Skip to content

Latest commit

 

History

History
120 lines (58 loc) · 3.05 KB

README.md

File metadata and controls

120 lines (58 loc) · 3.05 KB
Raw

This is a machine learning library developed by Lingyi Fu (u1452585) for CS5350/6350 in University of Utah

How to Run the algorithms?

Please run "main_bank.py" file in the EnsembleLearning folder for the Adaboost, Bagging trees, Random Forest algorithm for the Bank dataset.

Please run "credit.py" file in the EnsembleLearning folder for the Adaboost, Bagging trees, Random forest algorithm for the credit dataset.

Please run "LMSregression.py" file in the LinearRegression folder for the LMS with batch-gradient and stochastic gradient.

Please run the "Perceptron.py" file in the Perceptron folder for the Standard Perceptron/Voted Perceptron/Averaged Perceptron.

Please run the "main.ipynb" file in the SVM folder for the SVM with Primal form, Dual form, Gaussian kernel, and kernel perceptron algorithm.

Please run the "Neuronetwork.ipynb" file in the NeuralNetworks folder for the NeuralNetwork algorithm.

Please run the "logisticregression.ipynb" file in the Logisticregression folder for the logistic regression with MAP estimation and ML estimation.

How to set the hyperparameters?

  1. Adaboost

  • features: features of the dataset

  • num_iterations: the number of epoches that will be runed to update the weights #500

  1. Bagging

  • features: features of the dataset

  • tree_counts = number of trees #500

  • max_depth = the depth of each tree #len(features)

  1. Random forest

  • features: features of the dataset

  • tree_counts: nuber of trees #500

  • feature_subsets: random subset of features #[2, 4, 6]

  • max_depth: the depth of each tree #len(features)

  1. LMS with batch-gradient
  • learning_rate1: the step size at each iteration #0.0001
  1. Stochastic gradient
  • learning_rate1: the step size at each iteration #0.0001
  1. Standard perceptron/Voted perceptron/Averaged perceptron

  • epoches: number of iterations

  • learning_rate: the spead of learning

  1. SVM - PrimalSVM

  • Cs: regularization parameter

  • gamma_values: learning rate, te spead of learning

  • a_values: parametrs for calculating "schedule of learning rate"

  • N: Number of training samples, if it is Stochastic regression, each iteration is 1.

  • epochs: number of iterations

  1. SVM - DualSVM
  • Cs: regularization parameter
  1. SVM - Gaussian SVM

  • Cs: regularization parameter

  • gammas: learning rate

  1. SVM - Kernel Perceptron
  • gammas: learning rate
  1. Neural Networks

  • widths: width of hidden layer

  • depths: the number of hidden layers

  • gamma_0: learning rate

  • d: hyperparmeters for calculating scheduled learning rate

  • epochs: number of iterations

  • activation_functions: different active function ("tanh", "relu") for introducing non-linearity into a neural network.

  1. Logistic regression with MAP estimation

  • variances: hyperparmeters for calculating MAP estimation

  • gamma_0_values: learning rate

  • d_values: hyperparmeters for calculating scheduled learning rate

  • epochs: number of iterations

  1. Logistic regression with ME estimation

  • gamma_0_values: learning rate

  • d_values: hyperparmeters for calculating scheduled learning rate

  • epochs: number of iterations