Intermediate

Question 1

What is a good use case for implementing neural networks for supervised learning?

Answer 1

Neural networks are ideal for scenarios where traditional ML algorithms fail, especially with unstructured data like images, text, or audio. They are also preferred when the dataset size is large, allowing for improved pattern detection.

Question 2

Should neural networks be applied in healthcare for predicting diagnosis and medication?

Answer 2

No, neural networks are not suitable where interpretability is crucial, such as in healthcare. Decision trees or random forests are preferred for their transparency.

Question 3

Why do we mostly use the ReLU activation function in CNNs?

Answer 3

ReLU is simple to compute and avoids the vanishing gradient problem, making it effective for deep networks in CNNs.

Question 4

What is the role of fully connected layers in a CNN?

Answer 4

Fully connected layers learn non-linear combinations of high-level features extracted by convolutional layers, providing a meaningful feature space.

Question 5

What are some drawbacks of using CNNs on image datasets?

Answer 5

CNNs require a lot of labeled data and can be susceptible to spurious patterns. Transfer learning and data augmentation can address these issues.

Question 6

What text pre-processing steps are required for sentiment analysis of Twitter posts?

Answer 6

1) Lowercasing 2) Stemming 3) Lemmatization 4) Stop words removal 5) Noise removal 6) Remove emoticons.

Question 7

Which evaluation metric is suitable for sentiment analysis?

Answer 7

Precision, Recall, F-score, and Accuracy are suitable metrics. F1 scores are often used due to class imbalance in sentiment analysis.

Question 8

Which neural network configuration is likely to perform better on complex tasks?

Answer 8

The network with four hidden layers of four nodes each is likely to perform better due to its hierarchical learning capability.

Question 9

What are the pros and cons of Batch Gradient Descent vs Stochastic Gradient Descent?

Answer 9

Batch Gradient Descent has better convergence but is computationally expensive for large datasets. Stochastic Gradient Descent is faster but can be noisy and less stable.

Question 10

How is AUC different from ROC?

Answer 10

ROC is a curve showing a model’s performance at different thresholds, while AUC measures the area under the ROC curve.

Question 11

How do k-means and hierarchical clustering differ?

Answer 11

K-Means is centroid-based and requires a predefined number of clusters, while hierarchical clustering is connectivity-based and does not require this.

Question 12

Why are decision trees prone to overfitting?

Answer 12

Decision trees aim for homogeneity in leaf nodes, leading to capturing noise and complex patterns in training data, which results in overfitting.

Question 13

How does multicollinearity affect linear regression models?

Answer 13

Multicollinearity affects the interpretation of the model, making it difficult to distinguish the individual effects of correlated independent variables.

Question 14

Which evaluation metric should be used for linear regression with outliers?

Answer 14

MAE is preferred as it is robust to outliers, unlike MSE or RMSE which are sensitive to them.

Question 15

What is the difference between r-squared and adjusted r-squared?

Answer 15

R-squared measures the variance explained by independent variables, while adjusted r-squared accounts for the number of variables and penalizes unnecessary ones.

Question 16

How do we measure performance of a supervised learning model?

Answer 16

Performance metrics include R2, Adj. R2, RMSE, MAE for regression, and Accuracy, Precision, Recall, F1-Score for classification.

Question 17

How do we measure performance of an unsupervised learning model?

Answer 17

Performance metrics include silhouette score and cophenetic correlation.

Question 18

Can you plot 3D plots using matplotlib?

Answer 18

Yes, using the function: import numpy as np; import matplotlib.pyplot as plt; fig = plt.figure(); ax = plt.axes(projection =’3d’).

Question 19

How is get_dummies() different from one hot encoder?

Answer 19

OneHotEncoder cannot process strings directly and requires mapping to integers, while pandas.get_dummies converts string columns into one-hot representation by default.

Question 20

Name a tool to convert categorical columns into numeric columns.

Answer 20

LabelEncoder and OneHotEncoder from sklearn are popular tools for this purpose.

Question 21

How will you remove duplicate data from a dataframe?

Answer 21

Use the ‘drop_duplicates()’ function to eliminate redundant rows from the DataFrame.

Question 22

How do you select a sample from a dataframe?

Answer 22

Use df.sample() for random selection, df.sample(n=3) for n rows, df.sample(n=3, replace=True) for allowing duplicates, or df.sample(frac=0.50) for a fraction.

Question 23

How does the groupby function work in Python?

Answer 23

The groupby() function splits data into groups based on criteria, allowing for aggregation and analysis of grouped data.

Question 24

What are the parameters for the groupby function in Python?

Answer 24

Parameters include: by, axis, level, as_index, sort, group_keys, squeeze, and **kwargs.

Question 25

How do you check the distribution of data in Python?

Answer 25

A simple and commonly used plot to quickly check the distribution of a sample of data is the histogram. ## Footnote Example: from matplotlib import pyplot; pyplot.hist(data)

Question 26

Which libraries in SciPy have you worked with in your project?

Answer 26

SciPy contains modules for optimization, linear algebra, integration, interpolation, special functions, FFT, signal and image processing, ODE solvers, etc. Subpackages include: scipy.cluster, scipy.constants, scipy.fftpack, scipy.integrate, scipy.interpolation, scipy.linalg, scipy.io, scipy.ndimage, scipy.odr, scipy.optimize, scipy.signal, scipy.sparse, scipy.spatial, scipy.special, scipy.stats, scipy.weaves.

Question 27

How is the Python series different from a single column dataframe?

Answer 27

A Python series is the data structure for a single column of a DataFrame. It is a one-dimensional object that can hold any data type and does not have any name/header, whereas the DataFrame has column names.

Question 28

What does the function zip() do?

Answer 28

The zip() function takes iterables, aggregates them in a tuple, and returns it. ## Footnote Syntax: zip(*iterables)

Question 29

Can lambda function be used within a user-defined function?

Answer 29

Yes, a lambda function can be used as an anonymous function within another function.

Question 30

What does [::-1] do in Python?

Answer 30

[::] produces a copy of all the elements in order, while [::-1] produces a copy of all the elements in reverse order.

Question 31

How do you check missing values in a dataframe using Python?

Answer 31

The Pandas isnull() function detects missing values in the given object, returning a boolean same-sized object indicating if the values are NA.

Question 32

Explain a scenario where negative indices are used in Python.

Answer 32

Negative indexing allows access to elements from the end of an array. For example, -1 gives the last element, and -2 gives the second last element.

Question 33

What different methods can be used to standardize the data using Python?

Answer 33

Methods include Min Max Scaler, Standard Scaler, Max Abs Scaler, Robust Scaler, Quantile Transformer Scaler, Power Transformer Scaler, and Unit Vector Scaler.

Question 34

How would you define a block in Python?

Answer 34

A block is a group of statements in a program or script, consisting of at least one statement and declarations for the block.

Question 35

How do you do up-sampling of data? Name a Python function or explain the code.

Answer 35

Up-sampling is the process of randomly duplicating observations from the minority class. A common way is to resample with replacement. ## Footnote Module for resampling: from sklearn.utils import resample

Question 36

What is machine learning?

Answer 36

Machine learning is a branch of AI that focuses on using data and algorithms to mimic human learning, improving by learning from past events.

Question 37

What is the mathematical outcome of any machine learning model building exercise?

Answer 37

Machine learning models summarize patterns in data by establishing relationships between predictors and predicted values, often represented as equations.

Question 38

How do you address a performance gap in a machine learning model?

Answer 38

To address a performance gap, various regularization techniques can be applied to prevent overfitting, such as ridge regression, lasso regression, and pruning techniques.