Home / Past papers / MLP End Term / this paper

IIT M FOUNDATION DIPLOMA AN EXAM - 01 Sep 2024

MLP End Term37 questions100 marks19 MCQ · 13 MSQ · 5 SAFoundation / Diploma / afternoon

Elapsed 00:00

Comprehension passage - the questions below it refer to this

Consider the following common data and answer the subquestion: question figure

Consider the following common data and answer the subquestion:

Q1MSQ2 marksselect all that apply

Consider the following code snippet: question figure

Consider the following code snippet: Which of the following will be equivalent to the above code snippet?

Q2SA2 marks

What is the output of the following code snippet: question figure

What is the output of the following code snippet: Enter -1, if you think the above statement will generate an error.

Your numeric answer:

Q3SA2 marks

What is the output of the following code snippet: question figure

What is the output of the following code snippet: Enter -1, if you think the above statement will generate an error.

Your numeric answer:

Q4SA2 marks

What is the output of the following code snippet: question figure

What is the output of the following code snippet: Enter -1, if you think the above statement will generate an error.

Your numeric answer:

Q5MCQ2 marks

What is the output of the following code snippet: question figure

What is the output of the following code snippet: Choose correct options from following:

The number of rows will decrease in the dataset.
The number of columns will decrease in the dataset.
The number of columns and number of rows, both, will decrease in thedataset.
There will be no change.
Insufficient information.

Q6MCQ2 marks

To load datasets from openml.org, which method will be appropriate?

load_openml()
read_openl()
read_data()
fetch_openml()
load_data()
load_csv()
fetch_csv()

Q7MCQ2 marks

In which of the option given below data preprocessing is required?

In some columns which has values between 0 and 1.
A column contains entity names such as ‘Chennai’, ‘CHENNAI’ and ‘MADRAS’.
The data has only numbers in all the columns.

Q8MCQ2 marks

Q9MCQ2 marks

Which of the following scikit-learn classes is best suited for examining how the number of samples impacts training and testing errors?

Q10MCQ2 marks

Consider following two statements: Statement 1: The GaussianNB classifier can incrementally learn using partial_fit. Statement 2: GaussianNB performance suffers when features are dependent, as it assumes independence in calculating conditional probabilities.

Both statements are True
Only statement 1 is True
Only statement 2 is True
Both statements are False

Q11MCQ2 marks

Q12MCQ2 marks

Which of the following is true for a hard margin SVM algorithm ?

It does not create hyperplanes as a decision boundary
It will correctly classify all the data point if the data is linearly separable
It is robust to outliers
It is mostly used for clustering the data

Q13MCQ2 marks

Q14MCQ2 marks

Decision Trees are prone to:

Low bias, low variance
High bias, low variance
Low bias, high variance
High bias, high variance

Q15MCQ3 marks

Q16MCQ3 marks

‘l1’, ‘l2’
‘l1’, None
‘l2’, ‘l1’
‘l2’, None

Q17MCQ3 marks

True
False
Yes
No
None of these

Q18MCQ3 marks

It controls the number of weak learners.
It shrinks the contribution of each classifier.
It sets the weight of each weak learner.
It adjusts the speed at which the model learns.

Q19MCQ4 marks

Output score is likely to increase.
Output score is likely to decrease.
Output score may increase or decrease.
Output score will remain the same.

Q20MCQ4 marks

Q21MCQ4 marks

The final predictions are based on the majority vote.
The final predictions are based on the average of probabilities predicted byeach classifier.
The final predictions are based on the weighted sum of the predictions.
The final predictions are based on the classifier with the highest accuracy.

Q22MCQ4 marks

The mean accuracy across all cross-validation folds.
The mean precision across all cross-validation folds.
The mean recall across all cross-validation folds.
The mean F1 score across all cross-validation folds.

Q23MCQ4 marks

Plotting the sum of the squared distances of samples to their closest clustercenter
by Changing the distance metric for KMeans
By initialising clustering centroids very far from each other
All of these

Q24MSQ2 marksselect all that apply

Q25MSQ2 marksselect all that apply

Q26MSQ2 marksselect all that apply

Which of the following is a hyper parameter?

Q27MSQ2 marksselect all that apply

Which of the following models are inherently multiclass models?

Perceptron
DecisionTreeClassifier
KNeighborClassifier
LogisticRegression

Q28MSQ2 marksselect all that apply

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

SGDClassifier()
MLPClassifier()
NNClassifier()
MLPRegressor()

Q29MSQ3 marksselect all that apply

Which of the following is correct?

Q30MSQ3 marksselect all that apply

Which of the following processes can be done if we have a dataset with imbalanced target class distribution?

Remove all the minority classes
Up-sample the minority classes
Remove all the majority classes
Down-sample the majority classes
create synthetic samples to balance the classes

Q31MSQ3 marksselect all that apply

Which of the following is true about Naive Bayes algorithm ?

It is primarily used for regression problems
It is primarily used for classification problems
Hyperparameter tuning is required
Hyperparameter tuning is not required

Q32MSQ3 marksselect all that apply

Which of the following is/are correct regarding RadiusNeighborsClassifier

Only 5 neighbours in the range of some radius are used to compute the labelof a sample.
All the neighbours in the range of some radius are used to compute the labelof a sample.
It is sensitive to outliers.
It is not sensitive to outliers.

Q33MSQ3 marksselect all that apply

Data is continuously being generated
Data is generated every month
Whole data is generated and its in a huge file size
For very small dataset

Q34MSQ4 marksselect all that apply

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 3 neurons in the secondhidden layer
The neural network contains 2 hidden layers with 5 neurons in the first hiddenlayer
All of the given options are correct

Q35MSQ5 marksselect all that apply

Number of samples at node N = 15. If it is split, it will result in 9 samples in theleft child node and 6 sample in the right child node.
Number of samples at node N = 5. If it is split, it will result in 4 samples in theleft child node and 2 samples in the right child node.
Number of samples at node N = 12. If it is split, it will result in 3 samples in theleft child node and 9 samples in the right child node.
Number of samples at node N = 7. If it is split, it will result in 4 samples in theleft child node and 3 samples in the right child node.

Q36SA2 marks

Your numeric answer:

Q37SA4 marks

Your numeric answer:

Answer key

Q1 (MSQ, 2 marks): A, C
Q2 (SA, 2 marks): 120
Q3 (SA, 2 marks): 3
Q4 (SA, 2 marks): -1
Q5 (MCQ, 2 marks): A
Q6 (MCQ, 2 marks): D
Q7 (MCQ, 2 marks): B
Q8 (MCQ, 2 marks): B
Q9 (MCQ, 2 marks): D
Q10 (MCQ, 2 marks): A
Q11 (MCQ, 2 marks): C
Q12 (MCQ, 2 marks): B
Q13 (MCQ, 2 marks): B
Q14 (MCQ, 2 marks): C
Q15 (MCQ, 3 marks): E
Q16 (MCQ, 3 marks): A
Q17 (MCQ, 3 marks): A
Q18 (MCQ, 3 marks): B
Q19 (MCQ, 4 marks): A
Q20 (MCQ, 4 marks): D
Q21 (MCQ, 4 marks): B
Q22 (MCQ, 4 marks): A
Q23 (MCQ, 4 marks): A
Q24 (MSQ, 2 marks): A, D
Q25 (MSQ, 2 marks): A, D
Q26 (MSQ, 2 marks): B, C, D
Q27 (MSQ, 2 marks): B, C
Q28 (MSQ, 2 marks): B, D
Q29 (MSQ, 3 marks): A, D
Q30 (MSQ, 3 marks): B, D, E
Q31 (MSQ, 3 marks): B, D
Q32 (MSQ, 3 marks): B, D
Q33 (MSQ, 3 marks): A, B, C
Q34 (MSQ, 4 marks): C, D
Q35 (MSQ, 5 marks): A, D
Q36 (SA, 2 marks): 0.319 to 0.328
Q37 (SA, 4 marks): 180

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 the core topics, concepts, principles, and equations essential for the Diploma-Level Machine Learning Practice (Computer-Based Exam). The summary is organized by topic clusters for clarity, followed by Mermaid knowledge graphs to visualize relationships.

1. Core Topics & Concepts

A. Data Handling with Pandas

DataFrame Operations:
- Indexing: df.loc[] (label-based), df.iloc[] (position-based).
- Slicing: df.loc[23:24, 'Maths':'History'] ≡ df.iloc[2:4, 1:3] (Q2).
- Handling missing data: df.dropna() reduces rows (Q6).
- Data types: df.Hindi.apply(type).nunique() counts unique types (e.g., int, float, str) (Q4).

Chunked Data Loading:
- For large files: pd.read_csv("filename.csv", chunksize=1024, low_memory=False) (Q12).

B. Model Training & Evaluation

Train-Test Split:
- train_test_split(X, y, test_size=30) for exact sample counts (Q25).
- Default splits: test_size=0.2 (20% test data).

Cross-Validation:
- K-Fold CV: KFold(n_splits=3) splits data into 3 folds (Q9).
  - Output: Indices of train/test sets (e.g., [3,4,5,6,7,8] vs [0,1,2]).
- ShuffleSplit: Randomized splits (Q26).
- Learning Curves: sklearn.model_selection.learning_curve plots training/test errors vs. sample size (Q10).

Metrics:
- Regression: neg_mean_squared_error, r2 (Q26).
- Classification: accuracy, precision, recall, f1 (Q22).
- Clustering: Inertia (sum of squared distances to centroids) (Q24, Q38).

C. Algorithms & Hyperparameters

Decision Trees:
- Hyperparameters:
  - max_depth: Controls tree depth (higher → overfitting) (Q20).
  - min_samples_split: Minimum samples to split a node (Q36).
  - min_samples_leaf: Minimum samples in a leaf (Q36).
- Prone to: High variance (overfitting) (Q15).

SVM (Support Vector Machines):
- Hard Margin SVM: Correctly classifies all points if data is linearly separable (Q13).
- Kernel Trick: Not covered in questions, but implied for non-linear data.

Naive Bayes:
- Assumes feature independence (Q11).
- No hyperparameter tuning required (Q32).
- Incremental learning: partial_fit (Q11).

k-Nearest Neighbors (k-NN):
- RadiusNeighborsClassifier: Uses all neighbors within a radius (not fixed k) (Q33).
- Sensitive to scale (normalization recommended).

Ensemble Methods:
- Bagging: BaggingClassifier with DecisionTreeClassifier (Q23).
- Boosting: AdaBoost’s learning_rate shrinks weak learner contributions (Q19).
- Voting Classifier: voting='soft' averages probabilities (Q22).

Neural Networks (MLP):
- MLPClassifier/MLPRegressor: Defines hidden layers via hidden_layer_sizes=(5,3) (2 layers with 5 and 3 neurons) (Q35).
- Hyperparameters: max_iter, learning_rate, alpha (L2 regularization).

Stochastic Gradient Descent (SGD):
- Loss Functions:
  - loss='log_loss' → Logistic regression (Q18).
  - loss='hinge' → Linear SVM (Q17).
  - loss='perceptron' → Perceptron (Q30).
- Penalties: penalty='l1' (sparse weights) vs. penalty='l2' (Q17).
- Warm Start: warm_start=True enables incremental training (Q18).

Clustering (K-Means):
- Inertia: Sum of squared distances to centroids (Q24).
  - Inertia = 0 when n_clusters = n_samples (Q38).
- Elbow Method: Plot inertia vs. n_clusters to find optimal k.

Imbalanced Data:
- Techniques: Upsample minority, downsample majority, or synthetic samples (SMOTE) (Q31).

D. Preprocessing & Feature Engineering

Text Data:
- CountVectorizer: Converts text to word-count matrices (Q14).
  - Output: Vocabulary dictionary (e.g., {'this': 4, 'is': 2}).

Polynomial Features:
- PolynomialFeatures(degree=2) adds interaction terms (Q16).
- Hyperparameter tuning: param_grid = {'poly__degree': [2, 3, ...]}.

Scaling:
- MinMaxScaler or StandardScaler for SGD/neural networks (Q18).

Handling Categorical Data:
- Inconsistent Entries: E.g., 'Chennai', 'CHENNAI', 'MADRAS' → Requires preprocessing (Q8).

E. Model Selection & Hyperparameter Tuning

Grid Search:
- GridSearchCV: Exhaustively tests hyperparameter combinations (Q21).
  - Example: param_grid = {'max_depth': [1,3,5], 'min_samples_split': [1,4,7]} → 15 combinations.

Partial Fit:
- For streaming/large data: model.partial_fit(X, y) (Q12, Q34).

F. Specialized Topics

OpenML:
- Load datasets: fetch_openml() (Q7).

Model-Specific Notes:
- Perceptron/SGD: SGDClassifier(loss='perceptron') is a stochastic perceptron (Q30).
- Logistic Regression: SGDClassifier(loss='log_loss') (Q30).
- GaussianNB: Assumes normal distribution for features (Q11).

2. Key Equations (Mathematics)

A. Regression Metrics

Mean Absolute Error (MAE):

\text{MAE} = \frac{1}{n} \sum_{i=1}^n |y_i - \hat{y}_i|

Example (Q37): MAE for y_true = [0.5, 0.2, 0.7, 1], y_pred = [0.4, 0.2, 1, 0.1] → 0.325.

Mean Squared Error (MSE):

\text{MSE} = \frac{1}{n} \sum_{i=1}^n (y_i - \hat{y}_i)^2

B. Classification Metrics

Log Loss (Cross-Entropy):

\text{Log Loss} = -\frac{1}{n} \sum_{i=1}^n [y_i \log(\hat{y}_i) + (1 - y_i) \log(1 - \hat{y}_i)]

AdaBoost Weight Update:

\alpha_t = \frac{1}{2} \ln\left(\frac{1 - \epsilon_t}{\epsilon_t}\right)

$\epsilon_t$ : Error rate of weak learner at iteration $t$ (Q19).

C. Clustering

Inertia (K-Means):

\text{Inertia} = \sum_{i=1}^n \min_{\mu_j \in C} ||x_i - \mu_j||^2

$C$ : Set of centroids, $\mu_j$ : Centroid of cluster $j$ (Q24, Q38).

D. Probability (Naive Bayes)

Conditional Probability (GaussianNB):

P(y | X) \propto P(y) \prod_{i=1}^m P(x_i | y)

Assumes feature independence (Q11).

Graph 2: Model Training & Evaluation

Graph 3: Algorithms & Hyperparameters

Graph 4: Specialized Workflows

Graph 5: Clustering & Dimensionality

4. Exam Tips

Pandas:
- Master loc vs. iloc (Q2).
- Know how dropna() affects rows/columns (Q6).

Model Evaluation:
- For regression, use neg_mean_squared_error or r2 (Q26).
- For classification, accuracy or f1 (Q22).

Algorithms:
- Decision Trees: High variance (Q15), tune max_depth (Q20).
- SVM: Hard margin for separable data (Q13).
- Naive Bayes: Independent features (Q11).

Hyperparameter Tuning:
- GridSearchCV tests all combinations (Q21).
- Use param_grid with double underscores for pipelines (e.g., poly__degree) (Q16).

SGD:
- loss='log_loss' → Logistic regression (Q18).
- penalty='l1' → Sparse weights (Q17).

Clustering:
- Inertia = 0 when n_clusters = n_samples (Q38).
- Elbow method for optimal k (Q24).

Text Data:
- CountVectorizer outputs a vocabulary dictionary (Q14).

Imbalanced Data:
- Upsample minority or downsample majority (Q31).

Neural Networks:
- hidden_layer_sizes=(5,3) → 2 layers (Q35).

Code Snippets:
- Always check for random_state in splits/models for reproducibility.
- warm_start=True enables incremental training (Q18).