myalgo-core 0.1.3

ML algorithms with C++ backends

MyAlgo

MyAlgo is my Machine Learning project which is focused on building Linear Regression and Artificial Neural Network from scratch without using high level machine learning libraries such as Tensorflow, PyTorch and scikit-learn.

The main goal of this project is to combining

• python as a user facing python API
• C++ as a high performance computational backend (only for numerical calculations)

Read More : Documentation

Implemented Algorithms

Linear Regression

Feature	Supported values
Methods	Normal Equation, GD, SGD
Regularization	Lasso, Ridge, Elastic net

Example usage:

model = LinearRegression(method='gd', lr=0.01, epochs=1000)
model.fit(X_train, y_train)
predictions = model.predict(X_test)

2. ANN

Feature	Supported values
Layers	Flatten, Dense
Activations	ReLU, sigmoid, Tanh, Softmax
Loss functions	MSE, cross-entropy

Example usage:

model = NeuralNetwork(
  Flatten((28,28)),
  Dense(128, activation='relu'),
  Dense(10, activation='softmax')
)
model.compile(loss='categorical_crossentropy', lr=0.01)
model.train(X_train, y_train, epochs=11, batch_size=32)

Installation

pip install myalgo

Results of MINST dataset training

Author

Vitthal Humbe — github.com/vitthalhumbe B.Tech second year, AI/ML