Instance-Level Metalearning for Outlier Detection
Authors: Long Vu, Peter Kirchner, Charu C. Aggarwal, Horst Samulowitz
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This section presents the experimental evaluation of T-Auto OD on a variety of data sets. |
| Researcher Affiliation | Collaboration | Long Vu , Peter Kirchner , Charu C. Aggarwal , Horst Samulowitz IBM Research lhvu@us.ibm.com, pdk22@cornell.edu, {charu, samulowitz}@us.ibm.com |
| Pseudocode | Yes | Algorithm 1 T-Auto OD-Train and Algorithm 2 T-Auto OD-Test |
| Open Source Code | Yes | 5https://github.com/t-autood/t-autood |
| Open Datasets | Yes | All data sets were derived from Open ML 6, and were generated from classification data sets (after possible sparsification). 6https://www.openml.org |
| Dataset Splits | Yes | We then split these 520 derived data sets into a training set of 416 derived data sets and testing set of 104 derived data sets. The LGBM model was then trained using the training/validation data of the holdout partitions. |
| Hardware Specification | No | The paper does not mention any specific hardware used for running the experiments. |
| Software Dependencies | No | The paper mentions 'scikit-learn open-source package', 'LGBM classifier', and 'XGB model', but does not provide specific version numbers for any of these software dependencies. |
| Experiment Setup | Yes | The hyper-parameters for the pipelines were chosen at random from a recommended range as suggested in the scikit-learn implementations of these algorithms. We used an out-of-the-box LGBM classifier4 with its default hyperparameters in this paper. The value of d used was 20. Furthermore, we calculated the Mahalanobis distance of data instances in the data sets and added this distance as an additional feature. |