MyML

Just a machine learning lib written in Dotnet 8 framework

Features

• Naive Bayes Classifier (Multinomial): designed for text classification tasks. The Multinomial variant is suitable for handling text data with multiple classes and word occurrences.

• Similarity Search

  • Jaccard Similarity Search: Measures similarity between sets, ideal for categorical or keyword-based data.
  • Cosine Similarity Search: Measures the cosine similarity between text/vector pairs, commonly used in recommendation systems.

• Generic Design: The library is designed to be generic and adaptable, allowing you to train classifiers for various types of data and labels.

• Easy Training: Training a classifier is straightforward, and the library handles essential tasks such as feature extraction and Laplace (add-one) smoothing automatically.

Usage

Naive Bayes Classifier

// Create an instance of the classifier
var classifier = new MultinomialNaiveBayesClassifier();

// Train the classifier with your training data (ClassifierModel)
classifier.Train(trainingData);

// Make predictions
var prediction = classifier.Predict(textToClassify);

Similarity Search

Jaccard Similarity Search

// Create an instance of the Jaccard similarity search
SimilaritySearch search = new JaccardSimilaritySearch();

// Query and documents
string query = "gato no jardim";
var documents = new List<string>
{
    "O gato está no jardim.",
    "O cachorro brinca no parque.",
    "A árvore é alta e verde."
};

// Perform the search
var results = search.Search(query, documents);

Cosine Similarity Search

// Create an instance of the Cosine similarity search
SimilaritySearch search = new CosineSimilaritySearch();

// Query and documents
string query = "gato no jardim";
var documents = new List<string>
{
    "O gato está no jardim.",
    "O cachorro brinca no parque.",
    "A árvore é alta e verde."
};

// Perform the search
var results = search.Search(query, documents);

TODO

• Do unit tests with possible use cases
• Try other Supervised Learning (Decision Trees,Logistic Regression)
• Integrate these features into the classifier (e.g., similarity-based predictions or clustering).
• Explore ranking, leveraging these similarity metrics.