Scalable Non-linear Learning with Adaptive Polynomial Expansions
Authors: Alekh Agarwal, Alina Beygelzimer, Daniel J. Hsu, John Langford, Matus J Telgarsky
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | an extensive experimental study in Section 3 compares efficient implementations of these baselines with the proposed mechanism and gives strong evidence of the latter s dominant computation/prediction tradeoff ability (see Figure 1 for an illustrative summary). |
| Researcher Affiliation | Collaboration | Alekh Agarwal Microsoft Research alekha@microsoft.com Alina Beygelzimer Yahoo! Labs beygel@yahoo-inc.com Daniel Hsu Columbia University djhsu@cs.columbia.edu John Langford Microsoft Research jcl@microsoft.com Matus Telgarsky Rutgers University mtelgars@cs.ucsd.edu |
| Pseudocode | Yes | Algorithm 1 Adaptive Polynomial Expansion (apple) |
| Open Source Code | Yes | To this end, we implemented apple in the Vowpal Wabbit (henceforth VW) open source machine learning software1. ... 1Please see https://github.com/John Langford/vowpal_wabbit and the associated git repository, where -stage_poly and related command line options execute apple. |
| Open Datasets | Yes | To this end, we compiled a collection of 30 publicly available datasets, across a number of KDDCup challenges, UCI repository and other common resources (detailed in the appendix). |
| Dataset Splits | No | For all the datasets, we tuned the learning rate for each learning algorithm based on the progressive validation error (which is typically a reliable bound on test error) [24]. ... For each dataset, we performed a random split with 80% of the data used for training and the remainder for testing. While validation error is mentioned, no explicit validation split percentage or size is provided. |
| Hardware Specification | No | The paper mentions data storage on a "large Hadoop cluster" and experiments on "100 nodes" but does not provide specific hardware details like CPU/GPU models or memory specifications for running experiments. |
| Software Dependencies | No | The paper states "we implemented apple in the Vowpal Wabbit (henceforth VW) open source machine learning software" but does not provide specific version numbers for VW or any other software dependencies. |
| Experiment Setup | Yes | We implemented apple such that the total number of epochs was always 6 (meaning 5 rounds of adding new features). ... The number of bits in hashing was set to 18 for all algorithms, apart from cubic polynomials, where using 24 bits for hashing was found to be important for good statistical performance. For all datasets, we used squared-loss to train, and 0-1/squared-loss for evaluation in classification/regression problems. |