61 lines
2.2 KiB
Python
61 lines
2.2 KiB
Python
|
import numpy as np
|
||
|
|
|
||
|
|
|
||
|
|
class Walker:
|
||
|
|
""" The Walker knows how to walk at random on a context map. """
|
||
|
|
|
||
|
|
def __init__(self, sigma_i, sigma_j, context_map):
|
||
|
|
self.sigma_i = sigma_i
|
||
|
|
self.sigma_j = sigma_j
|
||
|
|
self.size = context_map.shape[0]
|
||
|
|
# Make sure that the context map is normalized
|
||
|
|
context_map /= context_map.sum()
|
||
|
|
self.context_map = context_map
|
||
|
|
|
||
|
|
# Pre-compute a 2D grid of coordinates for efficiency
|
||
|
|
self._grid_ii, self._grid_jj = np.mgrid[0:self.size, 0:self.size]
|
||
|
|
|
||
|
|
# --- Walker public interface
|
||
|
|
|
||
|
|
def sample_next_step(self, current_i, current_j, random_state=np.random):
|
||
|
|
""" Sample a new position for the walker. """
|
||
|
|
|
||
|
|
# Combine the next-step proposal with the context map to get a
|
||
|
|
# next-step probability map
|
||
|
|
next_step_map = self._next_step_proposal(current_i, current_j)
|
||
|
|
selection_map = self._compute_next_step_probability(next_step_map)
|
||
|
|
|
||
|
|
# Draw a new position from the next-step probability map
|
||
|
|
r = random_state.rand()
|
||
|
|
cumulative_map = np.cumsum(selection_map)
|
||
|
|
cumulative_map = cumulative_map.reshape(selection_map.shape)
|
||
|
|
i_next, j_next = np.argwhere(cumulative_map >= r)[0]
|
||
|
|
|
||
|
|
return i_next, j_next
|
||
|
|
|
||
|
|
# --- Walker non-public interface
|
||
|
|
|
||
|
|
def _next_step_proposal(self, current_i, current_j):
|
||
|
|
""" Create the 2D proposal map for the next step of the walker. """
|
||
|
|
|
||
|
|
# 2D Gaussian distribution , centered at current position,
|
||
|
|
# and with different standard deviations for i and j
|
||
|
|
grid_ii, grid_jj = self._grid_ii, self._grid_jj
|
||
|
|
sigma_i, sigma_j = self.sigma_i, self.sigma_j
|
||
|
|
|
||
|
|
rad = (
|
||
|
|
(((grid_ii - current_i) ** 2) / (sigma_i ** 2))
|
||
|
|
+ (((grid_jj - current_j) ** 2) / (sigma_j ** 2))
|
||
|
|
)
|
||
|
|
|
||
|
|
p_next_step = np.exp(-(rad / 2.0)) / (2.0 * np.pi * sigma_i * sigma_j)
|
||
|
|
return p_next_step / p_next_step.sum()
|
||
|
|
|
||
|
|
def _compute_next_step_probability(self, next_step_map):
|
||
|
|
""" Compute the next step probability map from next step proposal and
|
||
|
|
context map. """
|
||
|
|
next_step_probability = next_step_map * self.context_map
|
||
|
|
next_step_probability /= next_step_probability.sum()
|
||
|
|
return next_step_probability
|
||
|
|
|