def _stop_training(self):
self.data = numx.array(self.data, dtype=self.dtype)
self.data.shape = (self.tlen, self.input_dim)
# choose initial centroids unless they are already given
if not self._centroids:
import random
centr_idx = random.sample(xrange(self.tlen), self._num_clusters)
# numx_rand.permutation(self.tlen)[:self._num_clusters]
centroids = self.data[centr_idx]
else:
centroids = self._centroids
for step in xrange(self.max_iter):
# list of (sum_position, num_clusters)
new_centroids = [(0.0, 0.0)] * len(centroids)
# cluster
for x in self.data:
idx = self._nearest_centroid_idx(x, centroids)
# update position and count
pos_count = (new_centroids[idx][0] + x, new_centroids[idx][1] + 1.0)
new_centroids[idx] = pos_count
# get new centroid position
new_centroids = numx.array(
[c[0] / c[1] if c[1] > 0.0 else centroids[idx] for idx, c in enumerate(new_centroids)]
)
# check if we are stable
if numx.all(new_centroids == centroids):
self._centroids = centroids
return
centroids = new_centroids
def __init__(self, n, d, real_dtype="d", integer_dtype="l"):
"""
Input arguments:
n -- Number of maxima and minima to remember
d -- Minimum gap between two hits
real_dtype -- dtype of sequence items
integer_dtype -- dtype of sequence indices
Note: be careful with dtypes!
"""
self.n = int(n)
self.d = int(d)
self.iM = numx.zeros((n, ), dtype=integer_dtype)
self.im = numx.zeros((n, ), dtype=integer_dtype)
real_dtype = numx.dtype(real_dtype)
if real_dtype in mdp.utils.get_dtypes('AllInteger'):
max_num = numx.iinfo(real_dtype).max
min_num = numx.iinfo(real_dtype).min
else:
max_num = numx.finfo(real_dtype).max
min_num = numx.finfo(real_dtype).min
self.M = numx.array([min_num]*n, dtype=real_dtype)
self.m = numx.array([max_num]*n, dtype=real_dtype)
self.lM = 0
self.lm = 0
def __init__(self, n, d, real_dtype="d", integer_dtype="l"):
"""Initializes an object of type 'OneDimensionalHitParade'.
:param n: Number of maxima and minima to remember.
:type n: int
:param d: Minimum gap between two hits.
:type d: int
:param real_dtype: Datatype of sequence items
:type real_dtype: numpy.dtype or str
:param integer_dtype: Datatype of sequence indices
:type integer_dtype: numpy.dtype or str
"""
self.n = int(n)
self.d = int(d)
self.iM = numx.zeros((n, ), dtype=integer_dtype)
self.im = numx.zeros((n, ), dtype=integer_dtype)
real_dtype = numx.dtype(real_dtype)
if real_dtype in mdp.utils.get_dtypes('AllInteger'):
max_num = numx.iinfo(real_dtype).max
min_num = numx.iinfo(real_dtype).min
else:
max_num = numx.finfo(real_dtype).max
min_num = numx.finfo(real_dtype).min
self.M = numx.array([min_num] * n, dtype=real_dtype)
self.m = numx.array([max_num] * n, dtype=real_dtype)
self.lM = 0
self.lm = 0
def __init__(self, n, d, real_dtype="d", integer_dtype="l"):
"""Initializes an object of type 'OneDimensionalHitParade'.
:param n: Number of maxima and minima to remember.
:type n: int
:param d: Minimum gap between two hits.
:type d: int
:param real_dtype: Datatype of sequence items
:type real_dtype: numpy.dtype or str
:param integer_dtype: Datatype of sequence indices
:type integer_dtype: numpy.dtype or str
"""
self.n = int(n)
self.d = int(d)
self.iM = numx.zeros((n, ), dtype=integer_dtype)
self.im = numx.zeros((n, ), dtype=integer_dtype)
real_dtype = numx.dtype(real_dtype)
if real_dtype in mdp.utils.get_dtypes('AllInteger'):
max_num = numx.iinfo(real_dtype).max
min_num = numx.iinfo(real_dtype).min
else:
max_num = numx.finfo(real_dtype).max
min_num = numx.finfo(real_dtype).min
self.M = numx.array([min_num]*n, dtype=real_dtype)
self.m = numx.array([max_num]*n, dtype=real_dtype)
self.lM = 0
self.lm = 0
def __init__(self, n, d, real_dtype="d", integer_dtype="l"):
"""
Input arguments:
n -- Number of maxima and minima to remember
d -- Minimum gap between two hits
real_dtype -- dtype of sequence items
integer_dtype -- dtype of sequence indices
Note: be careful with dtypes!
"""
self.n = int(n)
self.d = int(d)
self.iM = numx.zeros((n, ), dtype=integer_dtype)
self.im = numx.zeros((n, ), dtype=integer_dtype)
real_dtype = numx.dtype(real_dtype)
if real_dtype in mdp.utils.get_dtypes('AllInteger'):
max_num = numx.iinfo(real_dtype).max
min_num = numx.iinfo(real_dtype).min
else:
max_num = numx.finfo(real_dtype).max
min_num = numx.finfo(real_dtype).min
self.M = numx.array([min_num]*n, dtype=real_dtype)
self.m = numx.array([max_num]*n, dtype=real_dtype)
self.lM = 0
self.lm = 0
def _stop_training(self):
self.data = numx.array(self.data, dtype=self.dtype)
self.data.shape = (self.tlen, self.input_dim)
# choose initial centroids unless they are already given
if not self._centroids:
import random
centr_idx = random.sample(xrange(self.tlen), self._num_clusters)
#numx_rand.permutation(self.tlen)[:self._num_clusters]
centroids = self.data[centr_idx]
else:
centroids = self._centroids
for step in xrange(self.max_iter):
# list of (sum_position, num_clusters)
new_centroids = [(0., 0.)] * len(centroids)
# cluster
for x in self.data:
idx = self._nearest_centroid_idx(x, centroids)
# update position and count
pos_count = (new_centroids[idx][0] + x,
new_centroids[idx][1] + 1.)
new_centroids[idx] = pos_count
# get new centroid position
new_centroids = numx.array([
c[0] / c[1] if c[1] > 0. else centroids[idx]
for idx, c in enumerate(new_centroids)
])
# check if we are stable
if numx.all(new_centroids == centroids):
self._centroids = centroids
return
centroids = new_centroids
def _set_dtype(self, dtype):
# when typecode is set, we set the whitening node if needed and
# the SFA and ICA weights
self._dtype = dtype
if not self.whitened and self.white.dtype is None:
self.white.dtype = dtype
self.icaweights = numx.array(self.icaweights, dtype)
self.sfaweights = numx.array(self.sfaweights, dtype)
def _set_dtype(self, dtype):
# when typecode is set, we set the whitening node if needed and
# the SFA and ICA weights
self._dtype = dtype
if not self.whitened and self.white.dtype is None:
self.white.dtype = dtype
self.icaweights = numx.array(self.icaweights, dtype)
self.sfaweights = numx.array(self.sfaweights, dtype)
def _get_nearest_nodes(self, x):
"""Return the two nodes in the graph that are nearest to x and their
squared distances.
:param x: Coordinates of point to compute distance to in order to
specifiy nearest nodes.
:type x: numpy.ndarray
:return: The coordinates of the nearest two nodes and their
distances to x. ([node1, node2], [dist1, dist2])
:rtype: tuple
"""
# distance function
def _distance_from_node(node):
#return norm(node.data.pos-x)**2
tmp = node.data.pos - x
return utils.mult(tmp, tmp)
g = self.graph
# distances of all graph nodes from x
distances = numx.array(list(map(_distance_from_node, g.nodes)))
ids = distances.argsort()[:2]
#nearest = [g.nodes[idx] for idx in ids]
#return nearest, distances[ids]
return (g.nodes[ids[0]], g.nodes[ids[1]]), distances.take(ids)
def _rank_nodes_by_distance(self, x):
"""Return the nodes in the graph in a list ranked by their squared
distance to x.
:param x: Point to compute distance to.
:type x: numpy.ndarray
:return: List of nodes ordered by the distance of the node to x.
:rtype: list
"""
#TODO: Refactor together with GNGNode._get_nearest_nodes
# distance function
def _distance_from_node(node):
tmp = node.data.pos - x
return utils.mult(tmp, tmp) # maps to mdp.numx.dot
g = self.graph
# distances of all graph nodes from x
distances = numx.array(list(map(_distance_from_node, g.nodes)))
ids = distances.argsort()
ranked_nodes = [g.nodes[id] for id in ids]
return ranked_nodes
def __init__(self, input_dim, connections):
"""Create a generic switchboard.
The input and output dimension as well as dtype have to be fixed
at initialization time.
Keyword arguments:
input_dim -- Dimension of the input data (number of connections).
connections -- 1d Array or sequence with an entry for each output
connection, containing the corresponding index of the
input connection.
"""
# check connections for inconsistencies
if len(connections) == 0:
err = "Received empty connection list."
raise SwitchboardException(err)
if numx.nanmax(connections) >= input_dim:
err = ("One or more switchboard connection "
"indices exceed the input dimension.")
raise SwitchboardException(err)
# checks passed
self.connections = numx.array(connections)
output_dim = len(connections)
super(Switchboard, self).__init__(input_dim=input_dim,
output_dim=output_dim)
# try to invert connections
if self.input_dim == self.output_dim and len(
numx.unique(self.connections)) == self.input_dim:
self.inverse_connections = numx.argsort(self.connections)
else:
self.inverse_connections = None
def train(self, data_iterables):
data_iterables = self._train_check_iterables(data_iterables)
if self.external_input_range is None:
external_input_range = []
else:
external_input_range = self.external_input_range
# train each Node successively
for i in range(len(self.flow)):
if self.verbose:
print "Training node #%d (%s)" % (i, str(self.flow[i]))
if not (data_iterables[i] == [] or data_iterables[i] is None):
# Delay the input timeseries with len(flow)-1, because every node connection introduces a one timestep delay
datax = [x[0:-i, :] for x in data_iterables[i][0]]
datay = []
for x in data_iterables[i][0]:
c = numx.array([True] * x.shape[1])
c[external_input_range] = False
datay.append(x[i:, c])
else:
datax, datay = [], []
self._train_node(zip(datax, datay), i)
if self.verbose:
print "Training finished"
self._close_last_node()
def _rank_nodes_by_distance(self, x):
"""Return the nodes in the graph in a list ranked by their squared
distance to x.
:param x: Point to compute distance to.
:type x: numpy.ndarray
:return: List of nodes ordered by the distance of the node to x.
:rtype: list
"""
#TODO: Refactor together with GNGNode._get_nearest_nodes
# distance function
def _distance_from_node(node):
tmp = node.data.pos - x
return utils.mult(tmp, tmp) # maps to mdp.numx.dot
g = self.graph
# distances of all graph nodes from x
distances = numx.array(list(map(_distance_from_node, g.nodes)))
ids = distances.argsort()
ranked_nodes = [g.nodes[id] for id in ids]
return ranked_nodes
def train(self, data_iterables):
data_iterables = self._train_check_iterables(data_iterables)
if self.external_input_range is None:
external_input_range = []
else:
external_input_range = self.external_input_range
# train each Node successively
for i in range(len(self.flow)):
if self.verbose:
print "Training node #%d (%s)" % (i, str(self.flow[i]))
if not data_iterables[i] == []:
# Delay the input timeseries with len(flow)-1, because every node connection introduces a one timestep delay
datax = [x[0:-i, :] for x in data_iterables[i][0]]
datay = []
for x in data_iterables[i][0]:
c = numx.array([True] * x.shape[1])
c[external_input_range] = False
datay.append(x[i:, c])
else:
datax, datay = [], []
self._train_node(zip(datax, datay), i)
if self.verbose:
print "Training finished"
self._close_last_node()
def __init__(self, input_dim, connections):
"""Create a generic switchboard.
The input and output dimension as well as dtype have to be fixed
at initialization time.
Keyword arguments:
input_dim -- Dimension of the input data (number of connections).
connections -- 1d Array or sequence with an entry for each output
connection, containing the corresponding index of the
input connection.
"""
# check connections for inconsistencies
if len(connections) == 0:
err = "Received empty connection list."
raise SwitchboardException(err)
if numx.nanmax(connections) >= input_dim:
err = ("One or more switchboard connection "
"indices exceed the input dimension.")
raise SwitchboardException(err)
# checks passed
self.connections = numx.array(connections)
output_dim = len(connections)
super(Switchboard, self).__init__(input_dim=input_dim,
output_dim=output_dim)
# try to invert connections
if (self.input_dim == self.output_dim and
len(numx.unique(self.connections)) == self.input_dim):
self.inverse_connections = numx.argsort(self.connections)
else:
self.inverse_connections = None
def test_RecursiveExpansionNode2():
"""Testing the tensor-base."""
data = 1e-6+np.random.rand(10, 4)
for functup in funcs:
func = functup[0]
degree = 4
name = functup[2]
recexpn = RecursiveExpansionNode(degree, recf=name,
check=False, with0=True)
resrec = recexpn.execute(data)
restensor = np.array([get_handcomputed_function_tensor(data[i, :], func, degree)
for i in range(data.shape[0])])
ressimplex = np.array(
[makeSimplex(restensor[i, :, :, :, :]) for i in range(data.shape[0])])
assert_array_almost_equal(resrec, ressimplex, decimal-3)
print('Multi dim ' + name + ' equal')
def _label(self, x):
if isinstance(x, (list, tuple, numx.ndarray)):
y = [0] * len(x)
p_labs, p_acc, p_vals = libsvmutil.svm_predict(y, x.tolist(), self.model)
return numx.array(p_labs)
else:
msg = "Data must be a sequence of vectors"
raise mdp.NodeException(msg)
def test_execute_routing(self):
"""Test the standard routing for messages."""
sboard = BiSwitchboard(input_dim=3, connections=[2,0,1])
x = n.array([[1,2,3],[4,5,6]])
msg = {
"string": "blabla",
"list": [1,2],
"data": x.copy(), # should be mapped by switchboard
"data2": n.zeros(3), # should not be modified
"data3": n.zeros((3,4)), # should not be modified
}
y, out_msg = sboard.execute(x, msg)
reference_y = n.array([[3,1,2],[6,4,5]])
assert (y == reference_y).all()
assert out_msg["string"] == msg["string"]
assert out_msg["list"] == msg["list"]
assert n.all(out_msg["data"] == reference_y)
assert out_msg["data2"].shape == (3,)
assert out_msg["data3"].shape == (3,4)
def _params(self):
"""Return the current parameters of the nodes."""
params = numx.array([])
for n in traverseHinet(self.gflow):
if hasattr(n, '_param_size') and n._param_size() > 0:
params = numx.concatenate((params, n.params()))
return params
def _label(self, x):
if isinstance(x, (list, tuple, numx.ndarray)):
y = [0] * len(x)
p_labs, p_acc, p_vals = libsvmutil.svm_predict(
y, x.tolist(), self.model)
return numx.array(p_labs)
else:
msg = "Data must be a sequence of vectors"
raise mdp.NodeException(msg)
def test_execute_routing(self):
"""Test the standard routing for messages."""
sboard = BiSwitchboard(input_dim=3, connections=[2,0,1])
x = n.array([[1,2,3],[4,5,6]])
msg = {
"string": "blabla",
"list": [1,2],
"data": x.copy(), # should be mapped by switchboard
"data2": n.zeros(3), # should not be modified
"data3": n.zeros((3,4)), # should not be modified
}
y, out_msg = sboard.execute(x, msg)
reference_y = n.array([[3,1,2],[6,4,5]])
assert (y == reference_y).all()
assert out_msg["string"] == msg["string"]
assert out_msg["list"] == msg["list"]
assert n.all(out_msg["data"] == reference_y)
assert out_msg["data2"].shape == (3,)
assert out_msg["data3"].shape == (3,4)
def _params(self):
"""Return the current parameters of the nodes."""
params = numx.array([])
for n in traverseHinet(self.gflow):
if hasattr(n, '_param_size') and n._param_size() > 0:
params = numx.concatenate((params, n.params()))
return params
def _label(self, x, threshold=0):
"""Retrieves patterns from the associative memory.
:param x: A matrix having different variables on different columns
and observations on rows.
:param threshold: numpy.ndarray
:return: The patterns.
"""
# todo: consider iterables
threshold = numx.zeros(self.input_dim) + threshold
return numx.array(
[self._label_one(pattern, threshold) for pattern in x])
def test_switchboard_gradient1(self):
"""Test that gradient is correct for a tiny switchboard."""
sboard = mdp.hinet.Switchboard(input_dim=4, connections=[2,0])
x = numx_rand.random((2,4))
mdp.activate_extension("gradient")
try:
result = sboard._gradient(x)
grad = result[1]["grad"]
ref_grad = numx.array([[[0,0,1,0], [1,0,0,0]],
[[0,0,1,0], [1,0,0,0]]], dtype=grad.dtype)
assert numx.all(grad == ref_grad)
finally:
mdp.deactivate_extension("gradient")
def test_RecursiveExpansionNode2():
"""Testing the tensor-base."""
data = 1e-6 + np.random.rand(10, 4)
for functup in funcs:
func = functup[0]
degree = 4
name = functup[2]
recexpn = RecursiveExpansionNode(degree,
recf=name,
check=False,
with0=True)
resrec = recexpn.execute(data)
restensor = np.array([
get_handcomputed_function_tensor(data[i, :], func, degree)
for i in range(data.shape[0])
])
ressimplex = np.array([
makeSimplex(restensor[i, :, :, :, :]) for i in range(data.shape[0])
])
assert_array_almost_equal(resrec, ressimplex, decimal - 3)
print('Multi dim ' + name + ' equal')
def _gradient(self):
"""Get the gradient with respect to the parameters.
This gradient has been calculated during the last backprop sweep.
"""
gradient = numx.array([])
for n in traverseHinet(self.gflow):
if hasattr(n, '_param_size') and n._param_size() > 0:
gradient = numx.concatenate((gradient, n.gradient()))
return gradient
def _gradient(self):
"""Get the gradient with respect to the parameters.
This gradient has been calculated during the last backprop sweep.
"""
gradient = numx.array([])
for n in traverseHinet(self.gflow):
if hasattr(n, '_param_size') and n._param_size() > 0:
gradient = numx.concatenate((gradient, n.gradient()))
return gradient
def test_inverse_message_routing(self):
"""Test the inverse routing for messages."""
sboard = BiSwitchboard(input_dim=3, connections=[2,0,1])
x = n.array([[1,2,3],[4,5,6]])
msg = {
"string": "blabla",
"method": "inverse",
"list": [1,2],
"data": x, # should be mapped by switchboard
"data2": n.zeros(3), # should not be modified
"data3": n.zeros((3,4)), # should not be modified
"target": "test"
}
y, out_msg, target = sboard.execute(None, msg)
assert y is None
assert target == "test"
reference_y = n.array([[2,3,1],[5,6,4]])
assert out_msg["string"] == msg["string"]
assert out_msg["list"] == msg["list"]
assert (out_msg["data"] == reference_y).all()
assert out_msg["data2"].shape == (3,)
assert out_msg["data3"].shape == (3,4)
def test_quadexpan_gradient1(self):
"""Test validity of gradient for QuadraticExpansionBiNode."""
node = mdp.nodes.QuadraticExpansionNode()
x = numx.array([[1, 3, 4]])
node.execute(x)
mdp.activate_extension("gradient")
try:
result = node._gradient(x)
grad = result[1]["grad"]
reference = numx.array(
[[[ 1, 0, 0], # x1
[ 0, 1, 0], # x2
[ 0, 0, 1], # x3
[ 2, 0, 0], # x1x1
[ 3, 1, 0], # x1x2
[ 4, 0, 1], # x1x3
[ 0, 6, 0], # x2x2
[ 0, 4, 3], # x2x3
[ 0, 0, 8]]]) # x3x3
assert numx.all(grad == reference)
finally:
mdp.deactivate_extension("gradient")
def _init_RBF(self, centers, sizes):
# initialize the centers of the RBFs
centers = numx.array(centers, self.dtype)
# define input/output dim
self.set_input_dim(centers.shape[1])
self.set_output_dim(centers.shape[0])
# multiply sizes if necessary
sizes = numx.array(sizes, self.dtype)
if sizes.ndim==0 or sizes.ndim==2:
sizes = numx.array([sizes]*self._output_dim)
else:
# check number of sizes correct
if sizes.shape[0] != self._output_dim:
msg = "There must be as many RBF sizes as centers"
raise mdp.NodeException, msg
if numx.isscalar(sizes[0]):
# isotropic RBFs
self._isotropic = True
else:
# anisotropic RBFs
self._isotropic = False
# check size
if (sizes.shape[1] != self._input_dim or
sizes.shape[2] != self._input_dim):
msg = ("Dimensionality of size matrices should be the same " +
"as input dimensionality (%d != %d)"
% (sizes.shape[1], self._input_dim))
raise mdp.NodeException, msg
# compute inverse covariance matrix
for i in range(sizes.shape[0]):
sizes[i,:,:] = mdp.utils.inv(sizes[i,:,:])
self._centers = centers
self._sizes = sizes
def test_inverse_message_routing(self):
"""Test the inverse routing for messages."""
sboard = BiSwitchboard(input_dim=3, connections=[2,0,1])
x = n.array([[1,2,3],[4,5,6]])
msg = {
"string": "blabla",
"method": "inverse",
"list": [1,2],
"data": x, # should be mapped by switchboard
"data2": n.zeros(3), # should not be modified
"data3": n.zeros((3,4)), # should not be modified
"target": "test"
}
y, out_msg, target = sboard.execute(None, msg)
assert y is None
assert target == "test"
reference_y = n.array([[2,3,1],[5,6,4]])
assert out_msg["string"] == msg["string"]
assert out_msg["list"] == msg["list"]
assert (out_msg["data"] == reference_y).all()
assert out_msg["data2"].shape == (3,)
assert out_msg["data3"].shape == (3,4)
def test_quadexpan_gradient1(self):
"""Test validity of gradient for QuadraticExpansionBiNode."""
node = mdp.nodes.QuadraticExpansionNode()
x = numx.array([[1, 3, 4]])
node.execute(x)
mdp.activate_extension("gradient")
try:
result = node._gradient(x)
grad = result[1]["grad"]
reference = numx.array([[
[1, 0, 0], # x1
[0, 1, 0], # x2
[0, 0, 1], # x3
[2, 0, 0], # x1x1
[3, 1, 0], # x1x2
[4, 0, 1], # x1x3
[0, 6, 0], # x2x2
[0, 4, 3], # x2x3
[0, 0, 8]
]]) # x3x3
assert numx.all(grad == reference)
finally:
mdp.deactivate_extension("gradient")
def test_switchboard_gradient1(self):
"""Test that gradient is correct for a tiny switchboard."""
sboard = mdp.hinet.Switchboard(input_dim=4, connections=[2, 0])
x = numx_rand.random((2, 4))
mdp.activate_extension("gradient")
try:
result = sboard._gradient(x)
grad = result[1]["grad"]
ref_grad = numx.array(
[[[0, 0, 1, 0], [1, 0, 0, 0]], [[0, 0, 1, 0], [1, 0, 0, 0]]],
dtype=grad.dtype)
assert numx.all(grad == ref_grad)
finally:
mdp.deactivate_extension("gradient")
def _get_nearest_nodes(self, x):
"""Return the two nodes in the graph that are nearest to x and their
squared distances. (Return ([node1, node2], [dist1, dist2])"""
# distance function
def _distance_from_node(node):
#return norm(node.data.pos-x)**2
tmp = node.data.pos - x
return utils.mult(tmp, tmp)
g = self.graph
# distances of all graph nodes from x
distances = numx.array(map(_distance_from_node, g.nodes))
ids = distances.argsort()[:2]
#nearest = [g.nodes[idx] for idx in ids]
#return nearest, distances[ids]
return (g.nodes[ids[0]], g.nodes[ids[1]]), distances.take(ids)
def _get_nearest_nodes(self, x):
"""Return the two nodes in the graph that are nearest to x and their
squared distances. (Return ([node1, node2], [dist1, dist2])"""
# distance function
def _distance_from_node(node):
#return norm(node.data.pos-x)**2
tmp = node.data.pos - x
return utils.mult(tmp, tmp)
g = self.graph
# distances of all graph nodes from x
distances = numx.array(map(_distance_from_node, g.nodes))
ids = distances.argsort()[:2]
#nearest = [g.nodes[idx] for idx in ids]
#return nearest, distances[ids]
return (g.nodes[ids[0]], g.nodes[ids[1]]), distances.take(ids)
def _rank_nodes_by_distance(self, x):
"""Return the nodes in the graph in a list ranked by their squared
distance to x. """
#TODO: Refactor together with GNGNode._get_nearest_nodes
# distance function
def _distance_from_node(node):
tmp = node.data.pos - x
return utils.mult(tmp, tmp) # maps to mdp.numx.dot
g = self.graph
# distances of all graph nodes from x
distances = numx.array(map(_distance_from_node, g.nodes))
ids = distances.argsort()
ranked_nodes = [g.nodes[id] for id in ids]
return ranked_nodes
def _rank_nodes_by_distance(self, x):
"""Return the nodes in the graph in a list ranked by their squared
distance to x. """
#TODO: Refactor together with GNGNode._get_nearest_nodes
# distance function
def _distance_from_node(node):
tmp = node.data.pos - x
return utils.mult(tmp, tmp) # maps to mdp.numx.dot
g = self.graph
# distances of all graph nodes from x
distances = numx.array(map(_distance_from_node, g.nodes))
ids = distances.argsort()
ranked_nodes = [g.nodes[id] for id in ids]
return ranked_nodes
def _stop_training(self):
super(GrowingNeuralGasExpansionNode, self)._stop_training()
# set the output dimension to the number of nodes of the neural gas
self._output_dim = self.get_nodes_position().shape[0]
# use the nodes of the learned neural gas as centers for a radial
# basis function expansion.
centers = self.get_nodes_position()
# use the mean distances to the neighbours as size of the RBF expansion
sizes = []
for i,node in enumerate(self.graph.nodes):
# calculate the size of the current RBF
pos = node.data.pos
sizes.append(numx.array([((pos-neighbor.data.pos)**2).sum()
for neighbor in node.neighbors() ]).mean())
# initialize the radial basis function expansion with centers and sizes
self.rbf_expansion = mdp.nodes.RBFExpansionNode(centers = centers,
sizes = sizes)
def _get_nearest_nodes(self, x):
"""Return the two nodes in the graph that are nearest to x and their
squared distances.
:param x: Coordinates of point to compute distance to in order to
specifiy nearest nodes.
:type x: numpy.ndarray
:return: The coordinates of the nearest two nodes and their
distances to x. ([node1, node2], [dist1, dist2])
:rtype: tuple
"""
# distance function
def _distance_from_node(node):
#return norm(node.data.pos-x)**2
tmp = node.data.pos - x
return utils.mult(tmp, tmp)
g = self.graph
# distances of all graph nodes from x
distances = numx.array(list(map(_distance_from_node, g.nodes)))
ids = distances.argsort()[:2]
#nearest = [g.nodes[idx] for idx in ids]
#return nearest, distances[ids]
return (g.nodes[ids[0]], g.nodes[ids[1]]), distances.take(ids)
def get_nodes_position(self):
return numx.array(map(lambda n: n.data.pos, self.graph.nodes),
dtype = self.dtype)
def _label(self, x, threshold=0):
"""Retrieves patterns from the associative memory.
"""
threshold = numx.zeros(self.input_dim) + threshold
return numx.array([self._label_one(pattern, threshold) for pattern in x])
def _stop_training(self, *args, **kwargs):
"""Transform the data and labels lists to array objects and reshape them."""
self.data = numx.array(self.data, dtype=self.dtype)
self.data.shape = (self.tlen, self.input_dim)
self.labels = numx.array(self.labels)
self.labels.shape = (self.tlen)
def makeSimplex(tensor):
simplex = np.concatenate(
(tensor[:, 0, 0, 0], tensor[0, 1:, 0, 0], tensor[0, 0, 1:,
0], tensor[0, 0, 0,
1:]))
x1 = tensor[1, 0, 0, 0]
x2 = tensor[2, 0, 0, 0]
x3 = tensor[3, 0, 0, 0]
y1 = tensor[0, 1, 0, 0]
y2 = tensor[0, 2, 0, 0]
y3 = tensor[0, 3, 0, 0]
z1 = tensor[0, 0, 1, 0]
z2 = tensor[0, 0, 2, 0]
z3 = tensor[0, 0, 3, 0]
w1 = tensor[0, 0, 0, 1]
w2 = tensor[0, 0, 0, 2]
w3 = tensor[0, 0, 0, 3]
simplex = np.concatenate((simplex, x1 * np.array([y1, y2, y3])))
simplex = np.concatenate((simplex, x2 * np.array([y1, y2])))
simplex = np.concatenate((simplex, np.array([x3 * y1])))
simplex = np.concatenate((simplex, x1 * np.array([z1, z2, z3])))
simplex = np.concatenate((simplex, x2 * np.array([z1, z2])))
simplex = np.concatenate((simplex, x3 * np.array([z1])))
simplex = np.concatenate((simplex, y1 * np.array([z1, z2, z3])))
simplex = np.concatenate((simplex, y2 * np.array([z1, z2])))
simplex = np.concatenate((simplex, y3 * np.array([z1])))
simplex = np.concatenate((simplex, x1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, x2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, x3 * np.array([w1])))
simplex = np.concatenate((simplex, y1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, y2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, y3 * np.array([w1])))
simplex = np.concatenate((simplex, z1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, z2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, z3 * np.array([w1])))
simplex = np.concatenate((simplex, x1 * np.array([y1 * z1, y1 * z2])))
simplex = np.concatenate((simplex, x2 * np.array([y1 * z1])))
simplex = np.concatenate((simplex, x1 * np.array([y2 * z1])))
simplex = np.concatenate((simplex, x1 * np.array([y1 * w1, y1 * w2])))
simplex = np.concatenate((simplex, x2 * np.array([y1 * w1])))
simplex = np.concatenate((simplex, x1 * np.array([y2 * w1])))
simplex = np.concatenate((simplex, x1 * np.array([z1 * w1, z1 * w2])))
simplex = np.concatenate((simplex, x2 * np.array([z1 * w1])))
simplex = np.concatenate((simplex, y1 * np.array([z1 * w1, z1 * w2])))
simplex = np.concatenate((simplex, y2 * np.array([z1 * w1])))
simplex = np.concatenate((simplex, x1 * np.array([z2 * w1])))
simplex = np.concatenate((simplex, y1 * np.array([z2 * w1])))
simplex = np.concatenate((simplex, x1 * np.array([y1 * z1 * w1])))
return simplex
def get_nodes_position(self):
return numx.array([n.data.pos for n in self.graph.nodes],
dtype=self.dtype)
def _stop_training(self, *args, **kwargs):
"""Transform the data and labels lists to array objects and reshape them."""
self.data = numx.array(self.data, dtype=self.dtype)
self.data.shape = (self.tlen, self.input_dim)
self.labels = numx.array(self.labels)
self.labels.shape = (self.tlen)
def test_symeig_fake_integer():
a = numx.array([[1,2],[2,7]])
b = numx.array([[3,1],[1,5]])
w,z = utils._symeig._symeig_fake(a)
w,z = utils._symeig._symeig_fake(a,b)
def _nearest_centroid_idx(self, data, centroids):
dists = numx.array([numx.linalg.norm(data - c) for c in centroids])
return dists.argmin()
def get_nodes_position(self):
return numx.array([n.data.pos for n in self.graph.nodes],
dtype = self.dtype)
def _label(self, x, threshold=0):
"""Retrieves patterns from the associative memory.
"""
threshold = numx.zeros(self.input_dim) + threshold
return numx.array(
[self._label_one(pattern, threshold) for pattern in x])
def _nearest_centroid_idx(self, data, centroids):
dists = numx.array([numx.linalg.norm(data - c) for c in centroids])
return dists.argmin()
def makeSimplex(tensor):
simplex = np.concatenate(
(tensor[:, 0, 0, 0], tensor[0, 1:, 0, 0], tensor[0, 0, 1:, 0],
tensor[0, 0, 0, 1:]))
x1 = tensor[1, 0, 0, 0]
x2 = tensor[2, 0, 0, 0]
x3 = tensor[3, 0, 0, 0]
y1 = tensor[0, 1, 0, 0]
y2 = tensor[0, 2, 0, 0]
y3 = tensor[0, 3, 0, 0]
z1 = tensor[0, 0, 1, 0]
z2 = tensor[0, 0, 2, 0]
z3 = tensor[0, 0, 3, 0]
w1 = tensor[0, 0, 0, 1]
w2 = tensor[0, 0, 0, 2]
w3 = tensor[0, 0, 0, 3]
simplex = np.concatenate((simplex, x1 * np.array([y1, y2, y3])))
simplex = np.concatenate((simplex, x2 * np.array([y1, y2])))
simplex = np.concatenate((simplex, np.array([x3 * y1])))
simplex = np.concatenate((simplex, x1 * np.array([z1, z2, z3])))
simplex = np.concatenate((simplex, x2 * np.array([z1, z2])))
simplex = np.concatenate((simplex, x3 * np.array([z1])))
simplex = np.concatenate((simplex, y1 * np.array([z1, z2, z3])))
simplex = np.concatenate((simplex, y2 * np.array([z1, z2])))
simplex = np.concatenate((simplex, y3 * np.array([z1])))
simplex = np.concatenate((simplex, x1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, x2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, x3 * np.array([w1])))
simplex = np.concatenate((simplex, y1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, y2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, y3 * np.array([w1])))
simplex = np.concatenate((simplex, z1 * np.array([w1, w2, w3])))
simplex = np.concatenate((simplex, z2 * np.array([w1, w2])))
simplex = np.concatenate((simplex, z3 * np.array([w1])))
simplex = np.concatenate((simplex, x1 * np.array([y1*z1, y1*z2])))
simplex = np.concatenate((simplex, x2 * np.array([y1*z1])))
simplex = np.concatenate((simplex, x1 * np.array([y2*z1])))
simplex = np.concatenate((simplex, x1 * np.array([y1*w1, y1*w2])))
simplex = np.concatenate((simplex, x2 * np.array([y1*w1])))
simplex = np.concatenate((simplex, x1 * np.array([y2*w1])))
simplex = np.concatenate((simplex, x1 * np.array([z1*w1, z1*w2])))
simplex = np.concatenate((simplex, x2 * np.array([z1*w1])))
simplex = np.concatenate((simplex, y1 * np.array([z1*w1, z1*w2])))
simplex = np.concatenate((simplex, y2 * np.array([z1*w1])))
simplex = np.concatenate((simplex, x1 * np.array([z2*w1])))
simplex = np.concatenate((simplex, y1 * np.array([z2*w1])))
simplex = np.concatenate((simplex, x1 * np.array([y1*z1*w1])))
return simplex
def __init__(self, lags=1, sfa_ica_coeff=(1., 1.), icaweights=None,
sfaweights=None, whitened=False, white_comp = None,
white_parm = None, eps_contrast=1e-6, max_iter=10000,
RP=None, verbose=False, input_dim=None, output_dim=None,
dtype=None):
"""Initializes an object of type 'ISFANode' to perform
Independent Slow Feature Analysis.
The notation is the same used in the paper by Blaschke et al. Please
refer to the paper for more information.
:param lags: A list of time-lags to generate the time-delayed covariance
matrices (in the paper this is the set of \tau). If lags is an
integer, time-lags 1,2,...,'lags' are used.
Note that time-lag == 0 (instantaneous correlation) is always
implicitly used.
:type lags: list or int
:param sfa_ica_coeff: A list of float with two entries, which defines the
weights of the SFA and ICA part of the objective function.
They are called b_{SFA} and b_{ICA} in the paper.
:type sfa_ica_coeff: list
:param icaweights: Weighting factors for the cov matrices relative
to the ICA part of the objective function (called
\kappa_{ICA}^{\tau} in the paper). Default is 1.
Possible values are:
- An integer ``n``: All matrices are weighted the same
(note that it does not make sense to have ``n != 1``).
- A list or array of floats of ``len == len(lags)``:
Each element of the list is used for weighting the
corresponding matrix.
- ``None``: Use the default values.
:type icaweights: int, list or array
:param sfaweights: Weighting factors for the covariance matrices relative
to the SFA part of the objective function (called
\kappa_{SFA}^{\tau} in the paper). Default is [1., 0., ..., 0.]
For possible values see the description of icaweights.
:type sfaweights: int, list or array
:param whitened: ``True`` if input data is already white, ``False``
otherwise (the data will be whitened internally).
:type whitened: bool
:param white_comp: If whitened is false, you can set ``white_comp`` to the
number of whitened components to keep during the
calculation (i.e., the input dimensions are reduced to
`white_comp`` by keeping the components of largest variance).
:type white_comp: int
:param white_parm: A dictionary with additional parameters for whitening.
It is passed directly to the WhiteningNode constructor.
Ex: white_parm = { 'svd' : True }
:type white_parm: dict
:param eps_contrast: Convergence is achieved when the relative
improvement in the contrast is below this threshold.
Values in the range [1E-4, 1E-10] are usually reasonable.
:type eps_contrast: float
:param max_iter: If the algorithms does not achieve convergence within
max_iter iterations raise an Exception.
Should be larger than 100.
:type max_iter: int
:param RP: Starting rotation-permutation matrix. It is an
input_dim x input_dim matrix used to initially rotate the
input components. If not set, the identity matrix is used.
In the paper this is used to start the algorithm at the
SFA solution (which is often quite near to the optimum).
:param verbose: Print progress information during convergence. This can
slow down the algorithm, but it's the only way to see
the rate of improvement and immediately spot if something
is going wrong.
:type verbose: bool
:param input_dim: The input dimensionality.
:type input_dim: int
:param output_dim: Sets the number of independent components that have to
be extracted. Note that if this is not smaller than input_dim,
the problem is solved linearly and SFA would give the same
solution only much faster.
:type output_dim: int
:param dtype: Datatype to be used.
:type dtype: numpy.dtype or str
"""
# check that the "lags" argument has some meaningful value
if isinstance(lags, (int, int)):
lags = list(range(1, lags+1))
elif isinstance(lags, (list, tuple)):
lags = numx.array(lags, "i")
elif isinstance(lags, numx.ndarray):
if not (lags.dtype.char in ['i', 'l']):
err_str = "lags must be integer!"
raise NodeException(err_str)
else:
pass
else:
err_str = ("Lags must be int, list or array. Found "
"%s!" % (type(lags).__name__))
raise NodeException(err_str)
self.lags = lags
# sanity checks for weights
if icaweights is None:
self.icaweights = 1.
else:
if (len(icaweights) != len(lags)):
err = ("icaweights vector length is %d, "
"should be %d" % (str(len(icaweights)), str(len(lags))))
raise NodeException(err)
self.icaweights = icaweights
if sfaweights is None:
self.sfaweights = [0]*len(lags)
self.sfaweights[0] = 1.
else:
if (len(sfaweights) != len(lags)):
err = ("sfaweights vector length is %d, "
"should be %d" % (str(len(sfaweights)), str(len(lags))))
raise NodeException(err)
self.sfaweights = sfaweights
# store attributes
self.sfa_ica_coeff = sfa_ica_coeff
self.max_iter = max_iter
self.verbose = verbose
self.eps_contrast = eps_contrast
# if input is not white, insert a WhiteningNode
self.whitened = whitened
if not whitened:
if white_parm is None:
white_parm = {}
if output_dim is not None:
white_comp = output_dim
elif white_comp is not None:
output_dim = white_comp
self.white = WhiteningNode(input_dim=input_dim,
output_dim=white_comp,
dtype=dtype, **white_parm)
# initialize covariance matrices
self.covs = [ DelayCovarianceMatrix(dt, dtype=dtype) for dt in lags ]
# initialize the global rotation-permutation matrix
# if not set that we'll eventually be an identity matrix
self.RP = RP
# initialize verbose structure to print nice and useful progress info
if verbose:
info = { 'sweep' : max(len(str(self.max_iter)), 5),
'perturbe': max(len(str(self.max_iter)), 5),
'float' : 5+8,
'fmt' : "%.5e",
'sep' : " | "}
f1 = "Sweep".center(info['sweep'])
f1_2 = "Pertb". center(info['perturbe'])
f2 = "SFA part".center(info['float'])
f3 = "ICA part".center(info['float'])
f4 = "Contrast".center(info['float'])
header = info['sep'].join([f1, f1_2, f2, f3, f4])
info['header'] = header+'\n'
info['line'] = len(header)*"-"
self._info = info
# finally call base class constructor
super(ISFANode, self).__init__(input_dim, output_dim, dtype)
def get_nodes_position(self):
return numx.array(map(lambda n: n.data.pos, self.graph.nodes),
dtype=self.dtype)
def __init__(self, lags=1, sfa_ica_coeff=(1., 1.), icaweights=None,
sfaweights=None, whitened=False, white_comp = None,
white_parm = None, eps_contrast=1e-6, max_iter=10000,
RP=None, verbose=False, input_dim=None, output_dim=None,
dtype=None):
"""
Perform Independent Slow Feature Analysis.
The notation is the same used in the paper by Blaschke et al. Please
refer to the paper for more information.
:Parameters:
lags
list of time-lags to generate the time-delayed covariance
matrices (in the paper this is the set of \tau). If
lags is an integer, time-lags 1,2,...,'lags' are used.
Note that time-lag == 0 (instantaneous correlation) is
always implicitly used.
sfa_ica_coeff
a list of float with two entries, which defines the
weights of the SFA and ICA part of the objective
function. They are called b_{SFA} and b_{ICA} in the
paper.
sfaweights
weighting factors for the covariance matrices relative
to the SFA part of the objective function (called
\kappa_{SFA}^{\tau} in the paper). Default is
[1., 0., ..., 0.]
For possible values see the description of icaweights.
icaweights
weighting factors for the cov matrices relative
to the ICA part of the objective function (called
\kappa_{ICA}^{\tau} in the paper). Default is 1.
Possible values are:
- an integer ``n``: all matrices are weighted the same
(note that it does not make sense to have ``n != 1``)
- a list or array of floats of ``len == len(lags)``:
each element of the list is used for weighting the
corresponding matrix
- ``None``: use the default values.
whitened
``True`` if input data is already white, ``False``
otherwise (the data will be whitened internally).
white_comp
If whitened is false, you can set ``white_comp`` to the
number of whitened components to keep during the
calculation (i.e., the input dimensions are reduced to
``white_comp`` by keeping the components of largest variance).
white_parm
a dictionary with additional parameters for whitening.
It is passed directly to the WhiteningNode constructor.
Ex: white_parm = { 'svd' : True }
eps_contrast
Convergence is achieved when the relative
improvement in the contrast is below this threshold.
Values in the range [1E-4, 1E-10] are usually
reasonable.
max_iter
If the algorithms does not achieve convergence within
max_iter iterations raise an Exception. Should be
larger than 100.
RP
Starting rotation-permutation matrix. It is an
input_dim x input_dim matrix used to initially rotate the
input components. If not set, the identity matrix is used.
In the paper this is used to start the algorithm at the
SFA solution (which is often quite near to the optimum).
verbose
print progress information during convergence. This can
slow down the algorithm, but it's the only way to see
the rate of improvement and immediately spot if something
is going wrong.
output_dim
sets the number of independent components that have to
be extracted. Note that if this is not smaller than
input_dim, the problem is solved linearly and SFA
would give the same solution only much faster.
"""
# check that the "lags" argument has some meaningful value
if isinstance(lags, (int, int)):
lags = list(range(1, lags+1))
elif isinstance(lags, (list, tuple)):
lags = numx.array(lags, "i")
elif isinstance(lags, numx.ndarray):
if not (lags.dtype.char in ['i', 'l']):
err_str = "lags must be integer!"
raise NodeException(err_str)
else:
pass
else:
err_str = ("Lags must be int, list or array. Found "
"%s!" % (type(lags).__name__))
raise NodeException(err_str)
self.lags = lags
# sanity checks for weights
if icaweights is None:
self.icaweights = 1.
else:
if (len(icaweights) != len(lags)):
err = ("icaweights vector length is %d, "
"should be %d" % (str(len(icaweights)), str(len(lags))))
raise NodeException(err)
self.icaweights = icaweights
if sfaweights is None:
self.sfaweights = [0]*len(lags)
self.sfaweights[0] = 1.
else:
if (len(sfaweights) != len(lags)):
err = ("sfaweights vector length is %d, "
"should be %d" % (str(len(sfaweights)), str(len(lags))))
raise NodeException(err)
self.sfaweights = sfaweights
# store attributes
self.sfa_ica_coeff = sfa_ica_coeff
self.max_iter = max_iter
self.verbose = verbose
self.eps_contrast = eps_contrast
# if input is not white, insert a WhiteningNode
self.whitened = whitened
if not whitened:
if white_parm is None:
white_parm = {}
if output_dim is not None:
white_comp = output_dim
elif white_comp is not None:
output_dim = white_comp
self.white = WhiteningNode(input_dim=input_dim,
output_dim=white_comp,
dtype=dtype, **white_parm)
# initialize covariance matrices
self.covs = [ DelayCovarianceMatrix(dt, dtype=dtype) for dt in lags ]
# initialize the global rotation-permutation matrix
# if not set that we'll eventually be an identity matrix
self.RP = RP
# initialize verbose structure to print nice and useful progress info
if verbose:
info = { 'sweep' : max(len(str(self.max_iter)), 5),
'perturbe': max(len(str(self.max_iter)), 5),
'float' : 5+8,
'fmt' : "%.5e",
'sep' : " | "}
f1 = "Sweep".center(info['sweep'])
f1_2 = "Pertb". center(info['perturbe'])
f2 = "SFA part".center(info['float'])
f3 = "ICA part".center(info['float'])
f4 = "Contrast".center(info['float'])
header = info['sep'].join([f1, f1_2, f2, f3, f4])
info['header'] = header+'\n'
info['line'] = len(header)*"-"
self._info = info
# finally call base class constructor
super(ISFANode, self).__init__(input_dim, output_dim, dtype)