Data Types and Machine Learning
Models in Data Science
Tabular data - CORRECT ANSWER โโโ Rows represent observations and columns
represent variables. This is considered traditional social science data. Example:
Employee records with variables like age, salary, and job title.
Text data - CORRECT ANSWER โโโ Rather than numbers, the data is text. It
needs to be converted into tabular data for processing. Techniques like tokenization
(e.g., Chat GPT-3 tokenizer, tiktoken, Byte pair encoding) are used for this purpose.
Example: Analyzing employee emails for sentiment or topic classification.
Image data - CORRECT ANSWER โโโ AI processes images as matrices of pixel
values. An image consists of a matrix of numbers between 0 and 255, indicating a
pixel's intensity. Example: Identifying defective products using machine vision on
factory lines.
Audio data - CORRECT ANSWER โโโ Audio data is often converted to arrays
based on waveform, quantifying the shape into numbers. Concepts related to audio
include pitch (measured in hertz), amplitude (loudness/quietness), and timbre
(uniqueness of sound). Example: Transcribing customer service calls for analysis.
Supervised machine learning - CORRECT ANSWER โโโ Involves training an
algorithm on a labeled dataset for predictions or decisions. Input data (only
features) is paired with output data (labels or targets). The algorithm learns
patterns between the input and output. Accuracy is a primary metric, measuring the
proportion of correctly classified instances. Example: Email spam detection where
emails are labeled as spam or not spam.
Unsupervised machine learning - CORRECT ANSWER โโโ The algorithm is trained
on an unlabeled dataset. It identifies patterns or relationships without guidance. The
algorithm might cluster similar data points based on shared features and can also
reduce data dimensionality. Evaluation is challenging due to the absence of a
'ground truth'. Example: Customer segmentation for targeted marketing campaigns.
Reinforcement learning - CORRECT ANSWER โโโ Has an agent interacting with an
environment to learn optimal decisions. The agent receives rewards or penalties
based on actions and updates its model to learn beneficial actions. The goal is to
maximize total reward. Total reward during training is often used for evaluation,
along with benchmarking against a baseline. Example: Training a robot to navigate
a warehouse while avoiding obstacles.
Generative AI - CORRECT ANSWER โโโ New examples of data that are useful.
Large Language Models (LLM) - CORRECT ANSWER โโโ Models that require large
amounts of data (billions) and parameters (billions).
Generative Pre-trained Transformer (GPT) - CORRECT ANSWER โโโ The algorithm
name for ChatGPT.
