def time_learn(rows, cols):
'''Return time elapsed to learn 10000 iterations on data'''
x, y = get_test_data(rows, cols)
net = geneticnetwork(x.shape[1], 8, 1)
connect_feedforward(net)
net.generations = 10
# Time it
start = time.time()
net.learn(x, y)
# Final time
elapsed = time.time() - start
return elapsed
def time_learn(rows, cols):
'''Return time elapsed to learn 10000 iterations on data'''
x, y = get_test_data(rows, cols)
net = geneticnetwork(x.shape[1], 8, 1)
connect_feedforward(net)
net.generations=10
# Time it
start = time.time()
net.learn(x, y)
# Final time
elapsed = time.time() - start
return elapsed
def time_learn(rows, cols):
"""Return time elapsed to learn 10000 iterations on data"""
x, y = get_test_data(rows, cols)
net = rpropnetwork(x.shape[1], 8, 1)
connect_feedforward(net)
net.maxError = 0
net.maxEpochs = 1000
# Time it
start = time.time()
net.learn(x, y)
# Final time
elapsed = time.time() - start
return elapsed
def test_output():
"""Numpy array results in different output from Python list"""
from ann import rpropnetwork
from ann.utils import connect_feedforward
data = np.random.normal(size=(10,6))
np.random.shuffle(data)
incols = list(range(2,6))
outcols = [1, 0]
net = rpropnetwork(len(incols), 3, len(outcols))
connect_feedforward(net)
for row in data[:3, incols]:
np_out = net.output(row)
py_out = net.output(list(row))
for n, p in zip(np_out, py_out):
assert n == p
def test_output():
"""Numpy array results in different output from Python list"""
from ann import rpropnetwork
from ann.utils import connect_feedforward
data = np.random.normal(size=(10, 6))
np.random.shuffle(data)
incols = list(range(2, 6))
outcols = [1, 0]
net = rpropnetwork(len(incols), 3, len(outcols))
connect_feedforward(net)
for row in data[:3, incols]:
np_out = net.output(row)
py_out = net.output(list(row))
for n, p in zip(np_out, py_out):
assert n == p
def test_matrixnetwork():
from ann import matrixnetwork
net = matrixnetwork(2, 2, 1)
# Default start is zero
xor_in, xor_out = getXOR()
for val in xor_in:
assert net.output(val) == 0, "Expected zero output as default"
length = net.input_count + net.hidden_count + net.output_count + 1
# Weights
#print("Length: ", len(net.weights))
#print(net.weights)
assert len(net.weights.ravel()) == length**2, "Wrong length of weight vector"
assert np.all(net.weights == 0) == True, "Expected weights to equal zero"
weights = net.weights.ravel()
for i in range(len(weights)):
weights[i] = 1.0
net.weights = weights
#print(net.weights)
assert np.all(net.weights == 1.0) == True, "Expected weights to equal 1"
# Connections
#print("Length: ", len(net.connections))
#print(net.connections)
assert len(net.connections.ravel()) == length**2, "Wrong length of conns vector"
assert np.all(net.connections == 0), "Expected conns to equal zero"
conns = net.connections.ravel()
for i in range(len(conns)):
conns[i] = 1
net.connections = conns
#print(net.connections)
assert np.all(net.connections == 1), "Expected conns to equal 1"
# Default dropout value
assert np.all(net.dropout_probabilities == 1), "Expected default dropout probs to be 1"
dp = net.dropout_probabilities
dp[:] = 0.5
net.dropout_probabilities = dp
assert np.all(net.dropout_probabilities == 0.5), "Expected dropout probs to be 0.5"
# Set one again
dp[:] = 1.0
net.dropout_probabilities = dp
assert np.all(net.dropout_probabilities == 1), "Expected dropout probs to be 1"
# ActFuncs
#print("Length: ", len(net.activation_functions))
#print(net.activation_functions)
assert len(net.activation_functions) == length, \
"Wrong length of funcs vector"
assert np.all(net.activation_functions == net.LOGSIG), \
"Expected funcs to be LINEAR"
actfuncs = net.activation_functions
#actfuncs = np.zeros(len(net.activation_functions))
for i in range(len(actfuncs)):
actfuncs[i] = net.TANH
#print(actfuncs)
net.activation_functions = actfuncs
#print(net.activation_functions)
assert np.all(net.activation_functions == net.TANH), \
"Expected actfuncs to be TANH"
# OUTPUTS
for val in xor_in:
assert net.output(val) != 0, "Expected some output"
# Solve XOR
# set all conns to zero first
conns[:] = 0
net.connections = conns
# Connect feedforward
from ann.utils import connect_feedforward
connect_feedforward(net)
# We care only about first two neurons and output
actfuncs[3] = net.LOGSIG
actfuncs[4] = net.LOGSIG
actfuncs[5] = net.LINEAR
net.activation_functions = actfuncs
weights[:] = 0
weights[3*length + 0] = -60
weights[3*length + 1] = 60
weights[3*length + 2] = -30
weights[4*length + 0] = 60
weights[4*length + 1] = -60
weights[4*length + 2] = -30
weights[(length-1)*length + 3] = 1
weights[(length-1)*length + 4] = 1
net.weights = weights
print(net.connections.reshape((6,6)))
print(net.weights.reshape((6,6)))
print(net.activation_functions)
for val in xor_in:
print("In:", val, " out:", net.output(val))
if sum(val) != 1:
assert net.output(val) < 0.1, "xor solution doesnt work"
else:
assert net.output(val) > 0.9, "xor solution doesnt work"
conns[:] = 0
net.connections = conns
for val in xor_in:
#print("In:", val, " out:", net.output(val))
assert net.output(val) == 0, "no conns should mean zero output!"
# Pickle test
import pickle
state = pickle.dumps(net, -1)
net2 = pickle.loads(state)
# Make sure it's the same
assert np.all(net.weights == net2.weights), "weights don't match"
assert np.all(net.connections == net2.connections), "conns don't match"
assert np.all(net.activation_functions == net2.activation_functions),\
"functions don't match"
def test_matrixnetwork():
from ann import matrixnetwork
net = matrixnetwork(2, 2, 1)
# Default start is zero
xor_in, xor_out = getXOR()
for val in xor_in:
assert net.output(val) == 0, "Expected zero output as default"
length = net.input_count + net.hidden_count + net.output_count + 1
# Weights
#print("Length: ", len(net.weights))
#print(net.weights)
assert len(
net.weights.ravel()) == length**2, "Wrong length of weight vector"
assert np.all(net.weights == 0) == True, "Expected weights to equal zero"
weights = net.weights.ravel()
for i in range(len(weights)):
weights[i] = 1.0
net.weights = weights
#print(net.weights)
assert np.all(net.weights == 1.0) == True, "Expected weights to equal 1"
# Connections
#print("Length: ", len(net.connections))
#print(net.connections)
assert len(
net.connections.ravel()) == length**2, "Wrong length of conns vector"
assert np.all(net.connections == 0), "Expected conns to equal zero"
conns = net.connections.ravel()
for i in range(len(conns)):
conns[i] = 1
net.connections = conns
#print(net.connections)
assert np.all(net.connections == 1), "Expected conns to equal 1"
# Default dropout value
assert np.all(net.dropout_probabilities ==
1), "Expected default dropout probs to be 1"
dp = net.dropout_probabilities
dp[:] = 0.5
net.dropout_probabilities = dp
assert np.all(
net.dropout_probabilities == 0.5), "Expected dropout probs to be 0.5"
# Set one again
dp[:] = 1.0
net.dropout_probabilities = dp
assert np.all(
net.dropout_probabilities == 1), "Expected dropout probs to be 1"
# ActFuncs
#print("Length: ", len(net.activation_functions))
#print(net.activation_functions)
assert len(net.activation_functions) == length, \
"Wrong length of funcs vector"
assert np.all(net.activation_functions == net.LOGSIG), \
"Expected funcs to be LINEAR"
actfuncs = net.activation_functions
#actfuncs = np.zeros(len(net.activation_functions))
for i in range(len(actfuncs)):
actfuncs[i] = net.TANH
#print(actfuncs)
net.activation_functions = actfuncs
#print(net.activation_functions)
assert np.all(net.activation_functions == net.TANH), \
"Expected actfuncs to be TANH"
# OUTPUTS
for val in xor_in:
assert net.output(val) != 0, "Expected some output"
# Solve XOR
# set all conns to zero first
conns[:] = 0
net.connections = conns
# Connect feedforward
from ann.utils import connect_feedforward
connect_feedforward(net)
# We care only about first two neurons and output
actfuncs[3] = net.LOGSIG
actfuncs[4] = net.LOGSIG
actfuncs[5] = net.LINEAR
net.activation_functions = actfuncs
weights[:] = 0
weights[3 * length + 0] = -60
weights[3 * length + 1] = 60
weights[3 * length + 2] = -30
weights[4 * length + 0] = 60
weights[4 * length + 1] = -60
weights[4 * length + 2] = -30
weights[(length - 1) * length + 3] = 1
weights[(length - 1) * length + 4] = 1
net.weights = weights
print(net.connections.reshape((6, 6)))
print(net.weights.reshape((6, 6)))
print(net.activation_functions)
for val in xor_in:
print("In:", val, " out:", net.output(val))
if sum(val) != 1:
assert net.output(val) < 0.1, "xor solution doesnt work"
else:
assert net.output(val) > 0.9, "xor solution doesnt work"
conns[:] = 0
net.connections = conns
for val in xor_in:
#print("In:", val, " out:", net.output(val))
assert net.output(val) == 0, "no conns should mean zero output!"
# Pickle test
import pickle
state = pickle.dumps(net, -1)
net2 = pickle.loads(state)
# Make sure it's the same
assert np.all(net.weights == net2.weights), "weights don't match"
assert np.all(net.connections == net2.connections), "conns don't match"
assert np.all(net.activation_functions == net2.activation_functions),\
"functions don't match"
