[FEATURE] Allow creating an ExtendedGrid directly from an existing base Grid

[juliettegiraud-2004]

Hello everyone!

Today, PowerGridModelInterface.create_grid_from_input_data() creates a base Grid from PGM input data (standard conversion step from the Power Grid Model engine input into a PGM‑DS Grid).

In many workflows, we want to work with an extended grid (Extended_Grid(Grid)) to add specific attributes (for example an extended LinkArray with a field like link_type) while keeping the rest of the grid unchanged. The documentation shows that PGM‑DS is explicitly designed to support grid extensions by subclassing arrays and the Grid dataclass.

However, from what I’ve understood, starting from a base Grid produced by create_grid_from_input_data() and converting it into an Extended_Grid is currently not implemented, and thus requires manually reconstructing / appending many arrays one-by-one. Having a direct base→extended conversion would avoid manual reconstruction and makes these workflows much easier to implement correctly.

What would be great is a built-in way to take the base Grid created from PGM input data and then promote/extend it into an Extended_Grid by only providing the extended parts, e.g. only node/line/link replaced by Node_Extended_Array, Line_Extended_Array, Link_Extended_Array, while all other component arrays (transformers, loads, sensors, etc.) are reused as-is.

Maybe it exists such functions but I haven't found them, so if you have more informations it would be great!

Thank you

[juliettegiraud-2004]

Just to clarify the feature request more precisely, what I’m looking for is a built-in conversion function from a base Grid to an ExtendedGrid, where I only have to provide the extended arrays and the library reuses all other arrays from the base grid automatically.

[TomvdPoll]

I'm not sure whether it does exactly what you want, but the Grid.from_extended classmethod might also work in reverse. You could try calling ExtendedGrid.from_extended(Grid) and see what happens? That way the extra arrays in the extended grid would be defaulted to empty, but you could fill them afterwards

[jaapschoutenalliander]

There is a supported way to load PGM input data directly into an ExtendedGrid that should fit your needs. By defining your ExtendedGrid upfront and passing it to PowerGridModelInterface, extra columns are automatically filled with defaults.

Step 1 – Define your extended arrays with defaults for any extra fields:

from numpy.typing import NDArray
import numpy as np
from power_grid_model_ds.arrays import NodeArray, LineArray, LinkArray

class ExtendedNodeArray(NodeArray):
    _defaults = {"node_type": 0}  # provide a default so PGM input data (which lacks this field) is accepted
    node_type: NDArray[np.int8]

class ExtendedLinkArray(LinkArray):
    _defaults = {"link_type": 0}
    link_type: NDArray[np.int32]

⚠️ The _defaults dict is important: because PGM input data won't contain your custom fields, a default value is required for every extra column. Without it, create_grid_from_input_data() will raise a ValueError: Missing required columns.

Step 2 – Define your ExtendedGrid dataclass using those arrays:

from dataclasses import dataclass
from power_grid_model_ds import Grid
from my_extended_arrays import ExtendedNodeArray, ExtendedLinkArray

@dataclass
class ExtendedGrid(Grid):
    node: ExtendedNodeArray
    link: ExtendedLinkArray
    # all other components (line, transformer, sym_load, etc.) are inherited from Grid as-is

Step 3 – Load PGM input data directly into it:

from power_grid_model_ds._core.power_grid_model_interface import PowerGridModelInterface
from my_extended_grid import ExtendedGrid

grid = ExtendedGrid.empty()

interface = PowerGridModelInterface(grid=grid, input_data=input_data_pgm)
grid = interface.create_grid_from_input_data()

# grid is already an ExtendedGrid — no manual reconstruction needed
assert isinstance(grid, ExtendedGrid)
assert isinstance(grid.node, ExtendedNodeArray)
assert isinstance(grid.link, ExtendedLinkArray)

All components you didn't override (transformers, loads, sensors, etc.) are populated from the base Grid definition unchanged — there's no need to reconstruct or copy them manually.

The relevant test demonstrating this is test_create_extended_grid_with_default_from_input_data

[nitbharambe]

To add to @jaapschoutenalliander's answer:
There are certain attributes which are not available in the default grid object. Like for eg. sk, rx_ratio for Source component. If the input_data contains them, then i imagine you also have to specially add these in extended array definition as above. Note that the type of that attribute should match with PGM. You can find PGM's attribtue type via power_grid_model's attribute_dtype function.
Check array definition docs for all attributes available in Grid