spateo.segmentation.bp#
Belief propagation to compute the marginal probabilities of a spot being occupied by a cell.
Module Contents#
Functions#
|
Helper function to convert a neighborhood mask to coordinate offsets. |
|
Compute the marginal probablity of each pixel being a cell, as opposed |
|
Compute the marginal probability of each pixel being a cell, using |
- spateo.segmentation.bp.create_neighbor_offsets(neighborhood: numpy.ndarray) numpy.ndarray[source]#
Helper function to convert a neighborhood mask to coordinate offsets.
- Parameters:
- neighborhood
Boolean mask (kernel) indicating the neighborhood to consider.
- Returns:
A Numpy array containing offsets for each dimension. Each row is the offsets required to reach a single neighbor.
- Raises:
ValueError – If neighborhood does not have odd size dimensions.
- spateo.segmentation.bp.cell_marginals(background_probs: numpy.ndarray, cell_probs: numpy.ndarray, neighborhood: numpy.ndarray | None = None, p: float = 0.6, q: float = 0.4, precision: float = 1e-05, max_iter: int = 100) numpy.ndarray[source]#
Compute the marginal probablity of each pixel being a cell, as opposed to background. This function calls a fast belief propagation library optimized for grid Markov Random Fields of arbitray dimension.
- Parameters:
- background_probs
The probability of each pixel being background (for instance, computed by taking the PDF of the parameters estmiated by EM).
- cell_probs
The probability of each pixel being a cell (for instance, computed by taking the PDF of the parameters estmiated by EM).
- neighborhood
A mask (kernel) indicating the neighborhood of each node to consider. The node is at the center of this array. Defaults to immediate neighbors (no diagonals).
- p
The potential indicating how likely two adjacent cells will be the same state. Does not necessarily have to be a probability.
- p
The potential indicating how likely two adjacent cells will be different states. Does not necessarily have to be a probability.
- precision
Stop iterations when desired precision is reached, as computed by the L2-norm of the messages from two consecutive iterations.
- max_iter
Maximum number of iterations.
- Returns:
The marginal probability, at each pixel, of the pixel being a cell.
- spateo.segmentation.bp.run_bp(background_cond: numpy.ndarray, cell_cond: numpy.ndarray, k: int = 3, square: bool = False, p: float = 0.6, q: float = 0.4, precision: float = 1e-06, max_iter: int = 100) numpy.ndarray[source]#
Compute the marginal probability of each pixel being a cell, using belief propagation.
- Parameters:
- background_cond
Probability of observing UMIs conditioned on background.
- cell_cond
Probability of observing UMIs conditioned on cell.
- k
Neighborhood size
- square
Whether the neighborhood of each node is a square around it. If false, the neighborhood is a circle.
- p
The potential indicating how likely two adjacent cells will be the same state. Does not necessarily have to be a probability.
- p
The potential indicating how likely two adjacent cells will be different states. Does not necessarily have to be a probability.
- precision
Stop iterations when desired precision is reached, as computed by the L2-norm of the messages from two consecutive iterations.
- max_iter
Maximum number of iterations.
- Returns:
Numpy array of marginal probabilities.