def make_action(locations, effects, keys, location_indices, effect_indices):
location_sum = np.sum(locations.vectors[:, location_indices], axis=1)
effect_sum = np.sum(effects.vectors[:, effect_indices], axis=1)
vector = hrr.bind(keys.get('LOCATION'), location_sum,
do_normalize=False) + hrr.bind(
keys.get('EFFECT'), effect_sum, do_normalize=False)
return vector
def make_thing(locations, goals, keys, location_indices, goal_indices):
# make an object HRR vector with these associated locations and goals
location_sum = np.sum(locations.vectors[:, location_indices], axis=1)
goal_sum = np.sum(goals.vectors[:, goal_indices], axis=1)
vector = hrr.bind(keys.get('LOCATION'), location_sum,
do_normalize=False) + hrr.bind(
keys.get('GOAL'), goal_sum, do_normalize=False)
return vector
def make_affordance(things, effects, keys, thing_index, effect_indices):
#normalize each thing vector for use as affordance component
v = things.vectors[:,thing_index].copy()
for i in range(v.shape[1]):
v[:,i] = hrr.normalize(v[:,i])
# thing_sum = np.sum(things.vectors[:,thing_index], axis=1)
thing_sum = np.sum(v, axis=1)
effect_sum = np.sum(effects.vectors[:,effect_indices], axis=1)
vector = hrr.bind(keys.get('THING'), thing_sum, do_normalize=False) + hrr.bind(keys.get('EFFECT'), effect_sum, do_normalize=False)
return vector
def make_affordance(things, effects, keys, thing_index, effect_indices):
#normalize each thing vector for use as affordance component
v = things.vectors[:, thing_index].copy()
for i in range(v.shape[1]):
v[:, i] = hrr.normalize(v[:, i])
# thing_sum = np.sum(things.vectors[:,thing_index], axis=1)
thing_sum = np.sum(v, axis=1)
effect_sum = np.sum(effects.vectors[:, effect_indices], axis=1)
vector = hrr.bind(keys.get('THING'), thing_sum,
do_normalize=False) + hrr.bind(
keys.get('EFFECT'), effect_sum, do_normalize=False)
return vector
def make_random_cleanup(n, prefix, keys, labels, cleanups, counts, D):
"""
n - number of concepts in cleanup
prefix - prefix for concept names (to be appended with a number)
keys - cleanup of data types (LOCATION, GOAL, etc.)
cleanups - tuple of cleanup memories from which parts of the created concepts will be drawn
labels - tuple of labels (e.g. 'LOCATION') of parts
counts - tuple of counts (each length n) of summed components from each cleanup
"""
assert len(labels) == len(cleanups)
assert len(labels) == len(counts)
cleanup = hrr.CleanupMemory([prefix + str(l) for l in range(n)], D)
indices = []
for i in range(n):
indices_i = []
vector = np.zeros(D)
for j in range(len(labels)):
indices_ij = np.random.randint(0,
len(cleanups[j].concepts),
size=counts[j][i])
s = np.sum(cleanups[j].vectors[:, indices_ij], axis=1)
vector = vector + hrr.bind(
keys.get(labels[j]), s, do_normalize=False)
indices_i.append(indices_ij)
cleanup.replace(cleanup.concepts[i], vector)
indices.append(indices_i)
cleanup.indices = indices
return cleanup
def make_random_cleanup(n, prefix, keys, labels, cleanups, counts, D):
"""
n - number of concepts in cleanup
prefix - prefix for concept names (to be appended with a number)
keys - cleanup of data types (LOCATION, GOAL, etc.)
cleanups - tuple of cleanup memories from which parts of the created concepts will be drawn
labels - tuple of labels (e.g. 'LOCATION') of parts
counts - tuple of counts (each length n) of summed components from each cleanup
"""
assert len(labels) == len(cleanups)
assert len(labels) == len(counts)
cleanup = hrr.CleanupMemory([prefix + str(l) for l in range(n)], D)
indices = []
for i in range(n):
indices_i = []
vector = np.zeros(D)
for j in range(len(labels)):
indices_ij = np.random.randint(0, len(cleanups[j].concepts), size=counts[j][i])
s = np.sum(cleanups[j].vectors[:,indices_ij], axis=1)
vector = vector + hrr.bind(keys.get(labels[j]), s, do_normalize=False)
indices_i.append(indices_ij)
cleanup.replace(cleanup.concepts[i], vector)
indices.append(indices_i)
cleanup.indices = indices
return cleanup
def make_action(locations, effects, keys, location_indices, effect_indices):
location_sum = np.sum(locations.vectors[:,location_indices], axis=1)
effect_sum = np.sum(effects.vectors[:,effect_indices], axis=1)
vector = hrr.bind(keys.get('LOCATION'), location_sum, do_normalize=False) + hrr.bind(keys.get('EFFECT'), effect_sum, do_normalize=False)
return vector
def make_thing(locations, goals, keys, location_indices, goal_indices):
# make an object HRR vector with these associated locations and goals
location_sum = np.sum(locations.vectors[:,location_indices], axis=1)
goal_sum = np.sum(goals.vectors[:,goal_indices], axis=1)
vector = hrr.bind(keys.get('LOCATION'), location_sum, do_normalize=False) + hrr.bind(keys.get('GOAL'), goal_sum, do_normalize=False)
return vector
