The dataset’s features are the columns in the dataset matrix. The features are used in the Input or Target block in the Modeling view.
Above each feature, you can see the feature distribution as well as the label and shape.
If you select a feature you can change the label, the feature encoding, e.g., from numeric to categorical, and whether and how it should be normalized.
A feature set consists of features that you want to group for use in the Input or Target block.
The feature sets are shown above the dataset features in the Dataset view. If you select a feature set only the included features will be shown in the Datasets view.
Use to narrow down large dataset
You can use feature sets when you want to narrow down a massive dataset with a high dimensionality, that is, if the dataset has many features. For example, if you have an idea and don’t want to use the whole of your big dataset since it will take too long to train. Then you can create a feature set with only a few features and test your idea with this feature set.
You can only combine features that use the same encoding.
Moreover, the data size must be the same for all the features. For instance, you cannot create a feature set that includes a 28x28 pixel image feature and a 32x32 pixel imaghe feature, but you can combine two 28x28 pixel image features in a set.
We want to build a model that predicts the price of a house. Your training data consists of an image of a house and some numeric values about the house including its sale price. Then you should, as an input feature, create a feature set with all tabular features, excluding the price.
You can do exactly this in the tutorial "Predict California house prices".
How to create a feature set
Click New feature set.
In the Edit feature set pop-up, select the features you want to combine.
You can only combine features with the compatible dimensions.
If you need to, change the order of the features.
Name the new feature set.
Change order of features in a feature set
In most cases, the order of the features in a feature set is not relevant. The only time the order is relevant is when you want to reuse a pretrained model that was trained on features in a specific order.
You want to use a model that has been trained on a feature set consisting of
weight (in this order).
The new feature set you create for the new model must then use the same order. Otherwise, someone that is 185 cm tall and weighs 80 kg may be interpreted as 80 cm tall and 185 kg heavy, that is, considerably shorter and fatter.
The order of the features is shown in the Selected features section in the Edit feature set pop-up. This pop-up opens when you create or edit a feature set.
To change the order of the features, hover over the feature you want to move and click on the arrow.