SGDLibrary: A MATLAB library for stochastic optimization algorithms

Authors: Hiroyuki Kasai

JMLR 2017 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Illustrative results additionally including SQN and SVRG-LBFGS are presented in Figure 1, which are generated by display_graph(), and display_classification_result() specialized for classification problems. Thus, SGDLibrary provides rich visualization tools as well.
Researcher Affiliation Academia Hiroyuki Kasai EMAIL Graduate School of Informatics and Engineering The University of Electro-Communications Tokyo, 182-8585, Japan
Pseudocode Yes Listing 1: Demonstration code for logistic regression problem. 1 % generate synthetic 300 samples of dimension 3 for logistic regression 2 d = logistic_regression_data_generator (300 ,3); 3 % define logistic regression problem 4 problem = logistic_regression (d.x_train ,d.y_train ,d.x_test ,d.y_test); 6 options.w_init = d.w_init; % set initial value 7 options.step_init = 0.01; % set initial stepsize 8 options.verbose = 1; % set verbose mode 9 [w_sgd , info_sgd] = sgd(problem , options); % perform SGD solver 10 [w_svrg , info_svrg] = svrg(problem , options); % perform SVRG solver 11 [w_svrg , info_svrg] = sqn(problem , options); % perform SQN solver 12 % display cost vs. number of gradient evaluations 13 display_graph( grad_calc_count , cost ,{ SGD , SVRG } ,... 14 {w_sgd ,w_svrg },{info_sgd ,info_svrg });
Open Source Code Yes The code is available at https://github.com/hiroyuki-kasai/SGDLibrary.
Open Datasets No First, we generate train/test datasets d using logistic_regression_data_generator(), where the input feature vector is with n = 300 and d = 3.
Dataset Splits No The paper mentions generating synthetic train/test datasets for logistic regression with n=300 and d=3, but does not specify the split ratio or counts for these datasets.
Hardware Specification No The paper does not provide any specific details about the hardware used for the experiments.
Software Dependencies No The paper mentions that SGDLibrary is a 'pure-MATLAB library' and is 'operable on GNU Octave', but it does not specify any version numbers for MATLAB, Octave, or any other software dependencies.
Experiment Setup Yes options.w_init = d.w_init; % set initial value options.step_init = 0.01; % set initial stepsize options.verbose = 1; % set verbose mode