diff --git a/examples/animation.py b/examples/animation.py
index 9f7170c787c612427bfe08c6dc8a770cbbf13230..cd591c401db5dff19604dbd352192503950b02ed 100644
--- a/examples/animation.py
+++ b/examples/animation.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import os
import numpy as np
import numpy.random as npr
@@ -7,7 +8,6 @@ plt.ion()
npr.seed(0)
import pyhsmm
-from pyhsmm.util.text import progprint_xrange
###############
# load data #
@@ -32,7 +32,7 @@ obs_hypparams = {'mu_0':np.zeros(obs_dim),
# instantiate a Sticky-HDP-HMM
obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams)
- for state in xrange(Nmax)]
+ for state in range(Nmax)]
model = pyhsmm.models.WeakLimitStickyHDPHMM(
kappa=50.,alpha=6.,gamma=6.,init_state_concentration=1.,
obs_distns=obs_distns)
diff --git a/examples/hmm-EM.py b/examples/hmm-EM.py
index 08b21d6f29031752fe59458056cfd5b053ae8eef..9294de7ffccc5553e2e8db04e5ae829e0ee53ebc 100644
--- a/examples/hmm-EM.py
+++ b/examples/hmm-EM.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import numpy as np
np.seterr(divide='ignore') # these warnings are usually harmless for this code
from matplotlib import pyplot as plt
@@ -24,7 +25,7 @@ obs_hypparams = {'mu_0':np.zeros(obs_dim),
'kappa_0':0.25,
'nu_0':obs_dim+2}
-obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in xrange(N)]
+obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in range(N)]
# Build the HMM model that will represent the fitmodel
fitmodel = pyhsmm.models.HMM(
@@ -32,7 +33,7 @@ fitmodel = pyhsmm.models.HMM(
obs_distns=obs_distns)
fitmodel.add_data(data)
-print 'Gibbs sampling for initialization'
+print('Gibbs sampling for initialization')
for idx in progprint_xrange(25):
fitmodel.resample_model()
@@ -41,7 +42,7 @@ plt.figure()
fitmodel.plot()
plt.gcf().suptitle('Gibbs-sampled initialization')
-print 'EM'
+print('EM')
likes = fitmodel.EM_fit()
diff --git a/examples/hmm.py b/examples/hmm.py
index 5df1ac6f060db6e07846ff94227cac6363345da4..e15810ee66bbad617d79113fb102bac8e5910ffa 100644
--- a/examples/hmm.py
+++ b/examples/hmm.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import numpy as np
np.seterr(divide='ignore') # these warnings are usually harmless for this code
from matplotlib import pyplot as plt
@@ -9,8 +10,7 @@ matplotlib.rcParams['font.size'] = 8
import pyhsmm
from pyhsmm.util.text import progprint_xrange
-print \
-'''
+print('''
This demo shows how HDP-HMMs can fail when the underlying data has state
persistence without some kind of temporal regularization (in the form of a
sticky bias or duration modeling): without setting the number of states to be
@@ -18,8 +18,7 @@ the correct number a priori, lots of extra states can be intsantiated.
BUT the effect is much more relevant on real data (when the data doesn't exactly
fit the model). Maybe this demo should use multinomial emissions...
-'''
-
+''')
###############
# load data #
###############
@@ -42,7 +41,7 @@ obs_hypparams = {'mu_0':np.zeros(obs_dim),
### HDP-HMM without the sticky bias
-obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in xrange(Nmax)]
+obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in range(Nmax)]
posteriormodel = pyhsmm.models.WeakLimitHDPHMM(alpha=6.,gamma=6.,init_state_concentration=1.,
obs_distns=obs_distns)
posteriormodel.add_data(data)
@@ -55,7 +54,7 @@ plt.gcf().suptitle('HDP-HMM sampled model after 100 iterations')
### Sticky-HDP-HMM
-obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in xrange(Nmax)]
+obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in range(Nmax)]
posteriormodel = pyhsmm.models.WeakLimitStickyHDPHMM(
kappa=50.,alpha=6.,gamma=6.,init_state_concentration=1.,
obs_distns=obs_distns)
diff --git a/examples/hsmm.py b/examples/hsmm.py
index b3834e7c5c40073bf5aab80844c3476246f5632e..f39149feb788179a5240f2d7fc745a9a99ea640b 100644
--- a/examples/hsmm.py
+++ b/examples/hsmm.py
@@ -1,4 +1,4 @@
-from __future__ import division
+
import numpy as np
np.seterr(divide='ignore') # these warnings are usually harmless for this code
from matplotlib import pyplot as plt
@@ -9,12 +9,11 @@ from pyhsmm.util.text import progprint_xrange
SAVE_FIGURES = False
-print \
-'''
+print('''
This demo shows the HDP-HSMM in action. Its iterations are slower than those for
the (Sticky-)HDP-HMM, but explicit duration modeling can be a big advantage for
conditioning the prior or for discovering structure in data.
-'''
+''')
###############
# load data #
diff --git a/examples/svi.py b/examples/svi.py
index 6da60d51efd5447270287a41f3b5b0ff2bdcab75..2547ae513ee6f5986a8269f37b8628ab1943061a 100644
--- a/examples/svi.py
+++ b/examples/svi.py
@@ -1,4 +1,4 @@
-from __future__ import division
+
import numpy as np
from numpy import newaxis as na
from matplotlib import pyplot as plt
@@ -14,19 +14,19 @@ datapath = str(join(dirname(__file__),'svi_data.gz'))
### load data
if not isfile(datapath):
- print 'download svi_data.gz data and put it in examples/'
- print 'https://github.com/mattjj/example_data'
+ print('download svi_data.gz data and put it in examples/')
+ print('https://github.com/mattjj/example_data')
import sys; sys.exit(1)
-print 'loading data...'
+print('loading data...')
alldata = np.loadtxt(datapath)
allseqs = np.array_split(alldata,250)
datas, heldout = hold_out(allseqs,0.05)
training_size = sum(data.shape[0] for data in datas)
-print '...done!'
+print('...done!')
-print '%d total frames' % sum(data.shape[0] for data in alldata)
-print 'split into %d training and %d test sequences' % (len(datas),len(heldout))
+print('%d total frames' % sum(data.shape[0] for data in alldata))
+print('split into %d training and %d test sequences' % (len(datas),len(heldout)))
### inference!
diff --git a/pyhsmm/basic/__init__.py b/pyhsmm/basic/__init__.py
index b379a33448ec094a1c03a754dde20e03abe2278d..bc4f9b7a5129e896dcb8e796b46a2d4c917c061f 100644
--- a/pyhsmm/basic/__init__.py
+++ b/pyhsmm/basic/__init__.py
@@ -1,3 +1,3 @@
-import models
-import distributions
-import abstractions
+from . import models
+from . import distributions
+from . import abstractions
diff --git a/pyhsmm/basic/abstractions.py b/pyhsmm/basic/abstractions.py
index 58a53dfa0723166843228ac4e478e1993896846a..ab8efaf9294269321f42b07d71a13f89ed5b1f71 100644
--- a/pyhsmm/basic/abstractions.py
+++ b/pyhsmm/basic/abstractions.py
@@ -1,4 +1,6 @@
from __future__ import division
+from future.utils import with_metaclass
+
import abc
import numpy as np
from matplotlib import pyplot as plt
@@ -8,8 +10,7 @@ import pyhsmm
from pyhsmm.util.stats import flattendata, sample_discrete, sample_discrete_from_log, combinedata
from pyhsmm.util.general import rcumsum
-class DurationDistribution(Distribution):
- __metaclass__ = abc.ABCMeta
+class DurationDistribution(with_metaclass(abc.ABCMeta, Distribution)):
# in addition to the methods required by Distribution, we also require a
# log_sf implementation
diff --git a/pyhsmm/internals/hmm_states.py b/pyhsmm/internals/hmm_states.py
index 328ac8ae75904533ca5d14ae16bde37e2fa64bd1..30ce024e7b56aca929c29118e62ff752d4d3b1cd 100644
--- a/pyhsmm/internals/hmm_states.py
+++ b/pyhsmm/internals/hmm_states.py
@@ -1,4 +1,6 @@
from __future__ import division
+from builtins import range
+from future.utils import with_metaclass
import numpy as np
from numpy import newaxis as na
import abc
@@ -16,8 +18,7 @@ from pyhsmm.util.general import rle
# Mixins and bases #
######################
-class _StatesBase(object):
- __metaclass__ = abc.ABCMeta
+class _StatesBase(with_metaclass(abc.ABCMeta, object)):
def __init__(self,model,T=None,data=None,stateseq=None,
generate=True,initialize_from_prior=True, fixed_stateseq=False):
@@ -141,7 +142,7 @@ class _SeparateTransMixin(object):
class _PossibleChangepointsMixin(object):
def __init__(self,model,data,changepoints=None,**kwargs):
changepoints = changepoints if changepoints is not None \
- else [(t,t+1) for t in xrange(data.shape[0])]
+ else [(t,t+1) for t in range(data.shape[0])]
self.changepoints = changepoints
self.segmentstarts = np.array([start for start,stop in changepoints],dtype=np.int32)
@@ -227,7 +228,7 @@ class HMMStatesPython(_StatesBase):
A = self.trans_matrix
stateseq = np.zeros(T,dtype=np.int32)
- for idx in xrange(T):
+ for idx in range(T):
stateseq[idx] = sample_discrete(nextstate_distn)
nextstate_distn = A[stateseq[idx]]
@@ -257,7 +258,7 @@ class HMMStatesPython(_StatesBase):
betal = np.zeros_like(aBl)
- for t in xrange(betal.shape[0]-2,-1,-1):
+ for t in range(betal.shape[0]-2,-1,-1):
betal[t] = logsumexp(Al + betal[t+1] + aBl[t+1],axis=1)
np.seterr(**errs)
@@ -278,7 +279,7 @@ class HMMStatesPython(_StatesBase):
alphal = np.zeros_like(aBl)
alphal[0] = np.log(init_state_distn) + aBl[0]
- for t in xrange(alphal.shape[0]-1):
+ for t in range(alphal.shape[0]-1):
alphal[t+1] = logsumexp(alphal[t] + Al.T,axis=1) + aBl[t+1]
np.seterr(**errs)
@@ -300,7 +301,7 @@ class HMMStatesPython(_StatesBase):
logtot = 0.
betan[-1] = 1.
- for t in xrange(T-2,-1,-1):
+ for t in range(T-2,-1,-1):
cmax = aBl[t+1].max()
betan[t] = A.dot(betan[t+1] * np.exp(aBl[t+1] - cmax))
norm = betan[t].sum()
@@ -327,7 +328,7 @@ class HMMStatesPython(_StatesBase):
logtot = 0.
in_potential = init_state_distn
- for t in xrange(T):
+ for t in range(T):
cmax = aBl[t].max()
alphan[t] = in_potential * np.exp(aBl[t] - cmax)
norm = alphan[t].sum()
@@ -369,7 +370,7 @@ class HMMStatesPython(_StatesBase):
stateseq = np.empty(T,dtype=np.int32)
nextstate_unsmoothed = init_state_distn
- for idx in xrange(T):
+ for idx in range(T):
logdomain = betal[idx] + aBl[idx]
logdomain[nextstate_unsmoothed == 0] = -np.inf
if np.any(np.isfinite(logdomain)):
@@ -392,7 +393,7 @@ class HMMStatesPython(_StatesBase):
stateseq = np.empty(T,dtype=np.int32)
nextstate_unsmoothed = init_state_distn
- for idx in xrange(T):
+ for idx in range(T):
logdomain = aBl[idx]
logdomain[nextstate_unsmoothed == 0] = -np.inf
stateseq[idx] = sample_discrete(nextstate_unsmoothed * betan * np.exp(logdomain - np.amax(logdomain)))
@@ -412,7 +413,7 @@ class HMMStatesPython(_StatesBase):
stateseq = np.empty(T,dtype=np.int32)
next_potential = np.ones(AT.shape[0])
- for t in xrange(T-1,-1,-1):
+ for t in range(T-1,-1,-1):
stateseq[t] = sample_discrete(next_potential * alphan[t])
next_potential = AT[stateseq[t]]
@@ -573,7 +574,7 @@ class HMMStatesPython(_StatesBase):
scores = np.zeros_like(aBl)
args = np.zeros(aBl.shape,dtype=np.int32)
- for t in xrange(scores.shape[0]-2,-1,-1):
+ for t in range(scores.shape[0]-2,-1,-1):
vals = Al + scores[t+1] + aBl[t+1]
vals.argmax(axis=1,out=args[t+1])
vals.max(axis=1,out=scores[t])
@@ -588,7 +589,7 @@ class HMMStatesPython(_StatesBase):
stateseq = np.empty(T,dtype=np.int32)
stateseq[0] = (scores[0] + np.log(init_state_distn) + aBl[0]).argmax()
- for idx in xrange(1,T):
+ for idx in range(1,T):
stateseq[idx] = args[idx,stateseq[idx-1]]
return stateseq
diff --git a/pyhsmm/internals/hsmm_inb_states.py b/pyhsmm/internals/hsmm_inb_states.py
index 428051f2a31807272f66bc2b27ffe6ff2bd82980..e1c6ce2be3c5fab769dec4d6b75e7bfee124dc4b 100644
--- a/pyhsmm/internals/hsmm_inb_states.py
+++ b/pyhsmm/internals/hsmm_inb_states.py
@@ -1,4 +1,6 @@
from __future__ import division
+from builtins import zip, range
+from future.utils import with_metaclass
import numpy as np
import abc
import scipy.stats as stats
@@ -11,16 +13,15 @@ except ImportError:
from ..util.stats import sample_markov
from ..util.general import top_eigenvector, cumsum
-from hmm_states import HMMStatesPython, HMMStatesEigen, _SeparateTransMixin
-from hsmm_states import HSMMStatesEigen
+from .hmm_states import HMMStatesPython, HMMStatesEigen, _SeparateTransMixin
+from .hsmm_states import HSMMStatesEigen
# TODO these classes are currently backed by HMM message passing, but they can
# be made much more time and memory efficient. i have the code to do it in some
# other branches, but dense matrix multiplies are actually competitive.
-class _HSMMStatesIntegerNegativeBinomialBase(HSMMStatesEigen, HMMStatesEigen):
- __metaclass__ = abc.ABCMeta
+class _HSMMStatesIntegerNegativeBinomialBase(with_metaclass(abc.ABCMeta, HSMMStatesEigen, HMMStatesEigen)):
@property
def rs(self):
@@ -169,7 +170,7 @@ class HSMMStatesIntegerNegativeBinomial(_HSMMStatesIntegerNegativeBinomialBase):
pmess = np.zeros(ends[-1])
# betal[-1] is 0
- for t in xrange(T-1,-1,-1):
+ for t in range(T-1,-1,-1):
pmess += self.hmm_aBl[t]
betastarl[t] = logsumexp(np.log(foo) + pmess, axis=1)
betal[t-1] = logsumexp(Al + betastarl[t], axis=1)
@@ -196,7 +197,7 @@ class HSMMStatesIntegerNegativeBinomial(_HSMMStatesIntegerNegativeBinomialBase):
self.expected_durations = np.zeros((self.num_states,self.T))
eye = np.eye(self.num_states)/num_r_samples
- for i in xrange(num_r_samples):
+ for i in range(num_r_samples):
self.model._resample_from_mf()
self.clear_caches()
@@ -206,7 +207,7 @@ class HSMMStatesIntegerNegativeBinomial(_HSMMStatesIntegerNegativeBinomialBase):
self.expected_transcounts += \
count_transitions(self.stateseq_norep,minlength=self.num_states)\
/ num_r_samples
- for state in xrange(self.num_states):
+ for state in range(self.num_states):
self.expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
@@ -225,7 +226,7 @@ class HSMMStatesIntegerNegativeBinomial(_HSMMStatesIntegerNegativeBinomialBase):
mf_aBl = self.mf_aBl
- for i in xrange(num_r_samples):
+ for i in range(num_r_samples):
for d in self.dur_distns:
d._resample_r_from_mf()
self.clear_caches()
@@ -248,9 +249,9 @@ class HSMMStatesIntegerNegativeBinomial(_HSMMStatesIntegerNegativeBinomialBase):
self.expected_transcounts += expected_transcounts / num_r_samples
# collect duration statistics by sampling from messages
- for j in xrange(num_stateseq_samples_per_r):
+ for j in range(num_stateseq_samples_per_r):
self._resample_from_mf(trans,init,aBl,hmm_alphal,hmm_betal)
- for state in xrange(self.num_states):
+ for state in range(self.num_states):
self.expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
@@ -445,7 +446,7 @@ class HSMMStatesTruncatedIntegerNegativeBinomial(HSMMStatesDelayedIntegerNegativ
def bwd_enter_rows(self):
As = [np.diag(np.repeat(p,r)) + np.diag(np.repeat(1-p,r-1),k=1) for r,p in zip(self.rs,self.ps)]
enters = [stats.binom.pmf(np.arange(r)[::-1],r-1,p) for A,r,p in zip(As,self.rs,self.ps)]
- # norms = [sum(v.dot(np.linalg.matrix_power(A,d))[-1]*(1-p) for d in xrange(delay))
+ # norms = [sum(v.dot(np.linalg.matrix_power(A,d))[-1]*(1-p) for d in range(delay))
# for A,v,p,delay in zip(As,enters,self.ps,self.delays)]
# enters = [v.dot(np.linalg.matrix_power(A,self.delays[state])) / (1.-norm)
enters = [v.dot(np.linalg.matrix_power(A,self.delays[state]))
diff --git a/pyhsmm/internals/hsmm_states.py b/pyhsmm/internals/hsmm_states.py
index c9663d2fdbca85dec952a9d4806ed6f7b45d0072..a4b4108bb7f6583523e9e2a2869d2bc23503c53b 100644
--- a/pyhsmm/internals/hsmm_states.py
+++ b/pyhsmm/internals/hsmm_states.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range, map
import numpy as np
from numpy import newaxis as na
from scipy.misc import logsumexp
@@ -6,8 +7,8 @@ from scipy.misc import logsumexp
from pyhsmm.util.stats import sample_discrete
from pyhsmm.util.general import rle, rcumsum, cumsum
-import hmm_states
-from hmm_states import _StatesBase, _SeparateTransMixin, \
+from . import hmm_states
+from .hmm_states import _StatesBase, _SeparateTransMixin, \
HMMStatesPython, HMMStatesEigen
@@ -398,7 +399,7 @@ class HSMMStatesPython(_StatesBase):
self.expected_durations = expected_durations = \
np.zeros((self.num_states,self.T))
- for state in xrange(self.num_states):
+ for state in range(self.num_states):
expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
@@ -476,11 +477,11 @@ class HSMMStatesPython(_StatesBase):
dur_potentials,cumulative_obs_potentials,
alphastarl,betal,normalizer):
if self.trunc is not None:
- raise NotImplementedError, "_expected_durations can't handle trunc"
+ raise NotImplementedError("_expected_durations can't handle trunc")
T = self.T
logpmfs = -np.inf*np.ones_like(alphastarl)
errs = np.seterr(invalid='ignore')
- for t in xrange(T):
+ for t in range(T):
cB, offset = cumulative_obs_potentials(t)
np.logaddexp(dur_potentials(t) + alphastarl[t] + betal[t:] +
cB - (normalizer + offset),
@@ -685,7 +686,7 @@ class _PossibleChangepointsMixin(hmm_states._PossibleChangepointsMixin,HSMMState
self.expected_durations = expected_durations = \
np.zeros((self.num_states,self.Tfull))
- for state in xrange(self.num_states):
+ for state in range(self.num_states):
expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
@@ -872,7 +873,7 @@ class HSMMStatesPossibleChangepoints(_PossibleChangepointsMixin,HSMMStatesPython
logpmfs = -np.inf*np.ones((self.Tfull,alphastarl.shape[1]))
errs = np.seterr(invalid='ignore') # logaddexp(-inf,-inf)
# TODO censoring not handled correctly here
- for tblock in xrange(self.Tblock):
+ for tblock in range(self.Tblock):
possible_durations = self.segmentlens[tblock:].cumsum()[:self.trunc]
cB, offset = cumulative_obs_potentials(tblock)
logpmfs[possible_durations -1] = np.logaddexp(
@@ -991,7 +992,7 @@ def hsmm_messages_backwards_log(
T, _ = betal.shape
betal[-1] = 0.
- for t in xrange(T-1,-1,-1):
+ for t in range(T-1,-1,-1):
cB, offset = cumulative_obs_potentials(t)
dp = dur_potentials(t)
betastarl[t] = logsumexp(
@@ -1020,7 +1021,7 @@ def hsmm_messages_forwards_log(
T, _ = alphal.shape
alphastarl[0] = initial_state_potential
- for t in xrange(T-1):
+ for t in range(T-1):
cB = reverse_cumulative_obs_potentials(t)
alphal[t] = logsumexp(
alphastarl[t+1-cB.shape[0]:t+1] + cB + reverse_dur_potentials(t), axis=0)
@@ -1098,7 +1099,7 @@ def hsmm_maximizing_assignment(
betastar_scores, betastar_args = np.empty((T,N)), np.empty((T,N),dtype=np.int)
beta_scores[-1] = 0.
- for t in xrange(T-1,-1,-1):
+ for t in range(T-1,-1,-1):
cB, offset = cumulative_obs_potentials(t)
vals = beta_scores[t:t+cB.shape[0]] + cB + dur_potentials(t)
diff --git a/pyhsmm/internals/transitions.py b/pyhsmm/internals/transitions.py
index 8ee8d6b2767dd55b9e108b1079a483ace13a8016..5921c66dc2f23f024e453ced5656c778aeb928d0 100644
--- a/pyhsmm/internals/transitions.py
+++ b/pyhsmm/internals/transitions.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import numpy as np
from numpy import newaxis as na
np.seterr(invalid='raise')
@@ -35,7 +36,7 @@ class _HMMTransitionsBase(object):
weights=row) for row in trans_matrix]
elif None not in (alpha,self.N) or alphav is not None:
self._row_distns = [Multinomial(alpha_0=alpha,K=self.N,alphav_0=alphav)
- for n in xrange(self.N)] # sample from prior
+ for n in range(self.N)] # sample from prior
@property
def trans_matrix(self):
diff --git a/pyhsmm/models.py b/pyhsmm/models.py
index eb7c4fbd957ecb679f1c5ce1094518b86e6718f7..1de2dd1fb37718ae2944c7267c2535a44d947d9b 100644
--- a/pyhsmm/models.py
+++ b/pyhsmm/models.py
@@ -1,4 +1,9 @@
from __future__ import division
+from future.utils import iteritems, itervalues
+from builtins import map
+
+import sys
+
import numpy as np
import itertools
import collections
@@ -81,7 +86,7 @@ class _HMMBase(Model):
# generating brand new data sequence
counts = np.bincount(s.stateseq,minlength=self.num_states)
obs = [iter(o.rvs(count)) for o, count in zip(s.obs_distns,counts)]
- s.data = np.squeeze(np.vstack([obs[state].next() for state in s.stateseq]))
+ s.data = np.squeeze(np.vstack([next(obs[state]) for state in s.stateseq]))
else:
# filling in missing data
data = s.data
@@ -89,7 +94,7 @@ class _HMMBase(Model):
counts = np.bincount(s.stateseq[nan_idx],minlength=self.num_states)
obs = [iter(o.rvs(count)) for o, count in zip(s.obs_distns,counts)]
for idx, state in zip(nan_idx, s.stateseq[nan_idx]):
- data[idx] = obs[state].next()
+ data[idx] = next(obs[state])
return s.data
@@ -141,7 +146,7 @@ class _HMMBase(Model):
prev_k = k
else:
from joblib import Parallel, delayed
- import parallel
+ from . import parallel
parallel.cmaxes = cmaxes
parallel.alphal = alphal
@@ -183,7 +188,8 @@ class _HMMBase(Model):
@property
def used_states(self):
'a list of the used states in the order they appear'
- canonical_ids = collections.defaultdict(itertools.count().next)
+ c = itertools.count()
+ canonical_ids = collections.defaultdict(c.__next__ if sys.version_info.major == 3 else c.next)
for s in self.states_list:
for state in s.stateseq:
canonical_ids[state]
@@ -471,7 +477,7 @@ class _HMMGibbsSampling(_HMMBase,ModelGibbsSampling):
def _joblib_resample_states(self,states_list,num_procs):
from joblib import Parallel, delayed
- import parallel
+ from . import parallel
# warn('joblib is segfaulting on OS X only, not sure why')
@@ -556,7 +562,7 @@ class _HMMMeanField(_HMMBase,ModelMeanField):
def _joblib_meanfield_update_states(self,states_list,num_procs):
if len(states_list) > 0:
from joblib import Parallel, delayed
- import parallel
+ from . import parallel
joblib_args = list_split(
[self._get_joblib_pair(s) for s in states_list],
@@ -1261,13 +1267,13 @@ class _SeparateTransMixin(object):
### Gibbs sampling
def resample_trans_distn(self):
- for group_id, trans_distn in self.trans_distns.iteritems():
+ for group_id, trans_distn in iteritems(self.trans_distns):
trans_distn.resample([s.stateseq for s in self.states_list
if hash(s.group_id) == hash(group_id)])
self._clear_caches()
def resample_init_state_distn(self):
- for group_id, init_state_distn in self.init_state_distns.iteritems():
+ for group_id, init_state_distn in iteritems(self.init_state_distns):
init_state_distn.resample([s.stateseq[0] for s in self.states_list
if hash(s.group_id) == hash(group_id)])
self._clear_caches()
@@ -1275,13 +1281,13 @@ class _SeparateTransMixin(object):
### Mean field
def meanfield_update_trans_distn(self):
- for group_id, trans_distn in self.trans_distns.iteritems():
+ for group_id, trans_distn in iteritems(self.trans_distns):
states_list = [s for s in self.states_list if hash(s.group_id) == hash(group_id)]
if len(states_list) > 0:
trans_distn.meanfieldupdate([s.expected_transcounts for s in states_list])
def meanfield_update_init_state_distn(self):
- for group_id, init_state_distn in self.init_state_distns.iteritems():
+ for group_id, init_state_distn in iteritems(self.init_state_distns):
states_list = [s for s in self.states_list if hash(s.group_id) == hash(group_id)]
if len(states_list) > 0:
init_state_distn.meanfieldupdate([s.expected_states[0] for s in states_list])
@@ -1290,23 +1296,23 @@ class _SeparateTransMixin(object):
vlb = 0.
vlb += sum(s.get_vlb() for s in self.states_list)
vlb += sum(trans_distn.get_vlb()
- for trans_distn in self.trans_distns.itervalues())
+ for trans_distn in itervalues(self.trans_distns))
vlb += sum(init_state_distn.get_vlb()
- for init_state_distn in self.init_state_distns.itervalues())
+ for init_state_distn in itervalues(self.init_state_distns))
vlb += sum(o.get_vlb() for o in self.obs_distns)
return vlb
### SVI
def _meanfield_sgdstep_trans_distn(self,mb_states_list,prob,stepsize):
- for group_id, trans_distn in self.trans_distns.iteritems():
+ for group_id, trans_distn in iteritems(self.trans_distns):
trans_distn.meanfield_sgdstep(
[s.expected_transcounts for s in mb_states_list
if hash(s.group_id) == hash(group_id)],
prob,stepsize)
def _meanfield_sgdstep_init_state_distn(self,mb_states_list,prob,stepsize):
- for group_id, init_state_distn in self.init_state_distns.iteritems():
+ for group_id, init_state_distn in iteritems(self.init_state_distns):
init_state_distn.meanfield_sgdstep(
[s.expected_states[0] for s in mb_states_list
if hash(s.group_id) == hash(group_id)],
diff --git a/pyhsmm/util/__init__.py b/pyhsmm/util/__init__.py
index f09c6507a2c62658374a632c694ae0088d967ad5..2f3d09d0213d443ea9a69fd1bef2c43ad83e2b37 100644
--- a/pyhsmm/util/__init__.py
+++ b/pyhsmm/util/__init__.py
@@ -1,2 +1,2 @@
__all__ = ['general','plot','stats','text']
-import general, plot, stats, text
+from . import general, plot, stats, text
diff --git a/pyhsmm/util/general.py b/pyhsmm/util/general.py
index ac85b7f577b26db88dbfe1bc860225ce80a2f3a0..4b354a50f5c320686371c4a85da310cf5d7d8dde 100644
--- a/pyhsmm/util/general.py
+++ b/pyhsmm/util/general.py
@@ -1,11 +1,14 @@
from __future__ import division
+from builtins import range, map, zip, filter
+
import numpy as np
from numpy.lib.stride_tricks import as_strided as ast
import scipy.linalg
import copy, collections, os, shutil, hashlib
from contextlib import closing
-from urllib2 import urlopen
-from itertools import izip, chain, count, ifilter
+from six.moves.urllib.request import urlopen
+from itertools import chain, count
+from functools import reduce
def solve_psd(A,b,chol=None,overwrite_b=False,overwrite_A=False):
if A.shape[0] < 5000 and chol is None:
@@ -145,7 +148,7 @@ def _sieve(stream):
# just for fun; doesn't work over a few hundred
val = stream.next()
yield val
- for x in ifilter(lambda x: x%val != 0, _sieve(stream)):
+ for x in filter(lambda x: x%val != 0, _sieve(stream)):
yield x
def primes():
@@ -172,7 +175,7 @@ def top_eigenvector(A,niter=1000,force_iteration=False):
else:
x1 = np.repeat(1./n,n)
x2 = x1.copy()
- for itr in xrange(niter):
+ for itr in range(niter):
np.dot(A.T,x1,out=x2)
x2 /= x2.sum()
x1,x2 = x2,x1
@@ -203,7 +206,7 @@ def count_transitions(stateseq,minlength=None):
if minlength is None:
minlength = stateseq.max() + 1
out = np.zeros((minlength,minlength),dtype=np.int32)
- for a,b in izip(stateseq[:-1],stateseq[1:]):
+ for a,b in zip(stateseq[:-1],stateseq[1:]):
out[a,b] += 1
return out
@@ -223,14 +226,14 @@ def hold_out(datalist,frac):
def sgd_passes(tau,kappa,datalist,minibatchsize=1,npasses=1):
N = len(datalist)
- for superitr in xrange(npasses):
+ for superitr in range(npasses):
if minibatchsize == 1:
perm = np.random.permutation(N)
- for idx, rho_t in izip(perm,sgd_steps(tau,kappa)):
+ for idx, rho_t in zip(perm,sgd_steps(tau,kappa)):
yield datalist[idx], rho_t
else:
minibatch_indices = np.array_split(np.random.permutation(N),N/minibatchsize)
- for indices, rho_t in izip(minibatch_indices,sgd_steps(tau,kappa)):
+ for indices, rho_t in zip(minibatch_indices,sgd_steps(tau,kappa)):
yield [datalist[idx] for idx in indices], rho_t
def sgd_sampling(tau,kappa,datalist,minibatchsize=1):
@@ -252,7 +255,7 @@ def minibatchsize(lst):
def random_subset(lst,sz):
perm = np.random.permutation(len(lst))
- return [lst[perm[idx]] for idx in xrange(sz)]
+ return [lst[perm[idx]] for idx in range(sz)]
def get_file(remote_url,local_path):
if not os.path.isfile(local_path):
diff --git a/pyhsmm/util/profiling.py b/pyhsmm/util/profiling.py
index 019daf3e55eb4bb06bfb02f8378b36c2527dd639..8d20227f7a855a19f5b58fed9a06713d621e0f4e 100644
--- a/pyhsmm/util/profiling.py
+++ b/pyhsmm/util/profiling.py
@@ -1,6 +1,7 @@
-from __future__ import division
+from __future__ import division, print_function
import numpy as np
-import sys, StringIO, inspect, os, functools, time, collections
+import sys, inspect, os, functools, time, collections
+from future.utils import iteritems
### use @timed for really basic timing
@@ -21,15 +22,15 @@ def show_timings(stream=None):
if len(_timings) > 0:
results = [(inspect.getsourcefile(f),f.__name__,
len(vals),np.sum(vals),np.mean(vals),np.std(vals))
- for f, vals in _timings.iteritems()]
+ for f, vals in iteritems(_timings)]
filename_lens = max(len(filename) for filename, _, _, _, _, _ in results)
name_lens = max(len(name) for _, name, _, _, _, _ in results)
fmt = '{:>%d} {:>%d} {:>10} {:>10} {:>10} {:>10}' % (filename_lens, name_lens)
- print >>stream, fmt.format('file','name','ncalls','tottime','avg time','std dev')
+ print(fmt.format('file','name','ncalls','tottime','avg time','std dev'), file=stream)
fmt = '{:>%d} {:>%d} {:>10} {:>10.3} {:>10.3} {:>10.3}' % (filename_lens, name_lens)
- print >>stream, '\n'.join(fmt.format(*tup) for tup in sorted(results))
+ print('\n'.join(fmt.format(*tup) for tup in sorted(results)), file=stream)
### use @line_profiled for a thin wrapper around line_profiler
diff --git a/pyhsmm/util/stats.py b/pyhsmm/util/stats.py
index 30524142ad36442e288907911cd9cfc741d33966..ae7922ae37345d96fc252d545ddf359ae6d3ea9e 100644
--- a/pyhsmm/util/stats.py
+++ b/pyhsmm/util/stats.py
@@ -1,5 +1,5 @@
from __future__ import division
-from itertools import izip
+from builtins import zip, range
import numpy as np
from numpy.random import random
na = np.newaxis
@@ -107,7 +107,7 @@ def diag_whiten(datalist):
def count_transitions(stateseq, num_states):
out = np.zeros((num_states,num_states),dtype=np.int32)
- for i,j in izip(stateseq[:-1],stateseq[1:]):
+ for i,j in zip(stateseq[:-1],stateseq[1:]):
out[i,j] += 1
return out
@@ -227,7 +227,7 @@ def sample_crp_tablecounts(concentration,customers,colweights):
for (i,j), n in np.ndenumerate(customers):
w = colweights[j]
- for k in xrange(n):
+ for k in range(n):
m[i,j] += randseq[starts[i,j]+k] \
< (concentration * w) / (k + concentration * w)
diff --git a/pyhsmm/util/text.py b/pyhsmm/util/text.py
index d219338ed3d1f528ed1791d604ff6753fa425084..b92e4511d2fc654246ef4a436c46b4ab6ee01aae 100644
--- a/pyhsmm/util/text.py
+++ b/pyhsmm/util/text.py
@@ -1,3 +1,4 @@
+from builtins import range
import numpy as np
import sys, time
@@ -20,7 +21,7 @@ def sec2str(seconds):
return '%0.2f' % seconds
def progprint_xrange(*args,**kwargs):
- xr = xrange(*args)
+ xr = range(*args)
return progprint(xr,total=len(xr),**kwargs)
def progprint(iterator,total=None,perline=25,show_times=True):
@@ -50,8 +51,8 @@ def progprint(iterator,total=None,perline=25,show_times=True):
sys.stdout.write(' [ %d ]\n' % (idx+1))
idx += 1
sys.stdout.flush()
- print ''
+ print('')
if show_times and len(times) > 0:
total = sec2str(seconds=np.sum(times))
- print '%7.2fsec avg, %s total\n' % (np.mean(times),total)
+ print('%7.2fsec avg, %s total\n' % (np.mean(times),total))
diff --git a/setup.py b/setup.py
index 39682e0c1c2d953b7352cc2737c32b9734c937cd..522690194b5d92200f530a85526c3c7ee4bee6a2 100644
--- a/setup.py
+++ b/setup.py
@@ -8,7 +8,7 @@ from warnings import warn
import os
import sys
from glob import glob
-from six.moves.urllib.request import urlretrieve
+from future.moves.urllib.request import urlretrieve
import tarfile
import shutil
@@ -109,8 +109,8 @@ setup(name='pyhsmm',
keywords=['bayesian', 'inference', 'mcmc', 'time-series', 'monte-carlo',
'variational inference', 'mean field', 'vb'],
install_requires=[
- "numpy", "scipy", "matplotlib", "nose", "pybasicbayes >= 0.1.3"],
- setup_requires=['numpy'],
+ "numpy", "scipy", "matplotlib", "nose", "pybasicbayes >= 0.1.3", "future"],
+ setup_requires=['numpy', "future"],
ext_modules=ext_modules,
classifiers=[
'Development Status :: 4 - Beta',
diff --git a/tests/test_hmm_geweke.py b/tests/test_hmm_geweke.py
index 31e6c3d12e32f01705cf9fe994b7426e326c1518..264d9939cc04bb0ee0e382faa7ce93335fc0d81b 100644
--- a/tests/test_hmm_geweke.py
+++ b/tests/test_hmm_geweke.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import numpy as np
from functools import wraps
from nose.plugins.attrib import attr
@@ -65,7 +66,7 @@ def discrete_geweke_test(fig):
# generate state sequences and parameters from the prior
prior_stateseqs = []
prior_weights = []
- for itr in xrange(num_iter):
+ for itr in range(num_iter):
hmm.resample_model() # sample parameters from the prior
_, stateseq = hmm.generate(T,keep=False)
prior_stateseqs.append(stateseq)
@@ -79,7 +80,7 @@ def discrete_geweke_test(fig):
gibbs_stateseqs = []
gibbs_weights = []
- for itr in xrange(num_iter):
+ for itr in range(num_iter):
s.data = None
hmm._generate_obs(s) # resamples data given state sequence, obs params
hmm.resample_model() # resamples everything else as usual
@@ -90,7 +91,7 @@ def discrete_geweke_test(fig):
# test that they look similar by checking probability of co-assignment
time_indices = np.arange(T)
- for itr in xrange(num_checks):
+ for itr in range(num_checks):
i,j = np.random.choice(time_indices,replace=False,size=2)
prior_prob_of_coassignment = (prior_stateseqs[:,i] == prior_stateseqs[:,j]).std()
gibbs_prob_of_coassignment = (gibbs_stateseqs[:,i] == gibbs_stateseqs[:,j]).std()
@@ -126,12 +127,12 @@ def discrete_geweke_multiple_seqs_test(fig):
obs_distns=obs_distns)
# generate state sequences and parameters from the prior
- prior_stateseqss = [[] for _ in xrange(num_seqs)]
+ prior_stateseqss = [[] for _ in range(num_seqs)]
prior_weights = []
- for itr in xrange(num_iter):
+ for itr in range(num_iter):
hmm.resample_model() # sample parameters from the prior
- for itr2 in xrange(num_seqs):
+ for itr2 in range(num_seqs):
_, stateseq = hmm.generate(T,keep=False)
prior_stateseqss[itr2].append(stateseq)
@@ -142,13 +143,13 @@ def discrete_geweke_multiple_seqs_test(fig):
prior_weights = np.array(prior_weights)
# generate state sequences and parameters using Gibbs
- for itr in xrange(num_seqs):
+ for itr in range(num_seqs):
hmm.generate(T,keep=True)
assert len(hmm.states_list) == num_seqs
- gibbs_stateseqss = [[] for _ in xrange(num_seqs)]
+ gibbs_stateseqss = [[] for _ in range(num_seqs)]
gibbs_weights = []
- for itr in xrange(num_iter):
+ for itr in range(num_iter):
for s in hmm.states_list:
s.data = None
hmm._generate_obs(s) # resamples data given state sequence, obs params
@@ -166,7 +167,7 @@ def discrete_geweke_multiple_seqs_test(fig):
# test that they look similar by checking probability of co-assignment
time_indices = np.arange(T)
seq_indices = np.arange(num_seqs)
- for itr in xrange(num_checks):
+ for itr in range(num_checks):
i,j = np.random.choice(time_indices,replace=False,size=2)
si,sj = np.random.choice(seq_indices,replace=True,size=2)
prior_prob_of_coassignment = \
diff --git a/tests/test_hmm_likelihood.py b/tests/test_hmm_likelihood.py
index 5ffb24894feb7897f8812d3f7d7967ba37ce2d34..58af94b5fd8da7161675cfa3db440247328a1eee 100644
--- a/tests/test_hmm_likelihood.py
+++ b/tests/test_hmm_likelihood.py
@@ -1,4 +1,5 @@
from __future__ import division
+from builtins import range
import numpy as np
from numpy import newaxis as na
from inspect import getargspec