!
""#$%&'%()*)+
&!,-#%').'!/%#%$
Q1 to Q8, Choose the correct option:
1. Which of the following in scikit-learn library is used for hyper
parameter tuning?
a. GridSearchCV
b. RandomizedCV
c. K-fold Cross Validation d. All
of the above
Answer: a. GridSearchCV and b. RandomizedCV
2. In which of the below ensemble techniques trees are trained in
parallel?
a. Random Forest. b.
Adaboost
c. Gradient Boosting d. All
of the above
Answer: a. Random Forest
3. In machine learning, if in the below line of code:
sklearn.svm.SVC(C=1.0, kernel='rbf', degree=3)
we increasing the Chyper parameter, what will happen?
a. The regularization will increase b. The
regularization will decrease c. No effect on
regularization
d. Kernel will be changed to linear
Answer: a. The regularization will increase
4. Check the below line of code and answer the following questions:
sklearn.tree.DecisionTreeClassifier(* criterion='gini',splitter='best',
max_depth=None, min_samples_split=2). Which of the following is true
regarding max_depth hyper parameter?
!""#$%&'%()*)+
a. It regularizes the decision tree by limiting the maximum
depth up to which a tree can be grown
b. It denotes the number of children a node can have c.
Both A & B
d. None of the above
Answer: a. It regularizes the decision tree by limiting the maximum depth
up to which a tree can be grown
5. Which of the following is true regarding Random Forests?
a. It's an ensemble of weak learners
b. The component trees are trained in series
c. In case of classification problem, the prediction is made by
taking mode of the class labels predicted by the component trees
d. None of the above
Answer: a. It’s an ensemble of weak learners and
6. What can be the disadvantage if the learning rate is very high in
gradient descent?
a. Gradient Descent algorithm can diverge from the optimal
solution.
b. Gradient Descent algorithm can keep oscillating around the
optimal solution and may not settle c.
Both of them
d. None of them
Answer: c) Both of them
7. As the model complexity increases, what will happen?
a. Bias will increase, Variance decrease
!""#$%&'%()*)+
b. Bias will decrease, Variance Increase c. Both
bias and variance increase
d. Both bias and variance decrease
Answer: b) Bias will decrease, Variance increases
8. Suppose I have a linear regression model which is performing as
follows: Train accuracy=0.95 and Test accuracy=0.75. Which of the
following is true regarding the model?
a. Model is underfitting b.
Model is overfitting
c. Model is performing good d.
None of the above
Answer: b. Model is overfitting
Q9 to Q15 are subjective answer type questions, Answer them briefly.
9. Suppose we have a dataset which have two classes A and B. The percentage of
class A is 40% and percentage of class B is 60%.
Calculate the Gini index and entropy of the dataset?
Answer: To calculate the Gini index of the dataset, we use the formula:
2 2
=1−( 1 + 2 )
Where;
• p1 is the probability of class A
• p2 is the probability of class B.
In this case, p1 = 0.4 and p2 = 0.6, so,
2 2
=1−(0.4 +0.6 )=0.48
To calculate the entropy of the dataset, we use the formula
= − 1 ( 1) − 2 ( 2)
In this case, p1 = 0.4 and p2 = 0.6, so,
= −0.4 (0.4) − 0.6 (0.6) = 0.97
𝐺
𝑒
𝑒
𝑛
𝑛
𝑖
𝑡
𝑛
𝑡
𝑟
𝑟
𝑖
𝑜
𝑜
𝑝
𝑝
𝑦
𝑦
𝐺
𝑖
𝑛
𝑖
𝑝
𝑙
𝑙
𝑜
𝑜
𝑔
𝑔
𝑝
𝑝
𝑝
𝑝
𝑙
𝑜
𝑙
𝑔
𝑜
𝑔
𝑝
!""#$%&'%()*)+
10.What are the advantages of Random Forests over Decision
Tree?
Answer: Some advantages of Random Forests over Decision Tree
are:
• Random forests are less prone to overfitting than a single
decision tree.
• Random forests can handle missing values and maintain
accuracy.
• Random forests can be used for both classification and
regression problems.
• Random forests can provide feature importance, which can
be used for feature selection.
11.Whatis the need of scaling all numerical features in a
dataset? Name any two techniques used for scaling?
Answer: Scaling all numerical features in a dataset is important because many
machine learning algorithms, such as those based
on distance measures, are sensitive to the scale of the features. Without scaling,
these algorithms would be affected by the presence of large scale features and
small scale features. Two techniques used for scaling are Min-Max Scaling
and Standardization.
12.Write down some advantages which scaling provides in
optimization using gradient descent algorithm?
Answer: Scaling provides the following advantages in optimization
using gradient descent algorithm:
!""#$%&'%()*)+
• It helps to converge faster by reducing the oscillations in the
optimization path.
• It helps to nd the global minimum by reducing the chances
of getting stuck in a local minimum.
13. In case of a highly imbalanced dataset for a classification
problem, is accuracy a good metric to measure the performance
of the model. If not, why?
Answer: In case of a highly imbalanced dataset for a classification problem,
accuracy is not a good metric to measure the performance of
the model. This is because accuracy does not take into account the
imbalance of the classes and can be misleading. Other metrics such as
precision, recall, F1-score, and AUC-ROCare more suitable for such cases.
14. What is “ f-score" metric? Write its mathematical formula.
Answer: F-score is a metric that balances precision and recall and is
commonly used in classification problems. The mathematical
formula for F-score is:
− = (2 ∗ ∗ )/( + )
15. What is the difference between fit(), transform() and
fit_transform()?
Answer: fit() is used to train a model on the training data,
transform() is used to apply a pre-trained model to the new data and
fit_transform() isused to train amodel on the training data and then apply
it to the new data.
The difference between fit() and transform() is that fit() learns the
parameters of the model, i.e., it trains the model on the training
fi
𝐹
𝑠
𝑐
𝑜
𝑟
𝑒
𝑝
𝑟
𝑒
𝑐
𝑖
𝑠
𝑖
𝑜
𝑛
𝑟
𝑒
𝑐
𝑎
𝑙
𝑙
𝑝
𝑟
𝑒
𝑐
𝑖
𝑠
𝑖
𝑜
𝑛
𝑟
𝑒
𝑐
𝑎
𝑙
𝑙
!""#$%&'%()*)+
data, while transform() applies the already trained model to the new data. On
the other hand, fit_transform() combines the functionality of fit() and
transform() in one step, it first trains the model on the training data and then
applies it to the new data in
one step.
