StochasticRank: Global Optimization of Scale-Free Discrete Functions
Authors: Aleksei Ustimenko, Liudmila Prokhorenkova
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our algorithm is implemented as a part of the Cat Boost gradient boosting library and outperforms the existing approaches on several learning-to-rank datasets. In addition to ranking metrics, our framework applies to any scale-free discrete loss function. Finally, Section 7 empirically compares the proposed algorithm with existing approaches, and Section 8 concludes the paper. The results are shown in Table 2. |
| Researcher Affiliation | Collaboration | 1Yandex, Moscow, Russia 2Moscow Institute of Physics and Technology, Dolgoprudny, Moscow Region, Russia 3Higher School of Economics, Moscow, Russia. |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Stochastic Rank is implemented within the official Cat Boost library (Prokhorenkova et al., 2018; Cat Boost, 2020). |
| Open Datasets | Yes | For our experiments, we use the following publicly available datasets. First, we use the data from YAHOO! Learning to Rank Challenge (Chapelle & Chang, 2011): there are two datasets, each is pre-divided into training, validation, and testing parts. The other datasets are WEB10K and WEB30K released by Microsoft (Qin & Liu, 2013). |
| Dataset Splits | Yes | First, we use the data from YAHOO! Learning to Rank Challenge (Chapelle & Chang, 2011): there are two datasets, each is pre-divided into training, validation, and testing parts. We tune the hyperparameters using 500 iterations of random search and select the best combination using the validation set, the details are given in the supplementary materials (Section F). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions "Cat Boost" as the library used, but does not provide specific version numbers for it or any other software dependencies. |
| Experiment Setup | Yes | We limited the tree depth parameter to 3, so one tree can separate all documents with different features. We set the number of iterations to 1000, learning rate to 0.1, diffusion temperature to 103, and model-shrink-rate to 10 3. For all algorithms, we set the maximum number of trees to 1000. |