Column DType

When a user creates a GNNTable object, the learning engine automatically infers the data type (dtype) for all columns in the table. A column’s dtype determines how it is encoded and interpreted by the graph neural network model. You can update the data type of existing column in the GNN table using the update_column_dtype method.

The following column dtypes are supported:

• ColumnDType('text') For columns containing free-form text. These will be encoded appropriately for NLP processing.

• ColumnDType('integer') For columns containing integer values.

• ColumnDType('float') For columns containing floating-point numbers.

• ColumnDType('category') For columns with discrete categorical values.

• ColumnDType('multi_categorical') For columns containing multiple categorical values, typically represented as a list (e.g., [‘car’, ‘red’]).

• ColumnDType('embedding') For columns containing fixed-size embedding vectors (e.g., outputs from external models). All embeddings in the column must have the same dimensionality.

• ColumnDType('datetime') For columns containing date or timestamp information. Time columns must always have a datetime dtype.