Dataset

Combines all tables and the task definition into a dataset ready for training.

Parameters

Name	Type	Description	Optional
`connector`	`SnowflakeConnector`	The connector object used for sending requests to the GNN reasoner.	No
`dataset_name`	`str`	A user-defined name for the dataset, which must comply with Snowflake object identifier rules.	No
`tables`	`list` of `GNNTable` instances	A collection of table objects that constitute the dataset.	No
`task_description`	`NodeTask` or `LinkTask`	The target task assigned to the GNN	No

Returns

An instance of the Dataset class.

Example

from relationalai_gnns import Dataset

dataset = Dataset(
    connector=connector,
    dataset_name="my_first_dataset",
    tables=[table_with_ckey_1, table_with_ckey_2, table_with_foreign_keys],
    task_description=node_task
)

Attributes

Name	Description	Type
`experiment_name`	A dataset is uniquely identified by its experiment name, which is automatically generated in the format `dataset_name_task_type_task_name`. The experiment name is also visible through the `JobMonitor`.	`str`
`metadata_dict`	A dictionary describing the dataset’s tables and task.	`dict`

Methods

Name	Description	Returns
`visualize_dataset`	Generates a visual representation of the dataset schema, encompassing its tables and the defined task.	`DatasetDiagram`
`print_data_config`	Prints in json format a dictionary describing the dataset’s tables and task.	`None`

.visualize_dataset()

Generates a visual representation of the dataset schema, encompassing its tables and the defined task.

Parameters

Name	Type	Description	Optional
`show_dtypes`	`bool`	Whether to show the data types of each column. Default is `False`.	Yes

Returns

A DatasetDiagram object. Call .display() to render it inline in a Jupyter notebook, .save("schema.png") to write it to a file, or access .source for the raw Graphviz DOT string.

Rendering requires the Graphviz dot command-line tool to be installed on your system (brew install graphviz on macOS, apt install graphviz on Linux).

# Auto-renders inline when returned from a Jupyter cell:
dataset.visualize_dataset()

# Or call display() explicitly:
dataset.visualize_dataset().display()

# Save to a file (format inferred from extension):
dataset.visualize_dataset().save("schema.svg")

import streamlit as st

diagram = dataset.visualize_dataset(show_dtypes=True)
st.image(diagram._render("png"))

.print_data_config()

Prints in json format a dictionary describing the dataset’s tables and task.

Example

dataset.print_data_config()