The IIT Madras BS Machine Learning Practice (MLP) End Term paper sat on 30 Apr 2023, in the January 2023 term, set QPD1-S2: 36 questions for 100 marks in 180 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.
______________ generates a bunch of normally-distributed clusters of points with specific mean and standard deviations for each cluster.
sklearn.datasets.make_clusters()
sklearn.datasets.make_centers()
sklearn.datasets.make_normal_clusters()
sklearn.datasets.make_blobs()
Correct answer
sklearn.datasets.make_blobs()
Why is data preprocessing necessary?
Some columns have values only between 0 and 1
The data is divided into multiple types of files i.e. html, csv, tsv, etc.
The data has only numbers in all the columns.
Correct answer
The data is divided into multiple types of files i.e. html, csv, tsv, etc.
Which of the following is correct with respect to R2-score?
R2 score is always positive and it may go up to infinity.
R2 score is always positive, but it ranges between 0 and 1 only.
R2 score can be negative. That happens if our model is worse than the mean model.
R2 score can be negative. That happens if the mean model is worse than our model.
Correct answer
R2 score can be negative. That happens if our model is worse than the mean model.
Which options represent the output of the following code block?
Correct answer
Consider the following code block:
from sklearn.datasets import make_regressionX, y = make_regression(n_samples = 10, n_features = 3)
from sklearn.model_selection import LeavePOutlpo = LeavePOut(p = 2)count = 0for train, test in lpo.split(X): print(train, test) count += 1print(count)What will be the value of the ‘count’?
20
30
45
15
Correct answer
45
Consider the following code block:
Correct answer
Which of the following is a hyperparameter?
L1-ratio in elasticnet
Pruning parameter in a decision tree
Learning rate in SGDRegressor
All of these
Correct answer
All of these
DecisionTreeClassifier
RandomForestRegressor
LogisticRegressor
SGDRegressor
Correct answer
SGDRegressor
Consider the following code
from sklearn.pipeline import Pipelinefrom sklearn.preprocessing import StandardScalerfrom sklearn.linear_model import LogisticRegression
pipe = ([ ('scaler', StandardScaler()), ('softmax', ____________ )])Which of the following options is True for blank space if I want to train the above pipeline as softmax regression ?
Correct answer
Consider the following code:
0.66
0.72
0.80
1.00
Correct answer
0.72
Which of the following are correct about sklearn.svm.SVC:
It will print number of support vectors
It will print an array of support vectors
It will print an array of probabilities representing distance from decision boundary with each data point.
It will print indices of the support vectors from the training set.
Correct answer
It will print an array of support vectors
It will print nearest neighbours from the test point
It will print indices of and distances to the neighbouring points (in training set) from test point
It will print indices, distance and nearest training point from the test point
It will throw an error
Correct answer
It will print indices of and distances to the neighbouring points (in training set) from test point
Which of the following APIs only supports conjoining transformers and estimators in series (i.e. one after another)?
Pipeline
ColumnTransformer
FeatureUnion
All of these
Correct answer
Pipeline
Consider the following ML task/steps for a regression dataset:
1. Read the data from a file (named ‘dataset.csv’). It has 7 columns. The last column is the target variable, all 6 features numerical.
2. Remove rows which have target values missing.
3. Fill the missing values in the features by KNN using 3 nearest neighbours.
4. Split the data into training and test sets. Take randomly the 70% of rows in the training set and the rest of them into the test set.
5. Train a simple linear regression model, with intercept, on the training set.
6. Report R2 score on the test set.
Which of the following code snippets correctly accomplishes the above task? Assume necessary imports.
Correct answer
(10, 3)
(10, 7)
(10, 8)
(10, 10)
(10, 11)
Correct answer
(10, 7)
Consider the following statements about SGDClassifier:
1. It can be used to train a model on large dataset that doesn’t fit in main memory
2. It can emulate a KNN model
3. It can emulate a decision tree model
4. It can emulate a perceptron
Choose the correction option(s)
1 and 4
1 and 2
2 and 3
3 and 4
Correct answer
1 and 4
Consider an image classification task: an image can have dogs, birds and trees. An image can have any combination of these three. The classifier is expected to report all these three for every sample. What kind of classification problem is this?
multi-label and multiclass problem.
multi-label and binary class problem.
multiclass problem.
binary class single label problem.
Correct answer
multi-label and multiclass problem.
Consider following code snippets, assuming necessary imports.
Correct answer
Which of the following options is true for the gamma parameter in a non-linear soft margin SVM?
For high values of gamma, the points need to be very close to each other in order to be considered in the same class
For low values of gamma, the points need to be very close to each other in order to be considered in the same class
Gamma doesn’t affect the SVM model at all
None of these
Correct answer
For high values of gamma, the points need to be very close to each other in order to be considered in the same class
Consider the following code:
from sklearn.tree import DecisionTreeClassifierfrom sklearn.datasets import load_wineX,y = load_wine(as_frame = True, return_X_y = True)
dtc1 = DecisionTreeClassifier(ccp_alpha = 0.0)dtc1.fit(X, y)dtc2 = DecisionTreeClassifier(ccp_alpha = 0.06)dtc2.fit(X, y)
dtc3 = DecisionTreeClassifier(ccp_alpha = 0.1)dtc3.fit(X, y)
dtc4 = DecisionTreeClassifier(ccp_alpha = 0.03)dtc4.fit(X, y)Which model is likely to overfit the most?
dtc1
dtc2
dtc3
dtc4
Correct answer
dtc1
Following is the code to tune the n_estimators parameter of a Bagging Classifier model.
from sklearn.model_selection import GridSearchCVfrom sklearn.pipeline import Pipelinefrom sklearn.preprocessing import StandardScalerfrom sklearn.ensemble import BaggingClassifier
param_grid = [{______________: [200, 300, 400, 500, 600]}]pipeline = Pipeline(steps=[('scaler', StandardScaler()),('bc', BaggingClassifier())])grid_search = GridSearchCV(pipeline, param_grid, cv=5,scoring='neg_mean_squared_error',return_train_score=True)grid_search.fit(X_train, y_train)What should the blank space contain?
Correct answer
Consider the following code. How many DecisionTreeClassifier models will be trained internally?
20
200
8
150
15
80
Correct answer
80
Consider two classifiers as shown in the following block of code:
from sklearn.datasets import load_winefrom sklearn.tree import DecisionTreeClassifierfrom sklearn.model_selection import train_test_splitX,y = load_wine(as_frame = True, return_X_y = True)X_train,X_test,y_train,y_test = train_test_split(X, y, test_size = 0.2, random_state = 1)clf1 = DecisionTreeClassifier(min_samples_split = 3, min_samples_leaf = 2, random_state = 5)clf1.fit(X_train, y_train)
clf2 = DecisionTreeClassifier(min_samples_split = 6, min_samples_leaf = 4, random_state = 5)clf2.fit(X_train, y_train)What can we say about the depths of the classifiers clf1 and clf2?
depth(clf1) ≥ depth(clf2)
depth(clf1) ≤ depth(clf2)
depth(clf1) = depth(clf2)
Insufficient Information
Correct answer
depth(clf1) ≥ depth(clf2)
Consider the following code:
from sklearn.tree import DecisionTreeClassifierfrom sklearn.datasets import load_wineX,y = load_wine(as_frame = True, return_X_y = True)
dtc1 = DecisionTreeClassifier(ccp_alpha = 0.0)dtc1.fit(X, y)
dtc2 = DecisionTreeClassifier(ccp_alpha = 0.03)dtc2.fit(X, y)
dtc3 = DecisionTreeClassifier(ccp_alpha = 0.06)dtc3.fit(X, y)
dtc4 = DecisionTreeClassifier(ccp_alpha = 0.1)dtc4.fit(X, y)
d1 = dtc1.get_depth()d2 = dtc2.get_depth()d3 = dtc3.get_depth()d4 = dtc4.get_depth()What can we say about d1, d2, d3 and d4?
d1 < d2 < d3 < d4
d1 ≤ d2 ≤ d3 ≤ d4
d1 > d2 > d3 > d4
d1 ≥ d2 ≥ d3 ≥ d4
Correct answer
d1 ≥ d2 ≥ d3 ≥ d4
Suppose that we have 10000 samples in a dataset. Suppose further we use the K − means algorithm to find clusters in the dataset. Then, the statement that K-Means algorithm always converges with zero inertia (or zero Sums Square Error) for some value of K is
Always True
Always False
True, sometimes
Correct answer
Always True
Suppose that we use k-means clustering for a dataset having 100 samples. The initial centroids for k clusters can be initialized in multiple ways. One such as way is shown below
20 centroids are randomly initialized 10 times
10 centroids are randomly initialized 20 times
20 samples in the dataset are selected as initialization point such that they are at least 10 units away from each other
10 samples in the dataset are selected as initialization point such that they are at least 20 units away from each other
Correct answer
20 centroids are randomly initialized 10 times
We wish to load the wine dataset from sklearn. Which of the following will throw an error?
Correct answers
Which of the following is correct?
Correct answers
Which of the following algorithms may get impacted by feature scaling?
LinearRegression
DecisionTree
SVM
BinomialNaiveBayes
Correct answers
LinearRegression
SVM
Which of the following class(es) is (are) used to instantiate a neural network in Sklearn.
Correct answers
Following information about X_train is given:
X_train is (100,6)from sklearn.preprocessing import OneHotEncoderOhe = OneHotEncoder()Encoded_X_train = ohe.fit_transform(X_train)Encoded_X_train.shapeWhich of the following is(are) correct option(s) for above information ?
Correct answers
Which of the following ways can help in feature selection?
Drop a Feature with many missing values
Drop a feature containing data with high standard deviation
Use SelectKBest or SelectKPercentile methods
Drop a feature which has high correlation with target variable
Correct answers
Drop a Feature with many missing values
Use SelectKBest or SelectKPercentile methods
Consider the following block of code:
from sklearn.datasets import load_breast_cancerfrom sklearn.tree import DecisionTreeClassifierfrom sklearn.model_selection import train_test_splitX,y = load_breast_cancer(as_frame = True, return_X_y = True)X_train,X_test,y_train,y_test = train_test_split(X, y, test_size = 0.2, random_state = 1)clf = DecisionTreeClassifier(min_samples_split = 6, min_samples_leaf = 4, random_state = 5)clf.fit(X_train, y_train)print(clf.score(X_test, y_test))In which of the following scenarios, the split will NOT be done at node N?
Number of samples at node N = 15. If it is split, it will result in 9 nodes in the left child and 6 nodes in the right child.
Number of samples at node N = 5. If it is split, it will result in 4 nodes in the left child and 2 nodes in the right child.
Number of samples at node N = 7. If it is split, it will result in 4 nodes in the left child and 3 nodes in the right child.
Number of samples at node N = 12. If it is split, it will result in 3 nodes in the left child and 9 nodes in the right child.
Correct answers
Number of samples at node N = 5. If it is split, it will result in 4 nodes in the left child and 2 nodes in the right child.
Number of samples at node N = 7. If it is split, it will result in 4 nodes in the left child and 3 nodes in the right child.
Number of samples at node N = 12. If it is split, it will result in 3 nodes in the left child and 9 nodes in the right child.
The code given below attempts to find the clusters in the dataset, X using the K-means algorithm.
The code attempts to find 2 clusters in the given dataset
The dataset X can be visualized in the Euclidean space
The code raises an error upon execution
The code gets executed without an error upon execution
Correct answers
The dataset X can be visualized in the Euclidean space
The code raises an error upon execution
The following line of code creates a neural network (assume necessary imports)
Select the correct statements from the following list of statements
The neural network contains 3 hidden layers with 5 neurons in each hidden layer
The neural network contains 5 hidden layers with 3 neurons in each hidden layer
The neural network contains 2 hidden layers with 5 neurons in the second hidden layer
The neural network contains 2 hidden layers with 3 neurons in the first hidden layer
None of the given options are correct
Correct answers
The neural network contains 2 hidden layers with 5 neurons in the second hidden layer
The neural network contains 2 hidden layers with 3 neurons in the first hidden layer
Consider the following code block:
from sklearn.model_selection import GridSearchCVfrom sklearn.tree import DecisionTreeClassifierfrom sklearn.datasets import make_classification
X, y = make_classification(n_samples = 100, n_features = 3,n_informative = 2, n_redundant = 1)
param_grid = [{'max_depth': [2, 3, 4, 5, 6], 'min_samples_split': [2, 3, 4, 5, 6]}, {'min_samples_leaf': [2, 3, 4, 5, 6]}, {'min_impurity_decrease': [0.2, 0.3, 0.4, 0.5, 0.6], 'ccp_alpha': [0.1, 0.2, 0.3, 0.4, 0.5, 0.6]}]
gscv = GridSearchCV(DecisionTreeClassifier(), param_grid, cv = 3)gscv.fit(X, y)
print(gscv.best_params_)How many parameter combinations will be tried by GridSearchCV?
Correct answer: 60