Bering.tools.node_classification

Bering.tools.node_classification(bg, df_spots, n_neighbors=10, beta=1.0, prob_threshold=0.3, max_num_spots=1500000, num_chunks=25)[source]

Node classification for all spots in the slice

Parameters:

  • bg (BrGraph) – Bering Graph object

  • df_spots (pd.DataFrame) – spots table. It can be bg.spots_all in case of whole slice prediction

  • n_neighbors (int) – number of neighbors for graph construction

  • prob_threshold (float) – minimal threshold of predicted probability for spots to be considered as foreground

  • max_num_spots (int) – maximum number of spots for node classification in each chunk. If the number of spots is larger than this number, the spots table will be split into chunks by coordinates.

  • num_chunks (int) – number of chunks for node classification. This is done by splitting the spots table into chunks by coordinates. This is used when the number of spots is too large. Refer to _get_node_embedding_prediction_byTiling for details.

Returns:

: preds_labels: np.array

predicted labels for all spots

graph_all: torch_geometric.data.Data

graph (torch_geometric.data.Data object) for the whole slice