def check_against_input(cls, module, args, inp_idx, grad, out_idx, y):
print('\t\tChecking input #', inp_idx)
x = args[inp_idx]
relates = list()
assert x.shape == grad.shape, \
'Gradient shape mismatch: {} and {}'.format(x.shape, grad.shape)
for each in range(cls.check_round):
pick = tuple(randint(dim) for dim in x.shape)
x1, x2 = x.copy(), x.copy()
x1[pick] -= cls.epsilon
x2[pick] += cls.epsilon
args[inp_idx] = x1
y1 = cls.tolist(module.forward(*args))[out_idx]
L1 = (y1 - y) * (y1 - y)
args[inp_idx] = x2
y2 = cls.tolist(module.forward(*args))[out_idx]
L2 = (y2 - y) * (y2 - y)
grad_pick = (L2.sum() - L1.sum())
grad_pick /= (2. * cls.epsilon)
if grad_pick == grad[pick]:
print('\t\t', grad_pick, grad[pick])
continue
relate = abs(grad_pick - grad[pick])
relate /= max(abs(grad_pick), abs(grad[pick]))
print('\t\t', grad_pick, grad[pick], relate)
relates.append(relate)
if len(relates) == 0: return
relate = np.mean(relates)
def test_create_ants(set_up_tree_nests_fixed):
"""Tests whether right amount of ants with type ant are created at nest positions.
set_up_tree_nests_fixed should not build two nests at one position."""
tree, positions = set_up_tree_nests_fixed
amount_ants = randint(1, 6)
for pos in positions:
nest = tree.get_at_position(pos)[0]
tree.create_ants(nest, amount_ants)
for pos in positions:
objects_at_pos = tree.get_at_position(pos)
assert len(objects_at_pos) == amount_ants + 1
assert sum([isinstance(obj, Ant) for obj in objects_at_pos]) == amount_ants
def move_randomly(self):
"""
changes the position of the ant using a random walk and combining it with the previous direction
direction is updated
:return: the updated position is returned
"""
while True: # to avoid standing still and divide by zero
movement = randint(low=-1, high=2, size=2) # random move
self.direction += self.direction_memory * self.direction + movement
if distance(self.direction) > 0.:
break
self.direction /= distance(self.direction)
self.position = self.position + self.direction
return self.position
def move_randomly(self):
movement = randint(low=-1, high=2, size=2) # random move
self.momentum += 0.5 * self.momentum + movement
self.momentum /= np.linalg.norm(self.momentum)
self.position = self.position + self.momentum
return self.position