Home / Past papers / MLP End Term / this paper

IIT M DIPLOMA ET1 EXAM QPD1 S2 30 Apr 2023 - 30 Apr 2023

MLP End Term36 questions100 marks26 MCQ · 9 MSQ · 1 SADiploma / QPD1

Elapsed 00:00

Q1MCQ2 marks

______________ 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()

Q2MCQ2 marks

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.

Q3MCQ2 marks

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 meanmodel.
R2 score can be negative. That happens if the mean model is worse than ourmodel.

Q4MCQ2 marks

Which options represent the output of the following code block? question figure

Which options represent the output of the following code block?

Q5MCQ2 marks

Consider the following code block: question figure

Consider the following code block: What will be the value of the ‘count’?

Q6MCQ2 marks

Consider the following code block: question figure

Consider the following code block:

Q7MCQ2 marks

Which of the following is a hyperparameter?

L1-ratio in elasticnet
Pruning parameter in a decision tree
Learning rate in SGDRegressor
All of these

Q8MCQ2 marks

DecisionTreeClassifier
RandomForestRegressor
LogisticRegressor
SGDRegressor

Q9MCQ2 marks

Consider the following code question figure

Consider the following code

Q10MCQ2 marks

Consider the following code: question figure

Consider the following code:

0.66
0.72
0.80
1.00

Q11MCQ2 marks

Which of the following are correct about sklearn.svm.SVC: question figure

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 decisionboundary with each data point.
It will print indices of the support vectors from the training set.

Q12MCQ2 marks

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

Q13MCQ3 marks

Which of the following APIs only supports conjoining transformers and estimators in series (i.e. one after another)?

Pipeline
ColumnTransformer
FeatureUnion
All of these

Q14MCQ3 marks

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.

Q15MCQ3 marks

(10, 3)
(10, 7)
(10, 8)
(10, 10)
(10, 11)

Q16MCQ3 marks

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

Q17MCQ3 marks

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.

Q18MCQ3 marks

Consider following code snippets, assuming necessary imports. question figure

Consider following code snippets, assuming necessary imports.

Q19MCQ3 marks

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 inorder to be considered in the same class
For low values of gamma, the points need to be very close to each other inorder to be considered in the same class
Gamma doesn’t affect the SVM model at all
None of these

Q20MCQ3 marks

Consider the following code: question figure

Consider the following code:

dtc1
dtc2
dtc3
dtc4

Q21MCQ3 marks

Following is the code to tune the n_estimators parameter of a Bagging Classifier model. question figure

Following is the code to tune the n_estimators parameter of a Bagging Classifier model.

Q22MCQ4 marks

Consider the following code. How many DecisionTreeClassifier models will be trained internally? question figure

Consider the following code. How many DecisionTreeClassifier models will be trained internally?

Q23MCQ4 marks

Consider two classifiers as shown in the following block of code: question figure

Consider two classifiers as shown in the following block of code:

depth(clf1) ≥ depth(clf2)
depth(clf1) ≤ depth(clf2)
depth(clf1) = depth(clf2)
Insufficient Information

Q24MCQ4 marks

Consider the following code: question figure

Consider the following code:

d1 < d2 < d3 < d4
d1 ≤ d2 ≤ d3 ≤ d4
d1 > d2 > d3 > d4
d1 ≥ d2 ≥ d3 ≥ d4

Q25MCQ4 marks

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

Q26MCQ4 marks

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 question figure

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 areat least 10 units away from each other
10 samples in the dataset are selected as initialization point such that they areat least 20 units away from each other

Q27MSQ2 marksselect all that apply

We wish to load the wine dataset from sklearn. Which of the following will throw an error?

Q28MSQ2 marksselect all that apply

Which of the following is correct?

Q29MSQ2 marksselect all that apply

Which of the following algorithms may get impacted by feature scaling?

LinearRegression
DecisionTree
SVM
BinomialNaiveBayes

Q30MSQ2 marksselect all that apply

Which of the following class(es) is (are) used to instantiate a neural network in Sklearn.

Q31MSQ3 marksselect all that apply

Q32MSQ3 marksselect all that apply

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

Q33MSQ4 marksselect all that apply

Consider the following block of code: question figure

Consider the following block of code:

Number of samples at node N = 15. If it is split, it will result in 9 nodes in theleft 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 leftchild 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 leftchild 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 theleft child and 9 nodes in the right child.

Q34MSQ4 marksselect all that apply

The code given below attempts to find the clusters in the dataset, X using the K-means algorithm. question figure

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

Q35MSQ4 marksselect all that apply

The following line of code creates a neural network (assume necessary imports) question figure

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 hiddenlayer
The neural network contains 5 hidden layers with 3 neurons in each hiddenlayer
The neural network contains 2 hidden layers with 5 neurons in the secondhidden layer
The neural network contains 2 hidden layers with 3 neurons in the first hiddenlayer
None of the given options are correct

Q36SA3 marks

Consider the following code block: question figure

Consider the following code block:

Your numeric answer:

Answer key

Q1 (MCQ, 2 marks): D
Q2 (MCQ, 2 marks): B
Q3 (MCQ, 2 marks): C
Q4 (MCQ, 2 marks): C
Q5 (MCQ, 2 marks): C
Q6 (MCQ, 2 marks): C
Q7 (MCQ, 2 marks): D
Q8 (MCQ, 2 marks): D
Q9 (MCQ, 2 marks): D
Q10 (MCQ, 2 marks): B
Q11 (MCQ, 2 marks): B
Q12 (MCQ, 2 marks): B
Q13 (MCQ, 3 marks): A
Q14 (MCQ, 3 marks): A
Q15 (MCQ, 3 marks): B
Q16 (MCQ, 3 marks): A
Q17 (MCQ, 3 marks): A
Q18 (MCQ, 3 marks): A
Q19 (MCQ, 3 marks): A
Q20 (MCQ, 3 marks): A
Q21 (MCQ, 3 marks): C
Q22 (MCQ, 4 marks): F
Q23 (MCQ, 4 marks): A
Q24 (MCQ, 4 marks): D
Q25 (MCQ, 4 marks): A
Q26 (MCQ, 4 marks): A
Q27 (MSQ, 2 marks): A, B
Q28 (MSQ, 2 marks): B, D
Q29 (MSQ, 2 marks): A, C
Q30 (MSQ, 2 marks): B, D
Q31 (MSQ, 3 marks): A, C
Q32 (MSQ, 3 marks): A, C
Q33 (MSQ, 4 marks): B, C, D
Q34 (MSQ, 4 marks): B, C
Q35 (MSQ, 4 marks): C, D
Q36 (SA, 3 marks): 60

Concepts this paper tests

Auto-generated summary from the source site. Handy as a checklist; the lessons above explain each idea from scratch.

Here’s a structured summary of core topics, concepts, principles, and equations essential for the "Diploma Level: Machine Learning Practice" exam, along with Mermaid knowledge graphs to visualize relationships.

1. Core Topics & Concepts

A. Data Preprocessing

Why Preprocess?
- Handle missing values, scale features, encode categorical variables, split data, and handle imbalanced datasets.
- Key Methods:
  - train_test_split (e.g., test_size=0.3 for 70-30 split).
  - StandardScaler/MinMaxScaler for feature scaling.
  - OneHotEncoder for categorical features (e.g., 3 categories → 3 columns).
  - KNNImputer for missing values (e.g., n_neighbors=3).
  - Pipeline to chain transformations and estimators.

Cross-Validation:
- LeavePOut(p=2): Generates C(n_samples, p) splits (e.g., for n_samples=10, p=2 → 45 splits).
- RepeatedKFold(n_splits=3, n_repeats=2): Repeats K-Fold n_repeats times.
- GridSearchCV: Exhaustive search over hyperparameters (e.g., cv=3 for 3-fold CV).

B. Supervised Learning

Classification:
- Metrics:
  - Confusion Matrix: TP, TN, FP, FN (e.g., confusion_matrix(y_true, y_pred)).
  - F1-Score: Harmonic mean of precision/recall. Formula:

F1 = 2 \times \frac{\text{precision} \times \text{recall}}{\text{precision} + \text{recall}}

R² Score: For regression, but can be negative if the model is worse than the mean model.

R^2 = 1 - \frac{\sum (y_i - \hat{y}_i)^2}{\sum (y_i - \bar{y})^2}

Algorithms:
- Logistic Regression:
  - multi_class='multinomial' + solver='lbfgs' for softmax regression.
  - partial_fit for online learning (incremental updates).
- Decision Trees:
  - Hyperparameters:
    - min_samples_split (default=2), min_samples_leaf (default=1).
    - ccp_alpha: Higher values → more pruning → shallower trees.
  - Depth Comparison: Lower min_samples_split/min_samples_leaf → deeper trees.
- SVM (SVC):
  - support_: Indices of support vectors.
  - gamma: High gamma → tighter decision boundary (points must be very close to be in the same class).
- k-NN:
  - Smaller n_neighbors → more complex (less smooth) decision boundary.
- Neural Networks:
  - MLPClassifier/MLPRegressor in sklearn (e.g., hidden_layer_sizes=(3,5) → 2 hidden layers with 3 and 5 neurons).
- SGDClassifier:
  - Supports loss="perceptron", loss="log" (logistic), loss="hinge" (SVM).
  - partial_fit for online learning.

Regression:
- Linear Regression: LinearRegression(fit_intercept=True).
- Polynomial Features: PolynomialFeatures(degree=2, interaction_only=True) generates interaction terms (e.g., for 3 features → 6 terms: original + 3 interactions).

C. Unsupervised Learning

K-Means Clustering:
- init='random': Randomly initialize centroids.
- n_init=10: Run algorithm 10 times with different centroids.
- Convergence: Always converges (inertia ≥ 0), but may not reach zero inertia unless K = n_samples.
- Error Cases: If n_clusters > n_samples, it raises an error.

D. Model Evaluation & Hyperparameter Tuning

Hyperparameters vs. Parameters:
- Hyperparameters (set before training): C (SVM), n_estimators (Random Forest), learning_rate (SGD).
- Parameters (learned during training): weights in linear regression.
GridSearchCV:
- Searches over all combinations of hyperparameters (e.g., param_grid with max_depth and min_samples_split).
- Example: For param_grid = [{'A': [1,2]}, {'B': [3,4]}], total combinations = 2 (A) + 2 (B) = 4 (not 2×2, since separate dicts).
- Naming in Pipelines: Use stepname__param (e.g., bc__n_estimators for BaggingClassifier in a pipeline).

E. Special Cases

Multi-Label Classification: Each sample can have multiple labels (e.g., image with dogs, birds, trees).
Imbalanced Data: Use metrics like F1-score (not accuracy).
Feature Selection:
- Drop features with:
  - Many missing values.
  - Low variance or high correlation with other features.
- Use SelectKBest or SelectPercentile.

2. Key Equations

Concept	Equation
R² Score	$R^2 = 1 - \frac{\sum (y_i - \hat{y}_i)^2}{\sum (y_i - \bar{y})^2}$
F1-Score	$F1 = 2 \times \frac{\text{precision} \times \text{recall}}{\text{precision} + \text{recall}}$
Polynomial Features	For `degree=2`, `interaction_only=True`: $n_{\text{features}} = \binom{n}{2} + n$ (e.g., 3 features → 6 terms).
K-Means Inertia	$\text{Inertia} = \sum_{i=1}^{n} \min_{\mu_j \in C} \\|x_i - \mu_j\\|^2$
SGD Update (Perceptron)	$w = w + \eta \cdot (y_i - \hat{y}_i) \cdot x_i$

Graph 2: Unsupervised Learning (K-Means)

Graph 3: Model Evaluation & Hyperparameter Tuning

Graph 4: Decision Trees & Pruning

Graph 5: Data Preprocessing Pipeline

4. Common Pitfalls & Tips

Confusion Matrix:

Rows = True labels, Columns = Predicted labels.

Example:

y_true = ["Rainy", "Sunny", "Sunny", "Rainy", "Sunny", "Rainy"]
y_pred = ["Sunny", "Sunny", "Rainy", "Rainy", "Sunny", "Sunny"]

Output: [[1, 2], [2, 1]] (Option C).

Polynomial Features:
- interaction_only=True → Only interaction terms (no squares).
- For n_features=3, degree=2 → 6 features (original 3 + 3 interactions).

SGDClassifier:
- loss="perceptron" → Perceptron.
- loss="log" → Logistic regression.
- loss="hinge" → Linear SVM.

K-Means:
- If n_clusters > n_samples, it raises an error.
- init='random' initializes centroids randomly n_init times.

Hyperparameter Tuning:
- In GridSearchCV, separate dictionaries in param_grid are treated as separate grids (not combined).
- Example: param_grid = [{'A': [1,2]}, {'B': [3,4]}] → 2 + 2 = 4 combinations (not 4 total).

Pipelines:
- Use stepname__param for hyperparameter tuning (e.g., bc__n_estimators).

R² Score:
- Can be negative if the model is worse than the horizontal line (mean predictor).

Neural Networks in sklearn:
- MLPClassifier/MLPRegressor (not NNClassifier).
- hidden_layer_sizes=(3,5) → 2 hidden layers (3 neurons, then 5 neurons).