def do(self, a, b):
d = linalg.det(a)
(s, ld) = linalg.slogdet(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev))
assert_almost_equal(s * np.exp(ld), multiply.reduce(ev))
if s != 0:
assert_almost_equal(np.abs(s), 1)
else:
assert_equal(ld, -inf)
def do(self, a, b):
d = linalg.det(a)
(s, ld) = linalg.slogdet(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev))
assert_almost_equal(s * np.exp(ld), multiply.reduce(ev))
if s != 0:
assert_almost_equal(np.abs(s), 1)
else:
assert_equal(ld, -inf)
def DUMvNormal(*args, **kwargs):
v = pm.MvNormal(*args, **kwargs)
global BOUND
pm.Potential("bound {}".format(BOUND), tt.switch(multiply.reduce(\
multiply(0 <= v, v <= 1)), 0, -inf))
BOUND += 1
return v
def do(self, a, b):
d = linalg.det(a)
(s, ld) = linalg.slogdet(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev, axis=-1))
assert_almost_equal(s * np.exp(ld), multiply.reduce(ev, axis=-1))
s = np.atleast_1d(s)
ld = np.atleast_1d(ld)
m = (s != 0)
assert_almost_equal(np.abs(s[m]), 1)
assert_equal(ld[~m], -inf)
def do(self, a, b):
d = linalg.det(a)
(s, ld) = linalg.slogdet(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev, axis=-1))
assert_almost_equal(s * np.exp(ld), multiply.reduce(ev, axis=-1))
s = np.atleast_1d(s)
ld = np.atleast_1d(ld)
m = (s != 0)
assert_almost_equal(np.abs(s[m]), 1)
assert_equal(ld[~m], -inf)
def fit(self, X):
from numpy import sort, unique, allclose, multiply
from numpy.random import uniform
from pandas import DataFrame
from collections import deque
Centroids_history = deque(maxlen=2)
self.X = X
self.centroids_ = None
"""THE TRIALS LOOP - TO SELECT THE BEST RESULT"""
trials =[]
for trial_number in range(5): # max number of trials
"""GENERATE SATRTING RANDOM POINTS""" # n_clusters points; n_dim columns
space = sort(X, axis=0)[[0,-1],:]
while True:
C = current_centroids = uniform(low=space[0], high=space[1], size=(self.n_clusters, X.shape[-1]))
y_current = ((X[None, :, :] - C[:, None, :]) ** 2).sum(axis=-1).T.argmin(axis=1)
if len(unique(y_current)) == self.n_clusters:
break
"""THE INNER LOOP"""
df = DataFrame(X, dtype=X.dtype)
for loop_number in range(100):
"""RECALCULATE CENTROIDS"""
dfGrouped = df.groupby(by=y_current).mean()
"""IN CASE A CENTROID IS LOST"""
if len(unique(y_current))!=self.n_clusters:
lost_centroids = sorted(set(range(self.n_clusters)).difference(unique(y_current)))
appendage = DataFrame(C[lost_centroids], index=lost_centroids, dtype=C.dtype)
dfGrouped = dfGrouped.append(appendage).sort_index()
C = dfGrouped.values
"""CHECK IF IT IS TIME TO BREAK OUT OF THE LOOP"""
Centroids_history.append(C)
if len(Centroids_history)>=2 and allclose(Centroids_history[-2], Centroids_history[-1]):
break
else:
y_current = ((X[None, :, :] - C[:, None, :]) ** 2).sum(axis=-1).T.argmin(axis=1)
#continue the loop
else:
from warnings import warn
warn("max loops reached: {}".format(loop_number))
"""after (successful) breaking out of the loop: CALCULATE VARIANCE"""
cluster_variances = df.groupby(by=y_current).var(ddof=0).sum(axis=0) ** 0.5
hypervolume_as_metric_for_total_variance = multiply.reduce(cluster_variances.values)
trials.append((C, y_current, hypervolume_as_metric_for_total_variance))
else: "END OF THE TRIALS LOOP"
"""SELECTING THE BEST TRIAL"""
from operator import itemgetter
trials = sorted(trials, key=itemgetter(-1))
self.centroids_ = trials[0][0]
self._y_train_pred = trials[0][1]
return self
def do(self, a, b):
d = linalg.det(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev))
def do(self, a, b):
d = linalg.det(a)
if asarray(a).dtype.type in (single, double):
ad = asarray(a).astype(double)
else:
ad = asarray(a).astype(cdouble)
ev = linalg.eigvals(ad)
assert_almost_equal(d, multiply.reduce(ev))
def test_multiply(self):
from numpy import array, multiply, arange
a = array([-5.0, -0.0, 1.0])
b = array([3.0, -2.0, -3.0])
c = multiply(a, b)
for i in range(3):
assert c[i] == a[i] * b[i]
a = arange(15).reshape(5, 3)
assert (multiply.reduce(a) == array([0, 3640, 12320])).all()
def test_multiply(self):
from numpy import array, multiply, arange
a = array([-5.0, -0.0, 1.0])
b = array([ 3.0, -2.0,-3.0])
c = multiply(a, b)
for i in range(3):
assert c[i] == a[i] * b[i]
a = arange(15).reshape(5, 3)
assert(multiply.reduce(a) == array([0, 3640, 12320])).all()
def testpattern(self):
mm = MultiMethod()
@defmethod(mm, "lambda x : x>0")
def meth(x):
return x * mm(x-1)
@defmethod(mm, "anytype")
def meth(x):
return 1
self.failUnlessEqual(mm(0), 1)
self.failUnlessEqual(mm(1), 1)
self.failUnlessEqual(mm(2), 2)
self.failUnlessEqual(mm(4), 4*3*2)
from numpy import multiply
self.failUnlessEqual(mm(10), multiply.reduce(range(1,11)))
def mestimate(option):
"""
Use the M-estimate to estimate P(a_i|v_j); the probability of an attribute a_i given the classification v_j out of set of possible classifications V
M-estimate from http://www.inf.u-szeged.hu/~ormandi/ai2/06-naiveBayes-example.pdf
"""
# all of the records which have the classifier as option
total = self.data[self.data[self.result_column] == option]
total_num = len(total)
#arbitrary equivalent sample size
m = 3
#calculate the m-estimate. *1.0 to ensure a decimal always returned
mest = lambda nc: 1.0*(nc+m*p)/(total_num + m)
attr_estimates = fromiter(( mest(len(total[total[k]==v])) for k,v in attrs.items()),float ,count=len(attrs))
return p*multiply.reduce(attr_estimates)
def do(self, a, b):
d = gula.det(a)
s, ld = gula.slogdet(a)
assert_almost_equal(s * np.exp(ld), d)
if np.csingle == a.dtype.type or np.single == a.dtype.type:
cmp_type=np.csingle
else:
cmp_type=np.cdouble
ev = gula.eigvals(a.astype(cmp_type))
assert_almost_equal(d.astype(cmp_type),
multiply.reduce(ev.astype(cmp_type),
axis=(ev.ndim-1)))
if s != 0:
assert_almost_equal(np.abs(s), 1)
else:
assert_equal(ld, -inf)
def __init__(self, seq, parents_0, parents_1, this_index, weight):
"""
@param seq: the data sequence
@type seq: numpy array
@param parents_0: index of parent value in previous slice
5B
@type parents_0: int
@param parents_1: index of parent value in current slice
@type parents_1: int
@param this_index: index of node value itself in current slice
@type this_index: int
@param weight: sequence weight
@type weight: float in [0,1]
"""
# data
lng = len(seq)
self.lng = lng
self.seq = seq.flat
# self index
# Cache the indices
indices = []
# Step size
if len(seq.shape) == 1:
step = 1
else:
step = multiply.reduce(seq.shape[1:])
for l in range(0, lng):
index = []
if l > 0:
for pi in parents_0:
index.append((l - 1) * step + pi)
for pi in parents_1:
index.append(l * step + pi)
index.append(l * step + this_index)
indices.append(index)
self.this_index = this_index
self.step = step
self.indices = indices
self.weight = weight
def test_move_particle_one_over():
""" Check density is change by a particle hopping left or right. """
from numpy import nonzero, multiply
from numpy.random import randint
mc = MonteCarlo()
for i in range(
100): # Do this n times, to avoid issues with random numbers
# Create density
density = randint(50, size=randint(2, 6))
# Change it
new_density = mc.change_density(density)
# Make sure any movement is by one
indices = nonzero(density - new_density)[0]
assert_equal(len(indices), 2, "densities differ in two places")
assert_equal(multiply.reduce((density - new_density)[indices]), -1,
"densities differ by + and - 1")
def predict_proba(self, X):
if (self.X is None) or (self.y is None):
raise Exception("you must fit the model first")
from numpy import exp, pi, multiply, ndarray
"""ERROR CHECKING"""
assert all([(type(X) is ndarray),
(X.shape[-1] == self.X.shape[-1])]), "inconsistent array"
nd3D = exp(-(self.X - self._mus)**2 /
(self._sigmas**2 * 2)) * (1 / (self._sigmas *
((pi * 2)**0.5)))
nd2D = multiply.reduce(nd3D, axis=-1).T # P(X|C) likelihoods
mx = nd2D * self._probabilities_of_classes # the numerator for the Bayes formula (i.e. non-normalized probs)
ndVerticalArray = mx.sum(
axis=-1
)[:,
None] # marginal totals = priori probabilities = denominator for the Bayes formula (i.e. normalizer)
self._probabilities = mx / ndVerticalArray
return self._probabilities
def __init__(self, seq, parents_0, parents_1, this_index, weight):
"""
@param seq: the data sequence
@type seq: numpy array
@param parents_0: index of parent value in previous slice
5B
@type parents_0: int
@param parents_1: index of parent value in current slice
@type parents_1: int
@param this_index: index of node value itself in current slice
@type this_index: int
@param weight: sequence weight
@type weight: float in [0,1]
"""
# data
lng=len(seq)
self.lng=lng
self.seq=seq.flat
# self index
# Cache the indices
indices=[]
# Step size
if len(seq.shape)==1:
step=1
else:
step=multiply.reduce(seq.shape[1:])
for l in range(0, lng):
index=[]
if l>0:
for pi in parents_0:
index.append((l-1)*step+pi)
for pi in parents_1:
index.append(l*step+pi)
index.append(l*step+this_index)
indices.append(index)
self.this_index=this_index
self.step=step
self.indices=indices
self.weight=weight
def pagerank(H):
n = len(H)
for i in range(n):
H[i] = [x / float(sum(H[i])) for x in H[i]]
w = zeros(n)
rho = 1. / n * ones(n)
for i in range(n):
if multiply.reduce(H[i] == zeros(n)):
w[i] = 1
newH = H + outer(1. / n * w, ones(n))
theta = 0.85
G = theta * newH + (1 - theta) * outer(1. / n * ones(n), ones(n))
print 'initial pagerank scores:'
print rho
for j in range(10):
rho = dot(rho, G)
print 'pagerank scores after %d iteration:' % (j + 1)
print rho
def test_move_particle_one_over():
""" Check density is change by a particle hopping left or right. """
from numpy import nonzero, multiply
from numpy.random import randint
mc = MonteCarlo()
for i in range(100): # Do this n times, to avoid issues with random numbers
# Create density
density = randint(50, size=randint(2, 6))
# Change it
new_density = mc.change_density(density)
# Make sure any movement is by one
indices = nonzero(density - new_density)[0]
assert_equal(len(indices), 2, "densities differ in two places")
assert_equal(
multiply.reduce((density - new_density)[indices]),
-1,
"densities differ by + and - 1"
)
def probabilistic_and(p, axis=0):
"""
Probabilistic version of AND
"""
from numpy import array, multiply
return multiply.reduce(array(p), axis=axis)
def makepattern(smallprimes):
pattern = ones(multiply.reduce(smallprimes), dtype=bool_)
pattern[0] = 0
for p in smallprimes:
pattern[p::p] = 0
return pattern
def probabilistic_and(p, axis=0):
"""
Probabilistic version of AND
"""
from numpy import array, multiply
return multiply.reduce(array(p), axis=axis)
array([[1, 0, 0, 0],
[1, 0, 0, 0],
[1, 0, 0, 0],
[1, 0, 0, 0]]),
array([[1, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1]]),
array([[0, 0, 0, 1],
[0, 0, 1, 0],
[0, 1, 0, 0],
[1, 0, 0, 0]])
]
results = []
x, y = data.shape
for mask in masks:
xm, ym = mask.shape
for i in range(x-xm+1):
for j in range(y-ym+1):
tmp = (mask * data[i:i+xm, j:j+ym]).ravel()
results.append(multiply.reduce(tmp[tmp > 0]))
print(max(results))
