Day 2 Practice Problems

You are not logged in.
Please Log In for full access to the web site.
Note that this link will take you to an external site (https://petrock.mit.edu) to authenticate, and then you will be redirected back to this page.

1) String Matching

Type a string that matches the following regex pattern ^Ba+.*n$

Type a string that matches the following regex pattern ^Hello{5,} (W|w)orld

When using the stream editor tool 'sed', what is the purpose of the syntax  within a regular expression?

To capture a portion of the match for later refernce in the replacement string

To specify that the enclosed pattern is optional

To define a character class of literal parentheses

To indicated the beginning and end of the entired line

Which component of a standard 'sed' command is responsible for ensuring that all instances of a pattern on a single line are replaced, rather than just the first one?

The use of backslashes  to group patterns

The s prefix at the start of the expression

The g flag appended to the end of the substitution command

The pipe character | used to combine Unix commands

Which 'awk' feature is used to define actions that should occur before any lines of the input file are processed?

The -F command line flag

The END block

The BEGIN block

The $0 variable assignment

2) Similarity

What is the primary goal of applying a 'Stemming' algorithm like Porter's Algorithm in text processing?

To convert every word in a document into a high-dimensional vector

To identify and remove common words like 'the' and 'and' from the dataset

To reduce words to their base or root form by removing suffixes

To determine the sentiment of a sentence by analyzing its structure

What is the Jaccard Similarity between S1 = {A, B, C, D} and S2 = {C,D,E}

What is the Jaccard Similarity between S1 = {A, B, C, D} and S2 = {C, H, A, S, E}

What is the Cosine similarity of the phrases:
"The best team"
and
"The best team"

According to the TF-IDF model, what happens to the importance of a term if it appears in almost every document in a large corpus?

Its importance remains constant regardless of document frequency

Its importance increases linearly with the total number of documents N

Its importance decreases because it has a high Inter-Document Frequency (IDF)

Its importance increases because it is highly representative of the corpus

3) Embeddings

What properties apply to Skip-gram? (Select all that apply)

uses a word to predict its surrounding context

Better for rare words than CBOW

predicts a target word from its context

needs less data than CBOW

What is the primary function of 'Negative Sampling' during the training of word embedding models?

To ensure that antonyms are placed on opposite sides of the vector space

To reduce the dimensionality of the output vectors through principal component analysis

To remove sentences with negative sentiment from the training corpus

To update only a small percentage of weights by selecting a few 'negative' words for each sample

4) Applications, RAG, and Vector Databases

In the context of Large Language Models (LLMs), what problem does Retrieval Augmented Generation (RAG) primarily help to solve?

The limited max context window length and the high cost of frequent retraining

The inability of LLMs to generate text in languages other than English

The requirement for high-quality human-written poetry in the training set

Slow inference speed cause by the Transformer architecture's attention mechanism

What is a major advantage of Retrieval Augmented Generation (RAG) when using Large Language Models (LLMs)?

It increases the model training time by years of GPU time

It completely eliminats the need for tokenization of the input query

It allows the model to compress its own parameters into a smaller vector database

It enables the LLM to access up-to-date or private information without requiring constant rtraining

How does Locality Sensitive Hashing (LSH) differ from traditional cryptographic hashing in vector databases?

Traditional hashing is O(1), while LSH is always O(n)

LHS is designed to maximize collisions for similar inputs, whereas traditional hashing tries to minimize them

LSH can only be applied to integer data, while traditional hashing works on strings

LSH uses much longer hash keys to ensure data security

In Navigable Small World (NSW) graphs used for vector indexing, what is the role of 'long-range' links?

They act as a backup in case the primary vector database fails

They allow for logarithmic scaling of search time by facilitating 'jumps' across the graph

They ensure that the graph is fully connected and has no isolated clusters

They provide the basic connectivity needed for greedy search to work at a local level