def regresswithpca(data, modelfile, regressmode):
"""Perform univariate regression,
followed by principal components analysis
to reduce dimensionality
:param data: RDD of data points as key value pairs
:param modelfile: model parameters (string with file location, array, or tuple)
:param regressmode: form of regression ("linear" or "bilinear")
:return stats: statistics of the fit
:return comps: compoents from PCA
:return scores: scores from PCA
:return latent: latent variances from PCA
"""
# create model
model = RegressionModel.load(modelfile, regressmode)
# do regression
betas, stats, resid = model.fit(data)
# do principal components analysis
scores, latent, comps = svd(betas, 2)
# compute trajectories from raw data
traj = model.fit(data, comps)
return stats, comps, latent, scores, traj
def crosscorr(data, sigfile, lag):
"""cross-correlate data points
(typically time series data)
against a signal over the specified lags
arguments:
data - RDD of data points
sigfile - signal to correlate with (string with file location or array)
lag - maximum lag (result will be 2*lag + 1)
returns:
betas - cross-correlations at different time lags
scores, latent, comps - result of applying pca if lag > 0
"""
# compute cross correlations
method = SigProcessingMethod.load("crosscorr", sigfile=sigfile, lag=lag)
betas = method.calc(data)
if lag is not 0:
# do PCA
scores, latent, comps = svd(betas, 2)
return betas, scores, latent, comps
else:
return betas
def regresswithpca(data, modelfile, regressmode, k=2):
"""Perform univariate regression,
followed by principal components analysis
to reduce dimensionality
:param data: RDD of data points as key value pairs
:param modelfile: model parameters (string with file location, array, or tuple)
:param regressmode: form of regression ("linear" or "bilinear")
:param k: number of principal components to compute
:return stats: statistics of the fit
:return comps: compoents from PCA
:return scores: scores from PCA
:return latent: latent variances from PCA
"""
# create model
model = RegressionModel.load(modelfile, regressmode)
# do regression
betas, stats, resid = model.fit(data)
# do principal components analysis
scores, latent, comps = svd(betas, k)
# compute trajectories from raw data
traj = model.fit(data, comps)
return stats, comps, latent, scores, traj
def regress(data, modelfile, regressmode):
"""perform mass univariate regression,
followed by principal components analysis
to reduce dimensionality
arguments:
data - RDD of data points
modelfile - model parameters (string with file location, array, or tuple)
regressmode - form of regression ("linear" or "bilinear")
returns:
stats - statistics of the fit
comps, latent, scores, traj - results of principal components analysis
"""
# create model
model = RegressionModel.load(modelfile, regressmode)
# do regression
betas, stats, resid = model.fit(data)
# do principal components analysis
scores, latent, comps = svd(betas, 2)
# compute trajectories from raw data
traj = model.fit(data, comps)
return stats, comps, latent, scores, traj
def crosscorr(data, sigfile, lag):
"""Cross-correlate data points
(typically time series data)
against a signal over the specified lags
:param data: RDD of data points as key value pairs
:param sigfile: signal to correlate with (string with file location or array)
:param lag: maximum lag (result will be length 2*lag + 1)
:return betas: cross-correlations at different time lags
:return scores: scores from PCA (if lag > 0)
:return latent: scores from PCA (if lag > 0)
:return comps: components from PCA (if lag > 0)
"""
# compute cross correlations
method = SigProcessingMethod.load("crosscorr", sigfile=sigfile, lag=lag)
betas = method.calc(data)
if lag is not 0:
# do PCA
scores, latent, comps = svd(betas, 2)
return betas, scores, latent, comps
else:
return betas
def crosscorr(data, sigfile, lag):
"""Cross-correlate data points
(typically time series data)
against a signal over the specified lags
:param data: RDD of data points as key value pairs
:param sigfile: signal to correlate with (string with file location or array)
:param lag: maximum lag (result will be length 2*lag + 1)
:return betas: cross-correlations at different time lags
:return scores: scores from PCA (if lag > 0)
:return latent: scores from PCA (if lag > 0)
:return comps: components from PCA (if lag > 0)
"""
# compute cross correlations
method = SigProcessingMethod.load("crosscorr", sigfile=sigfile, lag=lag)
betas = method.calc(data)
if lag is not 0:
# do PCA
scores, latent, comps = svd(betas, 2)
return betas, scores, latent, comps
else:
return betas
def ica(data, k, c, svdmethod="direct", maxiter=100, tol=0.000001, seed=0):
"""Perform independent components analysis
:param: data: RDD of data points
:param k: number of principal components to use
:param c: number of independent components to find
:param maxiter: maximum number of iterations (default = 100)
:param: tol: tolerance for change in estimate (default = 0.000001)
:return w: the mixing matrix
:return: sigs: the independent components
TODO: also return unmixing matrix
"""
# get count
n = data.count()
# reduce dimensionality
scores, latent, comps = svd(data, k, meansubtract=0, method=svdmethod)
# whiten data
whtmat = real(dot(inv(diag(latent / sqrt(n))), comps))
unwhtmat = real(dot(transpose(comps), diag(latent / sqrt(n))))
wht = data.mapValues(lambda x: dot(whtmat, x))
# do multiple independent component extraction
if seed != 0:
random.seed(seed)
b = orth(random.randn(k, c))
b_old = zeros((k, c))
iter = 0
minabscos = 0
errvec = zeros(maxiter)
while (iter < maxiter) & ((1 - minabscos) > tol):
iter += 1
# update rule for pow3 non-linearity (TODO: add others)
b = wht.map(lambda (_, v): v).map(
lambda x: outer(x,
dot(x, b)**3)).sum() / n - 3 * b
# make orthogonal
b = dot(b, real(sqrtm(inv(dot(transpose(b), b)))))
# evaluate error
minabscos = min(abs(diag(dot(transpose(b), b_old))))
# store results
b_old = b
errvec[iter - 1] = (1 - minabscos)
# get un-mixing matrix
w = dot(transpose(b), whtmat)
# get components
sigs = data.mapValues(lambda x: dot(w, x))
return w, sigs
def test_svd_direct(self):
data_local = [array([1.0, 2.0, 6.0]), array([1.0, 3.0, 0.0]), array([1.0, 4.0, 6.0]), array([5.0, 1.0, 4.0])]
data = self.sc.parallelize(zip(range(1, 5), data_local))
u, s, v = svd(data, 1, meansubtract=0, method="direct")
u_true, s_true, v_true = LinAlg.svd(array(data_local))
u_test = transpose(array(u.map(lambda (_, v): v).collect()))[0]
v_test = v[0]
assert allclose(s[0], s_true[0])
assert allclose(v_test, v_true[0, :]) | allclose(-v_test, v_true[0, :])
assert allclose(u_test, u_true[:, 0]) | allclose(-u_test, u_true[:, 0])
def ica(data, k, c, svdmethod="direct", maxiter=100, tol=0.000001, seed=0):
"""perform independent components analysis
arguments:
data - RDD of data points
k - number of principal components to use
c - number of independent components to find
maxiter - maximum number of iterations (default = 100)
tol - tolerance for change in estimate (default = 0.000001)
returns:
w - the mixing matrix
sigs - the independent components
"""
# get count
n = data.count()
# reduce dimensionality
scores, latent, comps = svd(data, k, meansubtract=0, method=svdmethod)
# whiten data
whtmat = real(dot(inv(diag(latent/sqrt(n))), comps))
unwhtmat = real(dot(transpose(comps), diag(latent/sqrt(n))))
wht = data.map(lambda x: dot(whtmat, x))
# do multiple independent component extraction
if seed != 0:
random.seed(seed)
b = orth(random.randn(k, c))
b_old = zeros((k, c))
iter = 0
minabscos = 0
errvec = zeros(maxiter)
while (iter < maxiter) & ((1 - minabscos) > tol):
iter += 1
# update rule for pow3 non-linearity (TODO: add others)
b = wht.map(lambda x: outer(x, dot(x, b) ** 3)).sum() / n - 3 * b
# make orthogonal
b = dot(b, real(sqrtm(inv(dot(transpose(b), b)))))
# evaluate error
minabscos = min(abs(diag(dot(transpose(b), b_old))))
# store results
b_old = b
errvec[iter-1] = (1 - minabscos)
# get un-mixing matrix
w = dot(transpose(b), whtmat)
# get components
sigs = data.map(lambda x: dot(w, x))
return w, sigs
def test_svd_em(self):
data_local = [array([1.0, 2.0, 6.0]), array([1.0, 3.0, 0.0]), array([1.0, 4.0, 6.0]), array([5.0, 1.0, 4.0])]
data = self.sc.parallelize(zip(range(1, 5), data_local))
u, s, v = svd(data, 1, meansubtract=0, method="em")
u_true, s_true, v_true = LinAlg.svd(array(data_local))
u_test = transpose(array(u.map(lambda (_, v): v).collect()))[0]
v_test = v[0]
tol = 10e-04 # allow small error for iterative method
assert allclose(s[0], s_true[0], atol=tol)
assert allclose(v_test, v_true[0, :], atol=tol) | allclose(-v_test, v_true[0, :], atol=tol)
assert allclose(u_test, u_true[:, 0], atol=tol) | allclose(-u_test, u_true[:, 0], atol=tol)
def pca(data, k, svdmethod="direct"):
"""Perform principal components analysis
using the singular value decomposition
:param data: RDD of data points as key value pairs
:param k: number of principal components to recover
:param svdmethod: which svd algorithm to use (default = "direct")
:return comps: the k principal components (as array)
:return latent: the latent values
:return scores: the k scores (as RDD)
"""
scores, latent, comps = svd(data, k, meansubtract=0, method=svdmethod)
return scores, latent, comps
def pca(data, k, svdmethod="direct"):
"""Perform principal components analysis
using the singular value decomposition
:param data: RDD of data points as key value pairs
:param k: number of principal components to recover
:param svdmethod: which svd algorithm to use (default = "direct")
:return comps: the k principal components (as array)
:return latent: the latent values
:return scores: the k scores (as RDD)
"""
scores, latent, comps = svd(data, k, meansubtract=0, method=svdmethod)
return scores, latent, comps
def test_svd_direct(self):
data_local = array([
array([1.0, 2.0, 6.0]),
array([1.0, 3.0, 0.0]),
array([1.0, 4.0, 6.0]),
array([5.0, 1.0, 4.0])
])
data = self.sc.parallelize(data_local)
u, s, v = svd(data, 1, meansubtract=0, method="direct")
u_true, s_true, v_true = LinAlg.svd(data_local)
u_test = transpose(array(u.collect()))[0]
v_test = v[0]
assert(allclose(s[0], s_true[0]))
assert(allclose(v_test, v_true[0, :]) | allclose(-v_test, v_true[0, :]))
assert(allclose(u_test, u_true[:, 0]) | allclose(-u_test, u_true[:, 0]))
def test_svd_direct(self):
data_local = [
array([1.0, 2.0, 6.0]),
array([1.0, 3.0, 0.0]),
array([1.0, 4.0, 6.0]),
array([5.0, 1.0, 4.0])
]
data = self.sc.parallelize(zip(range(1, 5), data_local))
u, s, v = svd(data, 1, meansubtract=0, method="direct")
u_true, s_true, v_true = LinAlg.svd(array(data_local))
u_test = transpose(array(u.map(lambda (_, v): v).collect()))[0]
v_test = v[0]
assert (allclose(s[0], s_true[0]))
assert (allclose(v_test, v_true[0, :])
| allclose(-v_test, v_true[0, :]))
assert (allclose(u_test, u_true[:, 0])
| allclose(-u_test, u_true[:, 0]))
def test_svd_em(self):
data_local = [
array([1.0, 2.0, 6.0]),
array([1.0, 3.0, 0.0]),
array([1.0, 4.0, 6.0]),
array([5.0, 1.0, 4.0])
]
data = self.sc.parallelize(zip(range(1, 5), data_local))
u, s, v = svd(data, 1, meansubtract=0, method="em")
u_true, s_true, v_true = LinAlg.svd(array(data_local))
u_test = transpose(array(u.map(lambda (_, v): v).collect()))[0]
v_test = v[0]
tol = 10e-04 # allow small error for iterative method
assert (allclose(s[0], s_true[0], atol=tol))
assert (allclose(v_test, v_true[0, :], atol=tol)
| allclose(-v_test, v_true[0, :], atol=tol))
assert (allclose(u_test, u_true[:, 0], atol=tol)
| allclose(-u_test, u_true[:, 0], atol=tol))
