def test_pickle(self):
params = base_config.get_config()
params['global']['epochs'] = 2
network = SRNetwork(params['network'])
constant = np.array([[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0
],
[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0
],
[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0
],
[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0
],
[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0
],
[
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0
]])
inputs = constant
for inp in inputs:
output = network.run(inp)
weights = network.layers[0].feedforward_node.weights
with open('test_network_serialized.pickle', 'wb') as f:
pickle.dump(network, f)
# network.serialize('test_network_serialized.pickle')
with open('test_network_serialized.pickle', "rb") as f:
x = pickle.load(f)
network_loaded = x
print network_loaded
self.assertTrue(
(weights[0] == network_loaded.layers[0].feedforward_node.weights[0]
).all())
def test_pickle(self):
params = base_config.get_config()
params['global']['epochs'] = 2
network = SRNetwork(params['network'])
constant = np.array([
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0]
])
inputs = constant
for inp in inputs:
output = network.run(inp)
weights = network.layers[0].feedforward_node.weights
with open('test_network_serialized.pickle', 'wb') as f:
pickle.dump(network, f)
# network.serialize('test_network_serialized.pickle')
with open('test_network_serialized.pickle', "rb") as f:
x = pickle.load(f)
network_loaded = x
print network_loaded
self.assertTrue((weights[0] == network_loaded.layers[0].feedforward_node.weights[0]).all())
def test_init(self):
logging.basicConfig(level=logging.INFO)
config = base_config.get_config()
# network = SRNetwork(config['network'])
network = SymmetricNetwork(config['network'])
long_sequence = np.array([
[0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0]
])
constant = np.array([
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0],
[0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0]
])
simple_sequence = np.array([
[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
])
input_series = simple_sequence #simple_sequence #long_sequence #constant
sd_input_series = preprocessing.scale(input_series)
feedforward_errors = {layer['name']: [] for layer in config['network']['layers']}
recurrent_errors = {layer['name']: [] for layer in config['network']['layers']}
feedback_errors = {layer['name']: [] for layer in config['network']['layers']}
output_mse = []
prev_output = None
for ep in range(config['global']['epochs']):
for ind, inp in enumerate(input_series):
output, mse = network.run(inp)
output_mse.append(mse)
if prev_output is not None:
prev_output_bin = np.zeros(prev_output.size).astype('int')
for i in range(0, prev_output.size):
if prev_output[i] > 0.5:
prev_output_bin[i] = 1
else:
prev_output_bin[i] = 0
mod_input = np.zeros(inp.size).astype('int')
for i in range(0, inp.size):
if inp[i] >= 0.5:
mod_input[i] = 1
print '############### epoch: {0}\n' \
'############### sequence: {1}\n' \
'############### input: \n' \
'{2}\n' \
'############### prev_output_bin: \n' \
'{3}\n' \
'############### prev_output: \n' \
'{4}\n' \
'output_error: {5}'.format(ep, ind, mod_input, prev_output_bin, prev_output, mse)
plt.ion()
plt.axis([-1, 4, -1, 4])
x_r, y_r = np.argwhere(mod_input.reshape(4, 4) == 1).T
x_t, y_t = np.argwhere(prev_output_bin.reshape(4, 4) == 1).T
plt.scatter(x_r, y_r, alpha=0.5, c='r', marker='s', s=150)
plt.scatter(x_t, y_t, alpha=0.5, c='b', marker='o', s=130)
plt.draw()
time.sleep(0.1)
plt.clf()
prev_output = output
for key, v in network.feedforward_errors.items():
if len(v) > 0:
feedforward_errors[key].append(v[0])
for key, v in network.recurrent_errors.items():
if len(v) > 0:
recurrent_errors[key].append(v[0])
for key, v in network.feedback_errors.items():
if len(v) > 0:
feedback_errors[key].append(v[0])
plt.ioff()
plt.subplot(4, 1, 1)
for k, v in feedforward_errors.items():
if len(v) > 0:
plt.plot(range(len(v)), v, label=k)
plt.xlabel('epochs')
plt.ylabel('feedforward_errors')
# plt.title('feedforward_errors')
plt.legend(loc=1)
plt.subplot(4, 1, 2)
for k, v in recurrent_errors.items():
if len(v) > 0:
plt.plot(range(len(v)), v, label=k)
plt.xlabel('epochs')
plt.ylabel('recurrent_errors')
# plt.title('recurrent_errors')
plt.legend(loc=1)
plt.subplot(4, 1, 3)
for k, v in feedback_errors.items():
if len(v) > 0:
plt.plot(range(len(v)), v, label=k)
plt.xlabel('epochs')
plt.ylabel('feedback_errors')
# plt.title('recurrent_errors')
plt.legend(loc=1)
plt.subplot(4, 1, 4)
plt.plot(range(len(output_mse)), output_mse, label="mse")
plt.xlabel('epochs')
plt.ylabel('output_mse')
# plt.title('recurrent_errors')
plt.legend(loc=1)
plt.show()
