ML End-Sem

Unsupervised learning
It is a type of machine learning where algorithms are used to uncover patterns or hidden
structures in unlabeled data. Unlike supervised learning, where the algorithm learns from labeled
data (input-output pairs), unsupervised learning deals with input data that doesn't have
corresponding output labels.

There are several approaches to unsupervised learning, each serving different purposes:
1. Clustering: Clustering algorithms aim to partition data points into groups or clusters based on
similarities in their features. Some popular clustering algorithms include:
• K-means: Divides data into K clusters, where each data point belongs to the cluster with the
nearest mean.
• Hierarchical clustering: Builds a hierarchy of clusters by either merging or splitting them
based on distance metrics.
• DBSCAN: Density-Based Spatial Clustering of Applications with Noise identifies clusters in
high-density areas separated by low-density regions.

2. Dimensionality Reduction: These techniques aim to reduce the number of features

(dimensions) in a dataset while preserving important information. They help in visualizing and
compressing data, as well as reducing computational complexity. Common methods include:
• Principal Component Analysis (PCA): Finds linear combinations of features that maximize
variance.

3. Association Rule Learning: This technique discovers interesting relationships or associations

between variables in large datasets. A famous algorithm is the Apriori algorithm, used for
mining frequent itemsets in transactional databases, often applied in market basket analysis.

Frequent itemset mining

It is a fundamental concept in machine learning and data mining used to discover interesting
associations or relationships between items in a dataset. It's commonly applied in market basket
analysis, recommendation systems, and other areas where understanding co-occurrences or
patterns among items is crucial.
Here's an explanation of frequent itemset mining:
1. Support and Itemset
• Itemset: A collection of one or more items grouped together. For instance, in a market
basket dataset, an itemset could be {bread, milk, eggs}.
• Support: It is a measure indicating how frequently an itemset appears in a dataset.
Mathematically, support is defined as the proportion of transactions in the dataset that
contain the itemset.

2. Frequent Itemset Mining

• Objective: Discover itemsets that have support greater than or equal to a predefined
minimum support threshold.
• Apriori Principle: This principle suggests that if an itemset is frequent, then all of its subsets
must also be frequent. This property helps in reducing the search space while mining
frequent itemsets.
 • Apriori Algorithm: A widely used algorithm for frequent itemset mining. It operates in
iterations, gradually finding itemsets with higher support.
o Initially, it finds frequent individual items (singletons) by scanning the dataset to
calculate their support.
o Then, it uses these singletons to generate candidate itemsets of length 2 (pairs) and
checks their support in the dataset.
o The algorithm continues this process, creating larger candidate itemsets by joining
frequent itemsets of length k to create candidates of length k+1, and then checking their
support.
o It stops when no new frequent itemsets can be found or when no candidate itemsets
meet the minimum support threshold.

3. Association Rule Generation

• Once frequent itemsets are discovered, association rules can be generated from these
itemsets. An association rule is an implication of the form X ➞ Y, where X and Y are
itemsets.
• Two common metrics used for association rules are:
o Confidence: Measures the likelihood of item Y being purchased when itemset X is
purchased. It's calculated as support(X ∪ Y) / support(X).
o Lift: Measures the strength of a rule by comparing the observed support of X and Y
appearing together to what would be expected if they were independent. Lift =
support(X ∪ Y) / (support(X) * support(Y)).

Applications:
• Market Basket Analysis: Understanding which items are frequently bought together to
drive product placement, marketing strategies, or bundle offerings.
• Recommendation Systems: Generating recommendations by analyzing user-item
interactions and suggesting items based on co-occurrence patterns.

Principal Component Analysis (PCA)

1. Dimensionality Reduction Technique: PCA is a technique used for reducing the
dimensionality of data by transforming it into a new coordinate system.
2. Maximizing Variance: It identifies the directions (principal components) in which the data
varies the most.
3. Orthogonal Components: Each principal component is orthogonal (uncorrelated) to each
other, capturing different aspects of the variation present in the data.
4. Preserving Information: PCA reorients data to preserve as much variance as possible in a
lower-dimensional space, often by selecting the top principal components that retain most
of the variance.
5. Mathematical Process: Involves eigenvalue decomposition or Singular Value Decomposition
(SVD) to compute the principal components.
6. Applications:
• Reducing dimensionality for visualization and computational efficiency.
• Feature extraction by transforming high-dimensional data into a lower-dimensional
space while retaining important information.
7. Assumptions:
• Linearity: PCA assumes a linear relationship between variables.
 • Gaussian Distribution: Assumes the data follows a Gaussian distribution.
8. Limitations:
• Assumes linear relationships which might not hold in all datasets.
• Might not perform well if the variance does not represent important information.
9. Use Cases:
• Image and signal processing.
• Preprocessing step in machine learning pipelines to reduce computational complexity.
• Exploratory data analysis to visualize high-dimensional data.

Tabular difference
Ensemble methods
Ensemble methods in machine learning refer to techniques that combine predictions from
multiple individual models to produce a stronger, more accurate predictive model. These methods
aim to improve the overall performance and robustness compared to using a single model.
Reinforcement Learning (RL)
It is a type of machine learning paradigm where an agent learns to make sequential decisions by
interacting with an environment to achieve a specific goal. In RL, the agent learns through a trial-
and-error process by receiving feedback in the form of rewards or penalties based on its actions.
Key components of reinforcement learning:
1. Agent: The learner or decision-maker that interacts with the environment. It observes the
environment, takes actions, and receives feedback.
2. Environment: The external system with which the agent interacts. It responds to the actions
taken by the agent and provides feedback in the form of rewards or penalties.
3. Actions: Choices made by the agent that influence the state of the environment.
4. State: Represents the current situation or configuration of the environment, which the agent
perceives before taking actions.
5. Rewards: Feedback signals provided by the environment to the agent after each action.
Rewards guide the agent toward maximizing cumulative reward over time, aligning with its
goal.
6. Policy: The strategy or set of rules that the agent uses to decide actions in different states.
7. Value Function: Estimates the expected cumulative reward an agent can obtain from a
particular state or action, helping the agent make better decisions.
8. Learning Process: The agent learns by interacting with the environment, using experiences
(state, action, reward) to update its policy or value function to make better decisions over
time.
9. Exploration vs. Exploitation: Balancing between exploring new actions and exploiting known
actions to maximize rewards while learning.

Applications of reinforcement learning:

• Game playing (e.g., AlphaGo, Atari games).
• Robotics (e.g., controlling robotic arms).
• Autonomous vehicles.
• Recommendation systems.
• Finance (e.g., portfolio management).

Temporal Difference (TD) learning

TD is a type of reinforcement learning technique used for estimating value functions or learning
from experiences in an environment without requiring a model of the environment's dynamics
(model-free learning). TD learning combines elements of Monte Carlo methods and dynamic
programming.
1. Advantages:
• More sample-efficient compared to Monte Carlo methods as it updates value estimates at
each time step rather than waiting until the end of an episode.
• Suitable for online and incremental learning scenarios.

2. Types of TD Learning:
• SARSA (State-Action-Reward-State-Action): TD learning algorithm that updates value
estimates based on the current state-action pair and the action taken next (on-policy
method).
 • Q-learning: TD learning algorithm that updates value estimates based on the current state
and the action that maximizes the value of the next state (off-policy method).

TD learning algorithms, such as SARSA and Q-learning, are fundamental in

reinforcement learning. They enable agents to learn from experiences by iteratively
updating value functions based on observed rewards and transitions between states,
facilitating efficient learning and decision-making in various environments.