Feature-Cost Sensitive Learning with Submodular Trees of Classifiers

Authors: Matt Kusner, Wenlin Chen, Quan Zhou, Zhixiang (Eddie) Xu, Kilian Weinberger, Yixin Chen

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate our approach on a real-world feature-cost sensitive ranking dataset: the Yahoo! Learning to Rank Challenge dataset. We begin by describing the dataset and show Precision@5 per cost compared against CSTC (Xu et al. 2014) and another cost-sensitive baseline. We then present results on a diverse set of non-cost sensitive datasets, demonstrating the flexibility of our approach. For all datasets we evaluate the training times of our approach compared to CSTC for varying tree budgets.
Researcher Affiliation Academia Washington University in St. Louis, 1 Brookings Drive, MO 63130 Tsinghua University, Beijing 100084, China {mkusner, wenlinchen, zhixiang.xu, kilian, ychen25}@wustl.edu zhouq10@mails.tsinghua.edu.cn
Pseudocode Yes Algorithm 1 ASTC in pseudo-code.
Open Source Code No The paper does not provide any statement or link indicating the release of open-source code for the described methodology.
Open Datasets Yes We evaluate our approach on a real-world feature-cost sensitive ranking dataset: the Yahoo! Learning to Rank Challenge dataset (Chen et al. 2012).
Dataset Splits No The paper mentions 'hyperparameter tuning on a validation set' but does not provide specific details on the dataset splits (e.g., percentages or counts) for training, validation, or test sets.
Hardware Specification No The paper mentions 'Computations were performed via the Washington University Center for High Performance Computing', but does not provide specific hardware details such as CPU/GPU models, memory, or other specifications.
Software Dependencies No The paper does not provide specific software dependencies or version numbers for its implementation.
Experiment Setup Yes For both algorithms we set a maximum tree depth of 5. and We set a new-feature budget B identical for each node in the tree and then greedily select new features up to cost B for each node. and Finally, we set node thresholds θk to send half of the training inputs to each child node.