GroupObservation

experimental.network_dynamics.GroupObservation(
    readouts,
    *,
    params=None,
    channels,
    fill_value=0.0,
    allow_partial_coverage=False,
)

Project group variables of interest into common graph-order channels.

Each mapping value is a variable-of-interest name, a tuple of names, or a readout(voi, params) -> [Q, n_group_nodes] callable. Equal channel width only establishes shape compatibility; users remain responsible for making group-specific transformations scientifically commensurable.

Partial coverage is allowed for ordinary projection. With reduce= it is rejected unless allow_partial_coverage=True. That opt-in is only for a reducer the user has verified is fill-aware: it does not prevent NaN rows in covariance reducers or stop BOLD reducers treating fill as real drive.

# GroupObservation { #tvboptim.experimental.network_dynamics.GroupObservation }

```python
experimental.network_dynamics.GroupObservation(
    readouts,
    *,
    params=None,
    channels,
    fill_value=0.0,
    allow_partial_coverage=False,
)
```

Project group variables of interest into common graph-order channels.

Each mapping value is a variable-of-interest name, a tuple of names, or a
``readout(voi, params) -> [Q, n_group_nodes]`` callable. Equal channel
width only establishes shape compatibility; users remain responsible for
making group-specific transformations scientifically commensurable.

Partial coverage is allowed for ordinary projection. With ``reduce=`` it is
rejected unless ``allow_partial_coverage=True``. That opt-in is only for a
reducer the user has verified is fill-aware: it does not prevent NaN rows
in covariance reducers or stop BOLD reducers treating fill as real drive.