner-partition entropy.

Answer 24

2 und 3

Question 25

Which of the following statements about continous attributes are correct?
1. Any threshold value between two adjacent values of a continuous attribute has the same effect for splitting.
2. It’s possible to handle continuous classes by testing all possible splits of an attribute.
3. Information gain based methods can’t handle continuous attributes.
4. Gini impurity assumes all attributes are continuous.

Answer 25

1 und 4

Question 26

Which of the following statements are correct?
1. Gini impurity assumes that all attributes are categorical.
2. Gini impurity has to be maximized.
3. Gini impurity is at its maximum value if all the records are equally distributed among all classes.
4. Gini impurity has to be calculated for every attribute and every possible split of each attribute.

Answer 26

3 und 4

Question 27

Which of the following statements are correct?
1. Gini impurity is at its maximum value if all the records are equally distributed among all classes.
2. Gini impurity has to be maximized.
3. Gini impurity assumes that all attributes are categorical.
4. Gini impurity has to be calculated for every attribute and every possible split of each attribute.

Answer 27

1 und 4