Please enable JavaScript.

Coggle requires JavaScript to display documents.

Machine Learning, Python library - Coggle Diagram

- - - - Binary: classification task where the output variable has only two possible classes
      - MultiClass: classification task where the output variable can have more than two classes
    - - Linear: is a statistical algorithm that models the relationship between dependent and independent variables.
        
        OLS regression is a statistical technique used to estimate the relationship between independent variables (predictors) and a dependent variable (outcome). OLS regression aims to minimize errors between observed data points and predicted values.
        
        Lasso regression, also like ridge, helps simplify the model by pushing some factors to be exactly zero. This not only prevents overfitting but also improves interpretability. In finance, lasso regression can be applied to select the most influential economic indicators for predicting stock prices.
        
        Think of elastic-net as a mix of ridge and lasso. It’s useful when you have many related predictors. It keeps the model stable like ridge and does some feature selection like lasso, giving you the best of both worlds. In genetics, when a trait is influenced by many genes, elastic-net regression helps find the important genes even when they’re connected to each other. It’s like untangling a web, making it easier to determine which genes matter for a specific trait.
        
        Ridge regression prevents the model from getting too complex, especially when there are relationships between predictors. It’s like a guardrail against overfitting, helping the model make more reliable predictions. For example, in scenarios such as predicting housing prices, where multiple correlated features (for example, number of bedrooms and square footage) can lead to overfitting, ridge regression helps maintain stability of the model.
        
        Non-negative least squares regression variant is a method that ensures that the weights assigned to the input features in the regression model are positive or zero (that is, non-negative), making it useful when dealing with such things as wind speed or travel times, where negative values can’t occur.
      - Logistic: Logistic regression is a statistical method where the goal is to predict the probability that an instance belongs to one of two classes.
      - Decision Tree: Decision trees is a robust algorithm and gets to the crux of complex problems with simple yes-or-no branches.
      - Support Vector Machinesare a type of supervised learning algorithm used for classification and regression tasks.
        
        In many real-world problems, the data cannot be divided into classes by a simple line. The SVM algorithm can still be applied using what is known as the kernel SVM trick, which uses curved lines instead of straight lines to separate the classes.
        
        The main objective of linear SVM is to find the line or plane that best separates classes in the target variable from each other.
      - K-nearest neighborsThe KNN algorithm is a non-parametric, supervised learning classifier, which uses proximity to make classifications or predictions about the grouping of an individual data point.2
  - - - The evaluation of model performance involves careful consideration of performance metrics, appropriate data splitting techniques, and interpretation of results to assess the model’s effectiveness in solving the task at hand
      - The adjustment of model parameters is an iterative phase for adjusting or fine-tuning the parameters for better results.
      - Model selection involves choosing the appropriate algorithm or architecture, while training involves optimizing the model’s parameters using labeled data to make accurate predictions on unseen examples.
      - Data Segmentation refers to dividing the available data set into distinct subsets for different purposes, a training set and a testing set
- - - - uniform manifold approximation and projection, or UMAP
      - T-distributed stochastic neighbor embedding
    - - Prinicipal component analysis
      - Singular value decomposition (SVD)
      - independent component analysis (ICA)