def test_neuron_update_fwd(self):
tf.reset_default_graph()
tf.disable_v2_behavior()
n_in, n_out = 10, 5
inp, out, w = random_dense(n_in, n_out)
env = blur_env.tf_env
pre, post, synapse, genome, network_spec = get_blur_state(env, inp, out, w)
pre_fwd, post_fwd = blur.dense_neuron_update(
pre,
post,
synapse,
inp_act=None,
out_act=sigmoid_with_grad,
neuron_genome=genome.neuron,
update_type=synapse_util.UpdateType.FORWARD,
global_spec=network_spec,
env=env)
inp = tf.constant(inp.astype(blur_env.NP_FLOATING_TYPE))
ww = w.astype(blur_env.NP_FLOATING_TYPE)
inp_with_bias = tf.concat([inp[Ellipsis, 0], [[[[1]]]]], axis=-1)
exp_results = inp_with_bias @ ww
with tf.Session() as s:
s.run(tf.initialize_all_variables())
self.assertAllClose(pre_fwd, pre)
self.assertAllClose(post_fwd[Ellipsis, 0], tf.math.sigmoid(exp_results))
self.assertAllClose(post_fwd[Ellipsis, 1],
d_sigmoid(out[Ellipsis, 0] + exp_results))
def test_neuron_update_bwd(self):
tf.reset_default_graph()
tf.disable_v2_behavior()
n_in, n_out = 10, 5
inp, out, w = random_dense(n_in, n_out)
env = blur_env.tf_env
pre, post, synapse, genome, network_spec = get_blur_state(
env, inp, out, w)
inp_act_fn = lambda x: x
out_act_fn = sigmoid_with_grad
pre_fwd, post_fwd = blur.dense_neuron_update(
pre,
post,
synapse,
inp_act=None,
out_act=out_act_fn,
neuron_genome=genome.neuron,
update_type=synapse_util.UpdateType.FORWARD,
global_spec=network_spec,
env=env)
pre_bkw, _ = blur.dense_neuron_update(
pre_fwd,
post_fwd,
synapse=synapse,
inp_act=inp_act_fn,
out_act=out_act_fn,
neuron_genome=genome.neuron,
update_type=synapse_util.UpdateType.BACKWARD,
global_spec=network_spec,
env=env)
inp = tf.constant(inp.astype(blur_env.NP_FLOATING_TYPE))
ww = w.astype(blur_env.NP_FLOATING_TYPE)
exp_result = post_fwd[Ellipsis, 1] @ tf.transpose(ww, (0, 1, 3, 2))
with tf.Session() as s:
s.run(tf.initialize_all_variables())
self.assertAllClose(pre_bkw[Ellipsis, 0], pre_fwd[Ellipsis, 0])
self.assertAllClose(
pre_bkw[Ellipsis, 1],
exp_result[Ellipsis, :-1] * pre_fwd[Ellipsis, 1])
def test_synapse_derivative(self):
tf.reset_default_graph()
tf.disable_v2_behavior()
n_in, n_out = 10, 5
inp, out, w = random_dense(n_in, n_out)
env = blur_env.tf_env
pre, post, synapse, genome, network_spec = get_blur_state(
env, inp, out, w)
post = np.concatenate(2 * [np.zeros_like(out)],
axis=-1).astype(blur_env.NP_FLOATING_TYPE)
pre_fwd, post_fwd = blur.dense_neuron_update(
pre,
post,
synapse,
inp_act=None,
out_act=sigmoid_with_grad,
neuron_genome=genome.neuron,
update_type=synapse_util.UpdateType.FORWARD,
global_spec=network_spec,
env=env)
hebbian_update = blur.get_hebbian_update(pre_fwd, post_fwd,
genome.synapse.transform,
network_spec, env)
hebbian_update_submatrix = synapse_util.synapse_submatrix(
hebbian_update, n_in, synapse_util.UpdateType.FORWARD)
inp = tf.constant(inp.astype(blur_env.NP_FLOATING_TYPE))
inp_with_bias = tf.concat([inp[Ellipsis, 0], [[[[1]]]]], axis=-1)
ww = tf.constant(w.astype(blur_env.NP_FLOATING_TYPE))
out = tf.nn.sigmoid(inp_with_bias @ ww)
grad_w = tf.gradients(out, ww)
with tf.Session() as s:
s.run(tf.initialize_all_variables())
hebb_update, grad_w_val = s.run(
[hebbian_update_submatrix[Ellipsis, 0], grad_w[0]])
self.assertAllClose(hebb_update, grad_w_val)