Evaluation datasets may be retrieved from the San Diego Supercomputer Center using the wget program. See retrieval of the low-frequency dataset and analyst annotations for details on the wget program and how to use it. There are two datasets:
The high-frequency evaluation data are approximately 1.94 TB and the wget program should be invoked from the command line as follows (all on one line):
wget --no-check-certificate –m https://cloud.sdsc.edu/v1/AUTH_dclde/SocalHFEvaluationData/
The low-frequency evaluation data are approximately 14 GB and the wget program should be invoked from the command line as follows (all on one line):
wget --no-check-certificate –m https://cloud.sdsc.edu/v1/AUTH_dclde/SocalLFEvaluationData/
Evaluation datasets are designed to provide a method of evaluating algorithms on data that are independent of the data used during the development of detector/classifiers. As such, these data should not be part of “data plowing:” the development cycle where one iteratively tunes detection/classification algorithms by adjusting methods to reduce error rates. In a sense, data plowing is a form of training, and the proper use of the evaluation dataset helps to provide a better indication of how well an algorithm might be expected to perform under field conditions. Consequently, do not use the DCLDE evaluation data as part of your development process.
Labels were established using the same methods as for the DCLDE development data and can be scored with the same tools that were provided for the development data. While the labels were checked by multiple analysts, humans make mistakes too, and should you find what you believe to be an error in the labels, please provide us constructive feedback at dcldeworkshop@gmail.com. Labels are provided as zip archives, and the date in the zip archive filename indicates the date that the labels were released/finalized.
Low frequency labels (2015-07-14-LF.zip)
High frequency labels (2015-07-14-HF.zip)