Skip to main content

Classification and Logistic Regression

1. Classification Problem

  • Definition: Classification is a supervised learning task where the output variable y is discrete-valued rather than continuous.
  • In particular, consider binary classification where y {0,1} (e.g., spam detection: spam =1, not spam =0).
  • Each training example is a pair (x(i), y(i)), where x(i)Rd is a feature vector, and y(i) is the label.

2. Why Not Use Linear Regression for Classification?

  • Linear regression tries to predict continuous values, which is problematic for classification as the prediction can be outside [0,1].
  • For example, predicting y1.5 or −0.2 is meaningless when y is binary.
  • Instead, we want the output (x) to be interpreted as the probability that y=1 given x.

3. Logistic Regression Model

Hypothesis:

(x)=g(θTx)=1+e−θTx1,

where:

  • g(z)=1+e−z1 is the sigmoid function, which maps any real value to the interval (0, 1).
  • θRd+1 are parameters (including intercept term).
  • (x) can be interpreted as the estimated probability P(y=1x;θ).

Decision Boundary:

  • Predict y=1 if (x)0.5; otherwise, predict y=0.
  • The decision boundary corresponds to θTx=0, which is a linear boundary in input space.

4. Loss Function and Cost Function

Probability Model:

  • Logistic regression models conditional probability directly:

P(y=1x;θ)=(x),P(y=0x;θ)=1(x).

  • Equivalently, likelihood for data point (x(i),y(i)):

p(y(i)x(i);θ)=((x(i)))y(i)(1(x(i)))1y(i).

Cost (Loss) Function:

  • Use negative log-likelihood (cross-entropy loss) as cost per example:

J(i)(θ)=[y(i)log(x(i))+(1y(i))log(1(x(i)))].

  • Overall cost function (average over n examples):

J(θ)=n1i=1nJ(i)(θ).

  • This loss is convex in θ, enabling efficient optimization.

5. Training Logistic Regression

·         Use methods such as gradient descent or more advanced optimization (Newton's method, quasi-Newton) to minimize cost J(θ).

·         The gradient of the cost function is:

θJ(θ)=n1i=1n((x(i))y(i))x(i).

  • Update rule in gradient descent:

θ:=θαθJ(θ),

where α is the learning rate.


6. Multi-class Classification

·         When y{1,2,...,k} for k>2, logistic regression generalizes to multinomial logistic regression or Softmax regression.

·         Model outputs hˉθ(x)Rk called logits.

·         The Softmax function converts logits into probabilities:

P(y=jx;θ)=s=1kexp(hˉθ(x)s)exp(hˉθ(x)j).

  • Loss for example (x(i),y(i)) is the negative log-likelihood:

J(i)(θ)=logP(y(i)x(i);θ).


7. Discriminative vs. Generative Classification Algorithms

  • Discriminative algorithms (like logistic regression) model p(yx) directly or learn a direct mapping from x to y.
  • Generative algorithms model the joint distribution p(x,y)=p(xy)p(y).
  • Example: Gaussian Discriminant Analysis (GDA).
  • Logistic regression is an example of a discriminative approach focusing purely on p(yx).

8. Linear Hypothesis Class and Decision Boundaries

  • Logistic regression hypothesis class:

H={:(x)=1{θTx0}},

which are classifiers with linear decision boundaries.

  • More generally, hypothesis classes can be extended to neural networks or other complex architectures.

9. Perceptron Learning as Contrast to Logistic Regression

·         Perceptron also uses a linear classifier but with a different loss and update rule.

·         Logistic regression provides probabilistic outputs and optimizes a convex cost function, generally yielding better statistical properties.


10. Practical Considerations

  • Feature scaling often improves numerical stability.
  • Regularization (e.g., L2) is frequently added to cost to prevent overfitting.
  • Logistic regression handles input features linearly; non-linear boundaries require feature engineering or kernel methods.

Summary:

Logistic regression is a fundamental classification algorithm that models the conditional probability of the positive class using a sigmoid of a linear function of input features. It is trained via maximizing likelihood (or minimizing cross-entropy loss) and extends naturally to multi-class problems via Softmax. It is a discriminative model focusing directly on p(yx) and yields linear decision boundaries. It contrasts with generative models by its direct approach to classification.

 

Comments

Popular posts from this blog

Research Process

The research process is a systematic and organized series of steps that researchers follow to investigate a research problem, gather relevant data, analyze information, draw conclusions, and communicate findings. The research process typically involves the following key stages: Identifying the Research Problem : The first step in the research process is to identify a clear and specific research problem or question that the study aims to address. Researchers define the scope, objectives, and significance of the research problem to guide the subsequent stages of the research process. Reviewing Existing Literature : Researchers conduct a comprehensive review of existing literature, studies, and theories related to the research topic to build a theoretical framework and understand the current state of knowledge in the field. Literature review helps researchers identify gaps, trends, controversies, and research oppo...

Mglearn

mglearn is a utility Python library created specifically as a companion. It is designed to simplify the coding experience by providing helper functions for plotting, data loading, and illustrating machine learning concepts. Purpose and Role of mglearn: ·          Illustrative Utility Library: mglearn includes functions that help visualize machine learning algorithms, datasets, and decision boundaries, which are especially useful for educational purposes and building intuition about how algorithms work. ·          Clean Code Examples: By using mglearn, the authors avoid cluttering the book’s example code with repetitive plotting or data preparation details, enabling readers to focus on core concepts without getting bogged down in boilerplate code. ·          Pre-packaged Example Datasets: It provides easy access to interesting datasets used throughout the book f...

Distinguishing Features of Vertex Sharp Transients

Vertex Sharp Transients (VSTs) have several distinguishing features that help differentiate them from other EEG patterns.  1.       Waveform Morphology : §   Triphasic Structure : VSTs typically exhibit a triphasic waveform, consisting of two small positive waves surrounding a larger negative sharp wave. This triphasic pattern is a hallmark of VSTs and is crucial for their identification. §   Diphasic and Monophasic Variants : While triphasic is the most common form, VSTs can also appear as diphasic (two phases) or even monophasic (one phase) waveforms, though these are less typical. 2.      Phase Reversal : §   VSTs demonstrate a phase reversal at the vertex (Cz electrode) and may show phase reversals at adjacent electrodes (C3 and C4). This characteristic helps confirm their midline origin and distinguishes them from other EEG patterns. 3.      Location : §   VSTs are primarily recorded from midl...

Distinguishing Features of K Complexes

  K complexes are specific waveforms observed in electroencephalograms (EEGs) during sleep, particularly in stages 2 and 3 of non-REM sleep. Here are the distinguishing features of K complexes: 1.       Morphology : o     K complexes are characterized by a sharp negative deflection followed by a slower positive wave. This biphasic pattern is a key feature that differentiates K complexes from other EEG waveforms, such as vertex sharp transients (VSTs). 2.      Duration : o     K complexes typically have a longer duration compared to other transient waveforms. They can last for several hundred milliseconds, which helps in distinguishing them from shorter waveforms like VSTs. 3.      Amplitude : o     The amplitude of K complexes is often similar to that of the higher amplitude slow waves present in the background EEG. However, K complexes can stand out due to their ...

Maximum Stimulator Output (MSO)

Maximum Stimulator Output (MSO) refers to the highest intensity level that a transcranial magnetic stimulation (TMS) device can deliver. MSO is an important parameter in TMS procedures as it determines the maximum strength of the magnetic field generated by the TMS coil. Here is an overview of MSO in the context of TMS: 1.   Definition : o   MSO is typically expressed as a percentage of the maximum output capacity of the TMS device. For example, if a TMS device has an MSO of 100%, it means that it is operating at its maximum output level. 2.    Significance : o    Safety : Setting the stimulation intensity below the MSO ensures that the TMS procedure remains within safe limits to prevent adverse effects or discomfort to the individual undergoing the stimulation. o Standardization : Establishing the MSO allows researchers and clinicians to control and report the intensity of TMS stimulation consistently across studies and clinical applications. o   Indi...