Fast and Accurate Least-Mean-Squares Solvers

Authors: Alaa Maalouf, Ibrahim Jubran, Dan Feldman

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results and complete open source code are also provided. Extensive experimental results on synthetic and real-world data for common LMS solvers of Scikit-learn library with either CPython or Intel s Python distributions. In this section we apply our fast construction of the Carathoodory Set S from the previous section to boost the running time of common LMS solvers in Table 1 by a factor of tens to hundreds, or to improve their numerical accuracy by a similar factor to support, e.g., 32 bit floating point representation as in Fig. 2v.
Researcher Affiliation Academia The Robotics and Big Data Lab, Department of Computer Science, University of Haifa, Haifa, Israel
Pseudocode Yes Algorithm 1 FAST-CARATHEODORY-SET(P, u, k); see Theorem 3.1; Algorithm 2 CARATHEODORY-MATRIX(A, k); see Theorem 3.2; Algorithm 3 LMS-CORESET(A, b, m, k); Algorithm 4 LINREG-BOOST(A, b, m, k); Algorithm 5 RIDGECV-BOOST(A, b, A, m, k); Algorithm 6 LASSOCV-BOOST(A, b, A, m, k); Algorithm 7 ELASTICCV-BOOST(A, b, m, A, ρ, k)
Open Source Code Yes Extensive experimental results and complete open source code are also provided. Open code for our algorithms is provided [29]. [29] Alaa Maalouf, Ibrahim Jubran, and Dan Feldman. Open source code for all the algorithms presented in this paper, 2019. Link for open-source code.
Open Datasets Yes (i) 3D Road Network (North Jutland, Denmark) [24]. (ii) Individual household electric power consumption [1]. (iii) House Sales in King County, USA [2].
Dataset Splits Yes m-folds cross validation (CV). We briefly discuss the CV technique which is utilized in common LMS solvers. Given a parameter m and a set of real numbers A, to select the optimal value α A of the regularization term, the existing Python s LMS solvers partition the rows of A into m folds (subsets) and run the solver m |A| times, each run is done on a concatenation of m 1 folds (subsets) and α A, and its result is tested on the remaining test fold .
Hardware Specification Yes All the experiments were conducted on a standard Lenovo Z70 laptop with an Intel i7-5500U CPU @ 2.40GHZ and 16GB RAM.
Software Dependencies No The paper mentions "Scikit-learn library" and "CPython or Intel s Python distributions" but does not provide specific version numbers for these software components.
Experiment Setup Yes The chosen number of clusters in Algorithm 3 is k = 2(d + 1)2 + 2 in order to have O(log n) iterations in Algorithm 1, and ρ = 0.5 for Algorithm 7.