Sign in to view source links and access this dataset
Description
DarkMachines secret dataset belongs to an unsupervised challenge requiring participants to submit anomaly scores. The challenge includes four separate data channels with specified submission file lengths: 75,528 for channel 1, 10,101 for channel 2a, 1,342,933 for channel 2b, and 1,994,639 for channel 3. Authored by Melissa van Beekveld of the University of Oxford and hosted on Papers with Code, this dataset is intended for a competition.
Use Cases
Develop unsupervised anomaly detection models based on the challenge's unlabeled data.
Benchmark anomaly scoring algorithms against the challenge's specified data channels.
Prepare competition submissions by generating files with 'S' or 'B' labels and anomaly scores.
Strengths
Provides exact row counts for four distinct data channels: 75,528, 10,101, 1,342,933, and 1,994,639.
Associated with a specific, structured challenge from a known academic author and institution.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is known only for the challenge submission files; the original dataset's full structure is unknown.
Provenance
Source
University of Oxford
Collection Method
Created for the DarkMachines unsupervised challenge.
Submission requires generating files with specific naming conventions (_chanx.csv) and formats (label and score separated by ';').