def test_sum_is_k(self):
hpc = HPC([49, 240, 1600, 480, 49],
0.67, 0.25, 0.04, # connection rates: (in_ec, ec_dg, dg_ca3)
0.10, 0.01, 0.04, # firing rates: (ec, dg, ca3)
0.7, 1, 0.1, 0.5, # gamma, epsilon, nu, turnover rate
0.10, 0.95, 0.8, 2.0) # k_m, k_r, a_i, alpha
I = np.asarray([[1, -1, 1, -1, 1, -1, 1] * 7], dtype=np.float32)
result = hpc.kWTA(I, 0.2)
k = np.round(I.shape[1] * 0.2)
self.assertEqual(sum(result[0]), k)
def test_equation_unconstrained_hebbian_learning_ca3_out(self):
hpc = HPC([49, 240, 1600, 480, 49],
0.67, 0.25, 0.04, # connection rates: (in_ec, ec_dg, dg_ca3)
0.10, 0.01, 0.04, # firing rates: (ec, dg, ca3)
0.7, 1, 0.1, 0.5, # gamma, epsilon, nu, turnover rate
0.10, 0.95, 0.8, 2.0) # k_m, k_r, a_i, alpha
empty_activation_values_l1 = np.zeros_like(hpc.ca3_values.get_value()).astype(np.float32)
empty_activation_values_l1.put([0, 0], 1)
hpc.set_ca3_values(empty_activation_values_l1)
empty_activation_values_l2 = np.zeros_like(hpc.output_values.get_value()).astype(np.float32)
empty_activation_values_l2.put([0, 0], 1)
hpc.set_output(empty_activation_values_l2)
current_weight_element = hpc.ca3_out_weights.get_value()[0][0]
next_weight_element = hpc._gamma * current_weight_element + 1
hpc.wire_ca3_out(hpc.ca3_values.get_value(return_internal_type=True),
hpc.output_values.get_value(return_internal_type=True),
hpc.ca3_out_weights.get_value(return_internal_type=True))
self.assertAlmostEqual(hpc.ca3_out_weights.get_value()[0][0], next_weight_element, places=6,
msg="Weight update did not correspond to the predicted update value of the equation: "
"next_weight_el != w_el : "+str(next_weight_element)+" != " +
str(hpc.ca3_out_weights.get_value()[0][0]))
io_dim = 49
turnover_rate = 0.30 # (Tools.get_parameter_counter() % 18) * 0.02 + 0.32
weighting_dg = 25 # Tools.get_experiment_counter() % 26
_ASYNC_FLAG = True
_TURNOVER_MODE = 1 # 0 for between every new set. 1 for every set iteration.
# print "TRIAL #", trial, "turnover rate:", turnover_rate
# dims,
# connection_rate_input_ec, perforant_path, mossy_fibers,
# firing_rate_ec, firing_rate_dg, firing_rate_ca3,
# _gamma, _epsilon, _nu, _turnover_rate, _k_m, _k_r, _a_i, _alpha):
hpc = HPC([io_dim, 240, 1600, 480, io_dim],
0.67, 0.25, 0.04, # connection rates: (in_ec, ec_dg, dg_ca3)
0.10, 0.01, 0.04, # firing rates: (ec, dg, ca3)
0.7, 100.0, 0.1, turnover_rate, # gamma, epsilon, nu, turnover rate
0.10, 0.95, 0.8, 2.0, weighting_dg, # k_m, k_r, a_i, alpha. alpha is 2 in 4.1
_ASYNC_FLAG=_ASYNC_FLAG, _TURNOVER_MODE=_TURNOVER_MODE)
# ============ Config. X: ============
for i in range(1):
for train_set_size_ctr in range(2, 3):
hpc.reset_hpc_module()
tar_patts = []
for p in training_patterns_associative[:5*train_set_size_ctr]:
tar_patts.append(p[1])
ann = NeocorticalNetwork(io_dim, 30, io_dim, 0.01, 0.9)
print "Starting experiment; HPC chaotic recall i iterations and HPC pseudopatterns..."
def test_equation_constrained_hebbian_learning_ec_ca3(self):
hpc = HPC([49, 240, 1600, 480, 49],
0.67, 0.25, 0.04, # connection rates: (in_ec, ec_dg, dg_ca3)
0.10, 0.01, 0.04, # firing rates: (ec, dg, ca3)
0.7, 1, 0.1, 0.5, # gamma, epsilon, nu, turnover rate
0.10, 0.95, 0.8, 2.0) # k_m, k_r, a_i, alpha
activation_values_l1 = np.zeros_like(hpc.ec_values.get_value()).astype(np.float32)
activation_values_l1.put([0, 0], 1)
activation_values_l1.put([0, 1], 1)
hpc.set_ec_values(activation_values_l1)
activation_values_l2 = np.zeros_like(hpc.ca3_values.get_value()).astype(np.float32)
activation_values_l2.put([0, 0], 1)
hpc.set_ca3_values(activation_values_l2)
current_weight_element_0 = hpc.ec_ca3_weights.get_value()[0][0]
next_weight_element_0 = current_weight_element_0 + hpc._nu * (1 - current_weight_element_0)
current_weight_element_1 = hpc.ec_ca3_weights.get_value()[1][0]
next_weight_element_1 = current_weight_element_1 + hpc._nu * (1 - current_weight_element_1)
# wire EC to CA3
n_rows = hpc.ec_values.get_value(return_internal_type=True).shape[1]
n_cols = hpc.ca3_values.get_value(return_internal_type=True).shape[1]
u_next_for_elemwise_ops = [hpc.ca3_values.get_value(return_internal_type=True)[0]] * n_rows
u_prev_for_elemwise_ops_transposed = [hpc.ec_values.get_value(return_internal_type=True)[0]] * n_cols
hpc.wire_ec_to_ca3(u_prev_for_elemwise_ops_transposed, u_next_for_elemwise_ops,
hpc.ec_ca3_weights.get_value(return_internal_type=True))
self.assertAlmostEqual(hpc.ec_ca3_weights.get_value()[0][0], next_weight_element_0, places=6,
msg="Weight update did not correspond to the predicted update value of the equation: "
"next_weight_el_0 != w_el : "+str(next_weight_element_0)+" != " +
str(hpc.ec_ca3_weights.get_value()[0][0]))
self.assertAlmostEqual(hpc.ec_ca3_weights.get_value()[1][0], next_weight_element_1, places=6,
msg="Weight update did not correspond to the predicted update value of the equation: "
"next_weight_el_0 != w_el : "+str(next_weight_element_1)+" != " +
str(hpc.ec_ca3_weights.get_value()[1][0]))
# last element in weight matrix:
weight_rows = activation_values_l1.shape[1]
weight_columns = activation_values_l2.shape[1]
# self.assertEqual(weight_columns, 480, msg="weight columns between ec-ca3 wasn't 480")
# self.assertEqual(weight_rows, 240, msg="weight rows between ec-ca3 wasn't 240")
activation_values_l1.put([0, weight_rows-1], 1) # set last element of activation values to 1
activation_values_l2.put([0, weight_columns-1], 1) # set last element of activation values to 1
hpc.set_ec_values(activation_values_l1)
hpc.set_ca3_values(activation_values_l2)
current_weight_element = hpc.ec_ca3_weights.get_value()[weight_rows-1][weight_columns-1]
next_weight_element = current_weight_element + hpc._nu * (1 - current_weight_element)
# wire EC to CA3
n_rows = hpc.ec_values.get_value(return_internal_type=True).shape[1]
n_cols = hpc.ca3_values.get_value(return_internal_type=True).shape[1]
u_next_for_elemwise_ops = [hpc.ca3_values.get_value(return_internal_type=True)[0]] * n_rows
u_prev_for_elemwise_ops_transposed = [hpc.ec_values.get_value(return_internal_type=True)[0]] * n_cols
hpc.wire_ec_to_ca3(u_prev_for_elemwise_ops_transposed, u_next_for_elemwise_ops,
hpc.ec_ca3_weights.get_value(return_internal_type=True))
self.assertAlmostEqual(hpc.ec_ca3_weights.get_value()[weight_rows-1][weight_columns-1], next_weight_element,
places=6, msg="Weight update did not correspond to the predicted update value of the "
"equation: next_weight_el != w_el : "+str(next_weight_element)+" != " +
str(hpc.ec_ca3_weights.get_value()[weight_rows-1][weight_columns-1]))
io_dim = 49
turnover_rate = 0.30 # (Tools.get_parameter_counter() % 18) * 0.02 + 0.32
weighting_dg = 25 # Tools.get_experiment_counter() % 26
_ASYNC_FLAG = True
_TURNOVER_MODE = 1 # 0 for between every new set. 1 for every set iteration.
# print "TRIAL #", trial, "turnover rate:", turnover_rate
# dims,
# connection_rate_input_ec, perforant_path, mossy_fibers,
# firing_rate_ec, firing_rate_dg, firing_rate_ca3,
# _gamma, _epsilon, _nu, _turnover_rate, _k_m, _k_r, _a_i, _alpha):
hpc = HPC([io_dim, 240, 1600, 480, io_dim],
0.67, 0.25, 0.04, # connection rates: (in_ec, ec_dg, dg_ca3)
0.10, 0.01, 0.04, # firing rates: (ec, dg, ca3)
0.7, 100.0, 0.1, turnover_rate, # gamma, epsilon, nu, turnover rate
0.10, 0.95, 0.8, 2.0, weighting_dg, # k_m, k_r, a_i, alpha. alpha is 2 in 4.1
_ASYNC_FLAG=_ASYNC_FLAG, _TURNOVER_MODE=_TURNOVER_MODE)
# ============ Config. 1: ============
for i in range(6):
for train_set_size_ctr in range(2, 6):
hpc.reset_hpc_module()
tar_patts = []
for p in training_patterns_associative[:5*train_set_size_ctr]:
tar_patts.append(p[1])
print "Starting experiment 4.1, HPC chaotic recall i iters and HPC pseudopatterns..."
# This also saves the experiment results:
# relative frequency as in successful 2x5 goodness of fit.
