def recall(self):
# Fire EC to CA3, CA3 to CA3
if(self._ASYNC_FLAG):
self.async_ca3_wrapper(self.ec_values.get_value(return_internal_type=True),
self.ec_ca3_weights.get_value(return_internal_type=True),
np.zeros_like(self.dg_values.get_value(), dtype=np.float32), # same as not including
np.zeros_like(self.dg_ca3_weights.get_value(), dtype=np.float32),
self.ca3_values.get_value(return_internal_type=True),
self.ca3_ca3_weights.get_value(return_internal_type=True),
self.eta_ca3.get_value(return_internal_type=True),
self.zeta_ca3.get_value(return_internal_type=True))
else:
self.fire_to_ca3_no_learning(self.ec_values.get_value(return_internal_type=True),
self.ec_ca3_weights.get_value(return_internal_type=True),
self.ca3_values.get_value(return_internal_type=True),
self.ca3_ca3_weights.get_value(return_internal_type=True),
self.eta_ca3.get_value(return_internal_type=True),
self.zeta_ca3.get_value(return_internal_type=True))
# kWTA CA3
self.set_ca3_values(
self.kWTA(self.ca3_values.get_value(return_internal_type=True), self.firing_rate_ca3)) # in-memory
# fire CA3 to output
self.fire_ca3_out(self.ca3_values.get_value(return_internal_type=True),
self.ca3_out_weights.get_value(return_internal_type=True))
# Bipolar output:
self.set_output(Tools.get_bipolar_in_out_values(self.output_values.get_value(return_internal_type=True)))
def evaluate_ann_with_bipolar_output(ann, set_size):
print "Evaluating the ANN-object.."
sum_corr = 0.
corr_ctr = 0.
neocortically_recalled_pairs = []
for [target_in, target_out] in training_patterns_associative[:5*set_size]:
obtained_in, obtained_out = ann.get_IO(target_in)
sum_corr += Tools.get_pattern_correlation(target_out, Tools.get_bipolar_in_out_values(obtained_out))
corr_ctr += 1
neocortically_recalled_pairs.append([obtained_in, obtained_out])
g = sum_corr / corr_ctr
goodness_str = "goodness of fit, g=" + "{:6.4f}".format(g)
print goodness_str
