Edit an imported dataset for use in experiments
Imported datasets are configured automatically, but you can do some editing to make them more appropriate to your experiment. Although the platform remembers the changes that you make, you need to manually save your changes as a new version of the dataset before you can use it to run an experiment.
You are not able to edit an already saved version of the dataset, so you will need to duplicate it if you want to make further changes.
Edit and inspect datasets
Dataset features
You can inspect your dataset either in the Features view or in the Table view. Just toggle the switch.
Features view to inspect features
Each row displays the name of a feature, its shape, and the distribution of values.
Click on the spanner icon to view or update the feature settings.
Table view to inspect features
Each column displays the name of a feature, its shape, and the distribution of values. You can also see a preview of the values for the first examples.
Click on the spanner icon to view or update the feature settings.
Note that to speed up processing, the platform only analyzes a fraction of the examples contained in a dataset. This has two consequences:
-
The information presented has a small degree of approximation.
-
The shape and the encodings available for a feature are determined only from the analyzed examples.
Dataset features
Group features into feature sets, making several features available as a single input to the model.
Subsets
Distribute the examples of the dataset into subsets. You always need to have at least two subsets, one for training and one for validating the model.
Incompatible examples
Every example must have its features compatible with the shape and encoding indicated in the dataset’s version, or running your experiment will fail.
If your experiment fails because of incompatible examples, you can click the icon under the experiment’s name to get more information about the problematic examples.