PASSCoDe: Parallel ASynchronous Stochastic dual Co-ordinate Descent
Authors: Cho-Jui Hsieh, Hsiang-Fu Yu, Inderjit Dhillon
ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that our methods are much faster than previous parallel coordinate descent solvers. |
| Researcher Affiliation | Academia | Cho-Jui Hsieh CJHSIEH@CS.UTEXAS.EDU Hsiang-Fu Yu ROFUYU@CS.UTEXAS.EDU Inderjit S. Dhillon INDERJIT@CS.UTEXAS.EDU Department of Computer Science, The University of Texas, Austin, TX 78721, USA |
| Pseudocode | Yes | Algorithm 1 Stochastic Dual Coordinate Descent (DCD) Input: Initial α and w = Pn i=1 αixi 1: while not converged do 2: Randomly pick i 3: Update αi αi + αi, where αi arg min δ 1 2 w +δxi 2 +ℓ i ( (αi +δ)) (4) 4: Update w by w w + αixi 5: end while |
| Open Source Code | Yes | Our code is available in http://www.cs.utexas.edu/ rofuyu/ exp-codes/passcode-icml15-exp/. |
| Open Datasets | Yes | We consider five datasets: news20, covtype, rcv1, webspam, and kddb. Detailed information is shown in Table 3. |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits, such as percentages or sample counts for each split. It mentions 'n is the number of test instances' in Table 3 but no specific breakdown. |
| Hardware Specification | No | All the experiments are performed on an Intel multi-core dualsocket machine with 256 GB memory. Each socket is associated with 10 computation cores. We explicitly enforce that all the threads use cores from the same socket to avoid inter-socket communication. |
| Software Dependencies | No | To have a fair comparison, we implement all methods in C++ using Open MP as the parallel programming framework. |
| Experiment Setup | Yes | We follow the description in (Liu & Wright, 2014; Liu et al., 2014) to implement Asy SCD with step length γ = 1/2 and the shuffling period p = 10 as suggested in (Liu et al., 2014). We implement a multi-core version of Co Co A (Jaggi et al., 2014) with βK = 1 and DCD as its local dual method. |