Data:AI&ML Challenge Dataset: Difference between revisions

Latest revision as of 15:44, 30 May 2019

AI&ML Challenge Dataset dataset information page

Short Description

Example Challenge Predictands

Statement: Machine Learning 'Challenge Dataset' for the Seabed

Abstract: We present data for use with ML, dealing with the sediment/rock substrates of the NE USA Continental Margin. The pointwise Labelled Data from seabed observations should be spatially extended over the entire area in an intelligent way. To aid that, environmental Feature Layers can be employed to train any chosen Machine Learning method. The trained ML model is then extended across all the vacant areas. The result predicts what the seabed is made of, so that survey operations (including research) can be planned, or biogeochemical budgets can be calculated. The idea of the Challenge Dataset is to permit people - researchers and students - to experiment with their own Machine Learning algorithms and data-preparation adjustments to achieve the BEST POSSIBLE mapping over the area. Metrics on the uncertainties should also be computed. For the time being the mappings are in terms of mud/sand/gravel, rock exposure, and carbonate and organic carbon contents. See the Powerpoint file in the Zipfile for further instructions.

Data format

Data type:	Substrates
Data origin:	Measured
Data format:
Other format:	zip
Data resolution:	~1km
Datum:	WGS84

Data Coverage

Spatial data coverage: NE USA Continental Margin
Temporal data coverage: Time averaged
Time period covered: Post 1930

Availability

Download data: http://instaar.colorado.edu/~jenkinsc/CSDMS AI&ML/DataChallenge/DataChallenge 4AI&ML.zip
Data source: http://instaar.colorado.edu/~jenkinsc

@@ Line 1: / Line 1: @@
 {{Data description
 |One-line data description=Machine Learning 'Challenge Dataset' for the Seabed
-|Extended data description=Data for use with ML, dealing with the sediment/rock substrates of the NE USA Continental Margin. Training data from seabed observations should be spatially extended over the entire area in an intelligent way. To aid that environmental Feature Layers are employed to train various Machine Learning methods on the sample data then the results are extended across all the vacant areas. The result predicts what the seabed is made of, so that survey operations (including research) can be planned, or biogeochemical budgets can be calculated. The idea of the Challenge Dataset is to permit people - researchers and students - to experiment with various Machine Learning algorithms and data preparation adjustments to achieve the BEST possible mapping over the area. The mappings are in terms of mud/sand/gravel, exposed rock, carbonate, organic carbon, but other parameters are possible also. See the PPT file in the Zipfile for further instructions.
+|Extended data description=We present data for use with ML, dealing with the sediment/rock substrates of the NE USA Continental Margin. The pointwise Labelled Data from seabed observations should be spatially extended over the entire area in an intelligent way. To aid that, environmental Feature Layers can be employed to train any chosen Machine Learning method. The trained ML model is then extended across all the vacant areas. The result predicts what the seabed is made of, so that survey operations (including research) can be planned, or biogeochemical budgets can be calculated. The idea of the Challenge Dataset is to permit people - researchers and students - to experiment with their own  Machine Learning algorithms and data-preparation adjustments to achieve the BEST POSSIBLE mapping over the area. Metrics on the uncertainties should also be computed. For the time being the mappings are in terms of mud/sand/gravel, rock exposure, and carbonate and organic carbon contents. See the Powerpoint file in the Zipfile for further instructions.
 |Upload image dataset=seabedStack v01.png
 |Caption dataset image=Example Challenge Predictands
 }}
 {{Data format