lazyslide.tl.feature_prediction

feature_prediction(wsi, model, feature_key=None, *, batch_size=1024, tile_key='tiles', key_added=None, amp=None, autocast_dtype=None, device=None, pbar=None)

Predict tile-level values from an existing feature matrix.

Parameters:

wsiWSIData

The whole-slide image object containing tile features.

modelstr or feature prediction model

A registered feature prediction model name or an object implementing predict(features).

feature_keystr, optional

Feature table used as model input. When omitted, this is inferred from model.features_model_name.

batch_sizeint, default: 1024

Number of tile feature vectors passed to the model at once.

tile_keystr, default: “tiles”

Tile table associated with the input and output features.

key_addedstr, optional

Key used to store the prediction table. Defaults to {model_name}_{tile_key}.

ampbool, optional

Whether to use automatic mixed precision.

autocast_dtypetorch.dtype, optional

Data type used for automatic mixed precision.

devicestr, optional

Device on which to run inference.

pbarbool, optional

Whether to display a progress bar.

Returns:

None

Notes

Dense input matrices are passed to the model as basic row slices. NumPy therefore supplies views rather than copying the full feature vectors.