class Levy(ExpDesign):
data_net_all = Corpus([
'manufacturing', 'fb_uc', 'blogs', 'emaileu', 'propro', 'euroroad',
'generator7', 'generator12', 'generator10', 'generator4'
])
net_all = data_net_all + Corpus(['clique6', 'BA'])
data_text_all = Corpus(['kos', 'nips12', 'nips', 'reuter50',
'20ngroups']) # lucene
#
# Poisson Point process :
# * stationarity / ergodicity of p(d_i) ?
# * Erny theorem (characterization by void probabilities ?
# * Inference ? Gamma Process ?
# * Sparsity ?
wmmsb = ExpTensor(
corpus=['BA'],
model='iwmmsb_scvb',
N=200,
chunk='adaptative_1',
K=6,
iterations=3,
hyper='auto',
testset_ratio=20,
delta=[[2, 2]],
chi_a=10,
tau_a=100,
kappa_a=0.6,
chi_b=10,
tau_b=500,
kappa_b=0.9,
h**o=0,
mask='unbalanced',
_data_format='w',
_data_type='networks',
_refdir='debug_scvb',
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}',
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _chi_a _tau_a _kappa_a _chi_b _tau_b _kappa_b _elbo _roc'
)
noelw3 = ExpGroup(wmmsb,
N='all',
chunk=['adaptative_0.1', 'adaptative_1'],
corpus=data_net_all,
mask=['unbalanced'],
_refdir='noel3')
def load(cls, expe, load=True):
""" Return the frontend suited for the given expe"""
corpus_name = expe.get('corpus') or expe.get('random')
if '.' in corpus_name:
c_split = corpus_name.split('.')
c_name, c_ext = '.'.join(c_split[:-1]), c_split[-1]
else:
c_name = corpus_name
c_ext = None
_corpus = Corpus.get(c_name)
if c_ext in GramExp._frontend_ext:
# graph-tool object
# @Todo: Corpus intragation!
_corpus.update(data_format=c_ext)
elif _corpus is False:
raise ValueError('Unknown Corpus `%s\'!' % corpus)
elif _corpus is None:
return None
if _corpus['data_type'] == 'text':
frontend = frontendText(expe)
elif _corpus['data_type'] == 'network':
if _corpus.get('data_format') == 'gt':
frontend = frontendNetwork_gt.from_expe(expe, load=load, corpus=_corpus)
else:
# Obsolete loading design. @Todo
frontend = frontendNetwork(expe)
if load is True:
frontend.load_data(randomize=False)
frontend.sample(expe.get('N'), randomize=False)
return frontend
def load(cls, expe, skip_init=False):
""" Return the frontend suited for the given expe
@TODO: skip_init is not implemented
"""
if skip_init:
cls.log.warning('skip init is not implemented')
corpus_name = expe.get('corpus') or expe.get('random') or expe.get(
'concept')
if expe.get('driver'):
corpus_name += '.' + expe.driver.strip('.')
if '.' in corpus_name:
c_split = corpus_name.split('.')
c_name, c_ext = '.'.join(c_split[:-1]), c_split[-1]
else:
c_name = corpus_name
c_ext = None
_corpus = Corpus.get(c_name)
if c_ext in cls._frontend_ext:
# graph-tool object
# @Todo: Corpus integration!
if not _corpus:
dt_lut = {'gt': 'network'}
_corpus = dict(data_type=dt_lut[c_ext])
_corpus.update(data_format=c_ext)
elif _corpus is False:
raise ValueError('Unknown Corpus `%s\'!' % c_name)
elif _corpus is None:
return None
if _corpus['data_type'] == 'text':
from .frontendtext import frontendText
frontend = frontendText(expe)
elif _corpus['data_type'] == 'network':
if _corpus.get('data_format') == 'gt':
from .frontendnetwork import frontendNetwork_gt
frontend = frontendNetwork_gt.from_expe(expe, corpus=_corpus)
else:
from .frontendnetwork import frontendNetwork
# Obsolete loading design. @Todo
frontend = frontendNetwork(expe)
frontend.load_data(randomize=False)
if hasattr(frontend, 'configure'):
frontend.configure()
return frontend
def load(cls, expe):
""" Return the frontend suited for the given expe"""
corpus_name = expe.get('corpus') or expe.get('random') or expe.get(
'concept')
if expe.get('driver'):
corpus_name += '.' + expe.ext.strip('.')
if '.' in corpus_name:
c_split = corpus_name.split('.')
c_name, c_ext = '.'.join(c_split[:-1]), c_split[-1]
else:
c_name = corpus_name
c_ext = None
_corpus = Corpus.get(c_name)
if c_ext in GramExp._frontend_ext:
# graph-tool object
# @Todo: Corpus intragation!
_corpus.update(data_format=c_ext)
elif _corpus is False:
raise ValueError('Unknown Corpus `%s\'!' % corpus)
elif _corpus is None:
return None
if _corpus['data_type'] == 'text':
from .frontendtext import frontendText
frontend = frontendText(expe)
elif _corpus['data_type'] == 'network':
if _corpus.get('data_format') == 'gt':
from .frontendnetwork import frontendNetwork_gt
frontend = frontendNetwork_gt.from_expe(expe, corpus=_corpus)
else:
from .frontendnetwork import frontendNetwork
# Obsolete loading design. @Todo
frontend = frontendNetwork(expe)
frontend.load_data(randomize=False)
frontend.configure()
return frontend
class Netw(ExpDesign):
# Use for Name on figure and table
_alias = dict((
('propro', 'Protein'),
('blogs', 'Blogs'),
('euroroad', 'Euroroad'),
('emaileu', 'Emaileu'),
('manufacturing', 'Manufacturing'),
('fb_uc', 'UC Irvine'),
('generator7', 'Network1'),
('generator12', 'Network2'),
('generator10', 'Network3'),
#('generator4' , 'Network4' ),
('generator4', 'Network2'),
#('pmk.ilfm_cgs' , 'ILFM' ),
#('pmk.immsb_cgs' , 'IMMSB' ),
))
# Networks Data
corpus_real_net = Corpus(
['manufacturing', 'fb_uc', 'blogs', 'emaileu', 'propro', 'euroroad'])
### Bursty
CORPUS_BURST_1 = Corpus([
'generator3', 'generator11', 'generator12', 'generator7', 'generator14'
])
### Non Bursty
CORPUS_NBURST_1 = Corpus([
'generator4', 'generator5', 'generator6', 'generator9', 'generator10'
])
CORPUS_SYN_ICDM = Corpus(
['generator7', 'generator12', 'generator10', 'generator4'])
CORPUS_REAL_ICDM = Corpus([
'manufacturing',
'fb_uc',
])
CORPUS_ALL_ICDM = CORPUS_SYN_ICDM + CORPUS_REAL_ICDM
CORPUS_REAL_PNAS = Corpus(
['manufacturing', 'fb_uc', 'blogs', 'emaileu', 'propro'])
CORPUS_ALL_PNAS = CORPUS_REAL_PNAS + CORPUS_SYN_ICDM
pnas_short = Corpus(['blogs', 'manufacturing', 'generator7', 'generator4'])
pnas_rest = (corpus_real_net + CORPUS_SYN_ICDM) - pnas_short
#data_net_all
#net_all
# Text Corpus
# intruder ?
data_text_all = Corpus(['kos', 'nips12', 'nips', 'reuter50',
'20ngroups']) # lucene
# Tensor Exp
EXPE_ICDM = ExpTensor((
('_data_type', ('networks', )),
('_refdir', ('debug10', 'debug11')),
#('corpus' , ('fb_uc', 'manufacturing')),
('corpus', CORPUS_ALL_ICDM),
('model', ('immsb_cgs', 'ilfm_cgs')),
('K', (5, 10, 15, 20)),
('N', ('all', )),
('hyper', ('fix', 'auto')),
('h**o', (0, )),
#('_repeat' , (0, 1, 2,3, 4, 5)),
))
PNAS1 = ExpTensor((
('corpus', CORPUS_ALL_PNAS),
('_data_type', 'networks'),
('_refdir', 'debug111111'), # ign in gen
#('model' , 'mmsb_cgs') ,
('model', ['immsb_cgs', 'ilfm_cgs']),
('K', 10),
('N', 'all'), # ign in gen
('hyper', ['auto', 'fix']), # ign in gen
('h**o', 0), # ign in gen
('_repeat', 1),
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
('iterations', '200'),
('_format', '{model}_{corpus}_{K}_{hyper}_{h**o}_{N}')))
PNAS2 = ExpTensor((
('corpus', CORPUS_ALL_PNAS),
('_data_type', 'networks'),
('_refdir', 'pnas2'), # ign in gen
#('model' , 'mmsb_cgs') ,
('model', ['immsb_cgs', 'ilfm_cgs']),
('K', 10),
('N', 'all'), # ign in gen
('hyper', ['fix', 'auto']), # ign in gen
('h**o', 0), # ign in gen
('_repeat', 0),
('_bind', [
'immsb_cgs.auto', 'ilfm_cgs.fix', 'ilfm_cgs.iterations.25',
'immsb_cgs.iterations.150'
]),
('iterations', [25, 150]),
('testset_ratio', [40, 60, 80]),
('_format', '{model}_{corpus}_{K}_{hyper}_{h**o}_{N}_{testset_ratio}'),
('_csv_typo',
'_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var'
),
))
PNAS3 = ExpTensor((
('corpus', pnas_short),
('_refdir', 'pnas3'), # ign in gen
('_data_type', 'networks'),
('model', ['immsb_cgs', 'ilfm_cgs']),
('K', 10),
('N', 'all'), # ign in gen
('hyper', ['fix', 'auto']), # ign in gen
('h**o', 0), # ign in gen
('_repeat', 0),
('_bind', [
'immsb_cgs.auto', 'ilfm_cgs.fix', 'ilfm_cgs.iterations.25',
'immsb_cgs.iterations.150'
]),
('iterations', [25, 150]),
('testset_ratio', 20),
('_format',
'{model}_{iterations}_{corpus}_{K}_{hyper}_{h**o}_{N}_{testset_ratio}'
),
('_csv_typo',
'_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var'
),
))
EXPE_ICDM_R = ExpTensor((
('_data_type', ('networks', )),
#('corpus' , ('fb_uc', 'manufacturing')),
('corpus', CORPUS_SYN_ICDM),
#('_refdir' , ('debug10', 'debug11')),
('_refdir', ('debug101010', 'debug111111')),
('model', ('immsb_cgs', 'ilfm_cgs')),
('K', (5, 10, 15, 20)),
('hyper', ('fix', 'auto')),
('h**o', (0, 1, 2)),
('N', ('all', )),
('_repeat', list(range(10))),
('iterations', '200'),
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
))
EXPE_ICDM_R_R = ExpTensor((
('_data_type', ('networks', )),
('corpus', ('fb_uc', 'manufacturing')),
('_refdir', ('debug101010', 'debug111111')),
('model', ('immsb_cgs', 'ilfm_cgs')),
('K', (5, 10, 15, 20)),
('hyper', ('fix', 'auto')),
('h**o', (0, 1, 2)),
('N', ('all', )),
('_repeat', list(range(10))),
('iterations', '200'),
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
))
# Single Expe
MODEL_FOR_CLUSTER_IBP = ExpSpace((
('_data_type', 'networks'),
('_refdir', 'debug11'),
('model', 'ilfm_cgs'),
('K', 20),
('N', 'all'),
('hyper', 'fix'),
('h**o', 0),
#('_repeat' , '*') ,
))
MODEL_FOR_CLUSTER_IMMSB = ExpSpace((
('_data_type', 'networks'),
('_refdir', 'debug11'),
('model', 'immsb_cgs'),
('K', 20),
('N', 'all'),
('hyper', 'auto'),
('h**o', 0),
#('_repeat' , '*') ,
))
default_gen = ExpTensor((
('corpus', CORPUS_SYN_ICDM),
('_data_type', 'networks'),
('_refdir', 'debug111111'), # ign in gen
#('model' , 'mmsb_cgs') ,
('model', ['immsb_cgs', 'ilfm_cgs']),
('K', 10),
('N', 'all'), # ign in gen
('hyper', ['auto', 'fix']), # ign in gen
('h**o', 0), # ign in gen
('_repeat', 1),
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
('alpha', 1),
('gmma', 1),
('delta', [(1, 5)]),
))
default_check = default_gen.copy()
default_check['model'] = 'immsb_cgs'
MODELS_GENERATE = ExpTensor((
('_data_type', 'networks'),
('_refdir', 'debug11'),
('model', ['immsb_cgs', 'ilfm_cgs']),
('K', 10),
('N', 'all'),
('hyper', ['fix', 'auto']),
('h**o', 0),
#('_repeat' , '*') ,
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
#### Temp
EXPE_ALL_ICDM_IBP = ExpTensor((
('_data_type', ('networks', )),
('_refdir', ('debug111111', 'debug101010')),
('corpus', CORPUS_ALL_ICDM),
('model', ('ilfm_cgs', )),
('K', (5, 10, 15, 20)),
('N', ('all', )),
('hyper', ('fix', )),
('h**o', (0, )),
('_repeat', (6, 7, 8, 9)),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
EXPE_ALL_ICDM_IMMSB = ExpTensor((
('_data_type', ('networks', )),
('_refdir', ('debug111111', 'debug101010')),
('corpus', CORPUS_ALL_ICDM),
('model', ('immsb_cgs', )),
('K', (5, 10, 15, 20)),
('N', ('all', )),
('hyper', ('auto', )),
('h**o', (0, )),
('_repeat', (6, 7, 8, 9)),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
RUN_DD = ExpTensor((
('_data_type', ('networks', )),
#('corpus' , ('fb_uc', 'manufacturing')),
('_refdir', ('test_temp', )),
('corpus', ('generator1', )),
('model', ('immsb_cgs', 'ilfm_cgs')),
('K', (5, )),
('N', ('all', )),
('hyper', ('auto', )),
('h**o', (0, )),
('hyper_prior', ('1 2 3 4', '20 2 10 2')),
('_repeat', (0, 1, 2, 4, 5)),
('_bind', ['immsb_cgs.auto', 'ilfm_cgs.fix']),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
EXPE_REAL_V2_IBP = ExpTensor((
('_data_type', ('networks', )),
('corpus', ('propro', 'blogs', 'euroroad', 'emaileu')),
(
'_refdir',
('debug111111'),
),
('model', ('ilfm_cgs', )),
('K', (10, )),
('hyper', ('fix', )),
('h**o', (0, )),
('N', ('all', )),
('_repeat', list(range(5))),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
EXPE_REAL_V2_IMMSB = ExpTensor((
('_data_type', ('networks', )),
('corpus', ('propro', 'blogs', 'euroroad', 'emaileu')),
(
'_refdir',
('debug111111', ),
),
('model', ('immsb_cgs', )),
('K', (10, )),
('hyper', ('auto', )),
('h**o', (0, )),
('N', ('all', )),
('_repeat', list(range(5))),
('_csv_typo ',
('_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var',
)),
))
RAGNRK = ExpTensor(
_data_type=['networks'],
corpus=['propro', 'blogs', 'euroroad', 'emaileu'],
_refdir=['ragnarok'],
model=['immsb_cgs'],
K=[10],
hyper=['auto'],
h**o=[0],
N=[10],
_repeat=list(range(2)),
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var'
)
default_expe = ExpTensor(
_data_type='networks',
corpus='clique4',
model='immsb_cgs',
hyper='auto',
_refdir='debug',
testset_ratio=20,
K=5,
N=50,
iterations=10,
h**o=0,
_format='{model}_{corpus}_{K}_{hyper}_{h**o}_{N}',
#_csv_typo = '_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var'
)
debug = ExpTensor(_data_type='networks',
corpus='clique4',
model=['immsb_cgs', 'ilfm_cgs', 'immsb_scvb'],
hyper='auto',
_refdir='debug',
testset_ratio=20,
K=4,
N=42,
iterations=3,
h**o=0,
_format='{model}_{corpus}_{K}_{hyper}_{h**o}_{N}',
_csv_typo='_iteration time_it _entropy _entropy_t _K')
model_all = Model(['immsb_cgs', 'ilfm_cgs', 'immsb_scvb', 'rescal'])
model_debug = Model(['mmsb_vb'])
class Levy(ExpDesign):
#refdir=gapv1 is old likelihood + sampling r/p.sampling
#rocw1: new likelihood + sampling r/p
#rocw2: newlikelihood + mixed samling/VB r/p
#rocw3: new likelihood + mixed samling/VB r/p and N_Phi added. (partial some correction and inversion sample/expectation)
#rocw4: new likelihood + samling r/p and
_alias = {'ml.iwmmsb_scvb3': 'WMMSB',
'ml.iwmmsb_scvb3_auto': 'WMMSB-bg',
'ml.immsb_scvb3': 'MMSB',
'ml.sbm_gt': 'SBM',
'ml.wsbm_gt': 'WSBM',
'link-dynamic-simplewiki': 'wiki-link',
'munmun_digg_reply': 'digg-reply',
'slashdot-threads': 'slashdot',
}
net_old = Corpus(['manufacturing', 'fb_uc', 'blogs', 'emaileu', 'propro', 'euroroad',
'generator7', 'generator12', 'generator10', 'generator4'])
net_gt = Corpus(['astro-ph', 'hep-th', 'netscience', 'cond-mat']) # all undirected
net_random = Corpus(['clique6', 'BA'])
net_test = Corpus(['manufacturing', 'fb_uc', 'netscience'])
net_large = Corpus(['link-dynamic-simplewiki', 'enron', 'foldoc'])
net_w = Corpus(['manufacturing', 'fb_uc']) + net_gt
# 'actor-collaboration' # Too big?
net_w2 = Corpus(['slashdot-threads', 'prosper-loans', 'munmun_digg_reply', 'moreno_names'])
# manufacturing
net_final = Corpus(['fb_uc',
#'manufacturing',
'hep-th',
'link-dynamic-simplewiki',
'enron',
'slashdot-threads',
'prosper-loans',
'munmun_digg_reply',
'moreno_names',
'astro-ph'])
net_all = net_old + net_gt + net_large
#
# Poisson Point process :
# * stationarity / ergodicity of p(d_i) ?
# * Erny theorem (characterization by void probabilities ?
# * Inference ? Gamma Process ?
# * Sparsity ?
warm = ExpTensor(
corpus=['manufacturing'],
model='iwmmsb_scvb3',
N='all',
K=10,
hyper='auto',
h**o=0,
testset_ratio=20,
validset_ratio=10,
# Sampling
chunk='stratify',
sampling_coverage=0.42,
#chi_a=10, tau_a=100, kappa_a=0.6,
#chi_b=10, tau_b=500, kappa_b=0.9,
chi_a=1, tau_a=1024, kappa_a=0.5,
chi_b=1, tau_b=1024, kappa_b=0.5,
zeros_set_prob=1/2,
zeros_set_len=50,
#delta = [[1, 1]],
#delta = [[0.5, 10]],
delta='auto',
fig_xaxis=[('_observed_pt', 'visited edges')],
fig_legend=4,
legend_size=12,
#ticks_size = 20,
title_size=20,
driver='gt', # graph-tool driver
_data_type='networks',
_refdir='debug_scvb3',
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}',
_measures='_observed_pt time_it _entropy _K _chi_a _tau_a _kappa_a _chi_b _tau_b _kappa_b _roc _wsim _pr',
)
# Sparse sampling
warm_sparse_chunk = ExpGroup(warm, sampling_coverage=1, chunk='sparse')
# Sampling sensiblity | hyper-delta sensibility
warm_sampling = ExpGroup(warm, delta=[[1, 1], [0.5, 10], [10, 0.5]],
zeros_set_prob=[1/2, 1/3, 1/4], zeros_set_len=[10, 50])
arm_sampling = ExpGroup(warm_sampling, delta='auto', model='immsb_scvb3')
sbm_base = ExpGroup(warm, model=['sbm_gt', 'wsbm_gt', 'rescal_als'],
zeros_set_prob=None, zeros_set_len=None, delta=None,
_measures='time_it _entropy _K _roc _wsim _pr')
wsbm2_base = ExpGroup(sbm_base, model=['wsm_g', 'wsbm2_gt'])
wsbm_base = ExpGroup(sbm_base, model=['wsbm_gt'])
# Compare sensibility
eta_b = ExpGroup([arm_sampling], _refdir='eta', corpus=net_w,
zeros_set_prob=[1/2, 1/4])
eta_w = ExpGroup([warm_sampling], _refdir='eta', corpus=net_w,
zeros_set_prob=[1/2, 1/4])
eta = ExpGroup([eta_b, eta_w])
eta_sbm = ExpGroup(sbm_base, _refdir='eta', corpus=net_w)
# test/visu
warm_visu = ExpGroup(warm, delta=[[1, 1], 'auto'],
zeros_set_prob=[1/2], zeros_set_len=[10])
# Best selection visu (eta)
best_mmsb = ExpGroup(eta_b, zeros_set_prob=1/4, zeros_set_len=50, delta='auto')
best_wmmsb = ExpGroup(eta_w, zeros_set_prob=1/4, zeros_set_len=50, delta=[[0.5, 10]])
best_scvb = ExpGroup([best_mmsb, best_wmmsb])
eta_best = ExpGroup([best_scvb, eta_sbm])
# Corrected zero sampling
eta2_base = ExpGroup(warm, testset_ratio=20, _refdir='roc5', corpus=net_w) # roc5§roc5_N
eta2_sbm = ExpGroup(eta2_base, model=['sbm_gt', 'wsbm_gt', 'rescal_als'],
zeros_set_prob=None, zeros_set_len=None, delta=None,
_measures='time_it _entropy _K _roc _wsim _pr')
eta2_b = ExpGroup(eta2_base, model="immsb_scvb3", zeros_set_prob=1/2, zeros_set_len=10, delta='auto')
eta2_b10 = eta2_b
eta2_b50 = ExpGroup(eta2_base, model="immsb_scvb3", zeros_set_prob=1/2, zeros_set_len=50, delta='auto')
eta0_w = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=10, delta=[[10, 0.5]])
eta1_w = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=10, delta=[[1, 1]])
eta2_w = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=10, delta=[[0.5, 10]])
eta3_w = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=10, delta=[[0.1, 10]])
eta0_w50 = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=50, delta=[[10, 0.5]])
eta1_w50 = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=50, delta=[[1, 1]])
eta2_w50 = ExpGroup(eta2_base, model="iwmmsb_scvb3", zeros_set_prob=1/2, zeros_set_len=50, delta=[[0.5, 10]])
eta2a_w = ExpGroup(eta2_base, model="iwmmsb_scvb3_auto", zeros_set_prob=1/2,
zeros_set_len=10, delta='auto', _model="ml.iwmmsb_scvb3")
eta2a_w10 = eta2a_w
eta2a_w50 = ExpGroup(eta2_base, model="iwmmsb_scvb3_auto", zeros_set_prob=1/2,
zeros_set_len=50, delta='auto', _model="ml.iwmmsb_scvb3")
eta4_full = ExpGroup([eta2_b50, eta2_w50, eta2a_w50]) # weighte are squared
roc_visu_sbm = ExpGroup(sbm_base, corpus=net_w, testset_ratio=20, model=['sbm_gt', 'wsbm_gt'], _refdir='roc5')
roc_visu_sbm_full = ExpGroup(sbm_base, corpus=net_w, testset_ratio=20, model=[
'rescal_als', 'sbm_gt', 'wsbm_gt'], _refdir='roc5')
roc_visu_full = ExpGroup([eta4_full, roc_visu_sbm_full])
roc_visu_final10 = ExpGroup([eta2_b, eta2_w, eta2a_w, roc_visu_sbm])
roc_visu_final50 = ExpGroup([eta2_b50, eta2_w50, eta2a_w50, roc_visu_sbm])
roc_visu_final = roc_visu_final50
roc_visu_final2 = ExpGroup([eta2_b50, eta2_w50, eta2a_w50], _refdir="roc5v")
roc_visu_final2 = ExpGroup([roc_visu_final2, roc_visu_sbm])
roc_visu_final2_full = ExpGroup([roc_visu_final2, roc_visu_sbm_full])
roc_visu_final3 = ExpGroup([eta2_b50, eta1_w50, eta2a_w50], _refdir="roc5v")
roc_visu_final3 = ExpGroup([roc_visu_final3, roc_visu_sbm])
online_roc = ExpGroup(roc_visu_final3, training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_ = ExpGroup([eta2_b50, eta2a_w50, roc_visu_sbm], training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_sbm = ExpGroup(roc_visu_sbm, training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_mmsb = ExpGroup([eta2_b50, eta1_w50, eta2a_w50], training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_w = ExpGroup([eta1_w50, eta2a_w50, wsbm_base], training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_wm = ExpGroup([eta1_w50, eta2a_w50], training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
online_roc_wsbm = ExpGroup([wsbm_base], training_ratio=[1, 5, 10, 20, 30, 50, 100], _refdir='online1w', corpus=net_final,
_seed='corpus', testset_ratio=20,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}--{training_ratio}',
)
roc_w = ExpGroup([eta0_w, eta1_w, eta2_w, eta0_w50, eta1_w50, eta2_w50]) # squared weight
roc_b = ExpGroup(eta2_base, model="immsb_scvb3", zeros_set_prob=1/2, zeros_set_len=[10, 50], delta='auto')
conv_w = ExpGroup([eta0_w50, eta1_w50, eta2_w50, eta2a_w50], corpus=['astro-ph', 'enron', 'munmun_digg_reply'])
param1 = ExpGroup(eta2a_w50, c0=[0.5, 1, 10, 100], r0=[0.1, 0.5, 1])
gap = ExpGroup([param1], _refdir='gap_hyper', testset_ratio=20, corpus=net_w,
ce=[100], eps=[1e-6],
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}-{c0}-{r0}-{ce}-{eps}',
)
gap_visu = ExpGroup([eta2a_w50], _refdir='gap_hyper', testset_ratio=20, corpus=net_w,
ce=[1, 10, 100], eps=[1e-5, 1e-6, 1e-7],
c0=10, r0=1,
_format='{corpus}_{model}_{N}_{K}_{hyper}_{h**o}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}_{delta}_{zeros_set_len}_{zeros_set_prob}-{c0}-{r0}-{ce}-{eps}',
)
class Netw2(ExpDesign):
data_net_all = Corpus([
'manufacturing', 'fb_uc', 'blogs', 'emaileu', 'propro', 'euroroad',
'generator7', 'generator12', 'generator10', 'generator4'
])
net_all = data_net_all + Corpus(['clique6', 'BA'])
# compare perplexity and rox curve from those baseline.
compare_scvb = ExpTensor(
corpus=['clique6', 'BA'],
model=['immsb_cgs', 'ilfm_cgs', 'rescal', 'immsb_cvb'],
N=200,
K=6,
iterations=30,
hyper='auto',
testset_ratio=20,
h**o=0,
mask='unbalanced',
_data_type='networks',
_refdir='debug_scvb',
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}',
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _alpha _gmma alpha_mean delta_mean alpha_var delta_var'
)
compare_scvb_m = ExpGroup(compare_scvb, model=['immsb_cgs', 'immsb_cvb'])
cvb = ExpGroup(compare_scvb, model='immsb_cvb')
scvb = ExpTensor(
corpus=['BA'],
model='immsb_scvb',
N=200,
chunk='adaptative_1',
K=6,
iterations=3,
hyper='auto',
testset_ratio=20,
#chi_a = 1,
#tau_a = 42,
kappa_a=0.75,
#chi_b = 42,
#tau_b = 300,
#kappa_b = 0.9,
h**o=0,
mask='unbalanced',
_data_type='networks',
_refdir='debug_scvb',
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}',
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _chi_a _tau_a _kappa_a _chi_b _tau_b _kappa_b _elbo _roc'
)
scvb_t = ExpGroup(scvb, _refdir='debug_')
scvb_chi = ExpGroup(
scvb,
chi_a=1,
tau_a=42,
kappa_a=[0.6, 0.7, 0.9],
chi_b=42,
tau_b=300,
kappa_b=[0.6, 0.7, 0.9],
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}'
)
scvb_chi2 = ExpGroup(
scvb,
chi_a=[10],
tau_a=[100],
kappa_a=[0.6],
chi_b=[10],
tau_b=[500],
kappa_b=[0.9],
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}'
)
scvb_chi_2 = ExpGroup(
scvb,
chi_a=[1, 10],
tau_a=[42, 100, 500],
kappa_a=[0.6],
chi_b=[1, 10],
tau_b=[42, 100, 500],
kappa_b=[0.6, 0.7, 0.9],
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}_{chi_a}-{tau_a}-{kappa_a}_{chi_b}-{tau_b}-{kappa_b}'
)
#
#
#
# * iter1 : don"t set masked, and gamma is not symmetric
#
# * iter2 : don"t set masked, and gamma is symmetric
#
#
#
scvb1_chi_a = ExpTensor(
corpus=['blogs', 'manufacturing', 'generator7', 'generator10'],
model='immsb_scvb',
N='all',
chunk='adaptative_1',
K=6,
iterations=1,
hyper='auto',
testset_ratio=20,
chi_a=[0.5, 1, 2, 10],
tau_a=[0.5, 1, 2, 16, 64, 256, 1024],
kappa_a=[0.51, 0.45, 1],
#h**o = 0,
#mask = 'unbalanced',
_data_type='networks',
_refdir='debug_scvb_chia_i1',
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}-{_name}-{_id}',
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _chi_a _tau_a _kappa_a _chi_b _tau_b _kappa_b'
)
scvb1_chi_b = ExpTensor(
corpus=['blogs', 'manufacturing', 'generator7', 'generator10'],
model='immsb_scvb',
N='all',
chunk='adaptative_1',
K=6,
iterations=1,
hyper='auto',
testset_ratio=20,
chi_b=[0.5, 1, 2, 10],
tau_b=[0.5, 1, 2, 16, 64, 256, 1024],
kappa_b=[0.51, 0.45, 1],
#h**o = 0,
#mask = 'unbalanced',
_data_type='networks',
_refdir='debug_scvb_chib_i1',
_format=
'{corpus}_{model}_{N}_{K}_{iterations}_{hyper}_{h**o}_{mask}_{testset_ratio}_{chunk}-{_name}-{_id}',
_csv_typo=
'_iteration time_it _entropy _entropy_t _K _chi_a _tau_a _kappa_a _chi_b _tau_b _kappa_b'
)
# Noel expe
noel = ExpGroup([scvb, compare_scvb],
N='all',
corpus=data_net_all,
mask=['balanced', 'unbalanced'],
_refdir='noel')
noel_cvb = ExpGroup(cvb,
N='all',
corpus=data_net_all,
mask=['balanced', 'unbalanced'],
_refdir='noel')
noel_scvb = ExpGroup(scvb,
N='all',
corpus=data_net_all,
mask=['balanced', 'unbalanced'],
_refdir='noel')
noel_scvb_ada = ExpGroup(noel_scvb,
chunk=[
'adaptative_0.1', 'adaptative_0.5',
'adaptative_1', 'adaptative_10'
])
noel_mmsb = ExpGroup([scvb, compare_scvb_m],
N='all',
corpus=data_net_all,
mask=['balanced', 'unbalanced'],
_refdir='noel')
compare_scvb2 = ExpGroup(compare_scvb,
N='all',
corpus=data_net_all,
iterations=100,
mask=['balanced', 'unbalanced'],
_refdir='noel2')
noel3 = ExpGroup(scvb_chi2,
N='all',
chunk=['adaptative_0.1', 'adaptative_1'],
corpus=data_net_all,
mask=['unbalanced'],
_refdir='noel3')
# cvb debug
pd = ExpGroup(compare_scvb,
iterations=150,
model='immsb_cvb',
_repeat='debug_cvb',
N='all',
corpus=data_net_all,
mask=['unbalanced'],
_refdir='noel2')
pd2n = ExpGroup(compare_scvb,
iterations=150,
model='immsb_cvb',
_repeat='debug_cvb_2n',
N='all',
corpus=data_net_all,
mask=['unbalanced'],
_refdir='noel2')
class Aistats19(ExpDesign):
_alias = {
'ml.iwmmsb_scvb3': 'WMMSB-bg',
'ml.immsb_scvb3': 'MMSB',
'ml.sbm_gt': 'SBM',
'ml.wsbm_gt': 'WSBM',
'ml.sbm_ai': 'SBM-ai',
'ml.wsbm_ai_n': 'WSBM-ai-n',
'ml.wsbm_ai_p': 'WSBM-ai-p',
'ml.epm': 'EPM',
'link-dynamic-simplewiki': 'wiki-link',
'munmun_digg_reply': 'digg-reply',
'slashdot-threads': 'slashdot',
}
net_final = Corpus([
'fb_uc',
#'manufacturing',
'hep-th',
'link-dynamic-simplewiki',
'enron',
'slashdot-threads',
'prosper-loans',
'munmun_digg_reply',
'moreno_names',
'astro-ph'
])
base_graph = dict(
corpus='manufacturing',
_seed='corpus',
testset_ratio=20,
validset_ratio=10,
training_ratio=100,
# Model global param
N='all',
K=10,
kernel='none',
# plotting
fig_legend=4,
legend_size=7,
#ticks_size = 20,
title_size=18,
fig_xaxis=('time_it', 'time'),
fig_yaxis=dict(wsim='MSE'),
driver='gt', # graph-tool driver
_write=True,
_data_type='networks',
_refdir='aistat_wmmsb2',
_format="{model}-{kernel}-{K}_{corpus}-{training_ratio}",
_measures=[
'time_it',
'entropy@data=valid',
'roc@data=test',
#'roc@data=test&measure_freq=10',
'pr@data=test&measure_freq=20',
'wsim@data=test&measure_freq=10',
'roc2@data=test&measure_freq=10',
'wsim2@data=test&measure_freq=10',
],
)
sbm_peixoto = ExpTensor(base_graph, model='sbm_gt')
wsbm_peixoto = ExpTensor(base_graph, model='wsbm_gt')
rescal_als = ExpTensor(base_graph, model='rescal_als')
wsbm = ExpTensor(
base_graph,
model='sbm_aicher',
kernel=['bernoulli', 'normal', 'poisson'],
#kernel = 'normal',
mu_tol=0.001,
tau_tol=0.001,
max_iter=100,
)
wsbm_1 = ExpTensor(
wsbm,
model='sbm_ai',
_model='ml.sbm_aicher',
#model='ml.sbm_aicher',
kernel='bernoulli',
)
wsbm_2 = ExpTensor(
wsbm,
model='wsbm_ai_n',
_model='ml.sbm_aicher',
#model='ml.sbm_aicher',
kernel='normal',
)
wsbm_3 = ExpTensor(
wsbm,
model='wsbm_ai_p',
_model='ml.sbm_aicher',
#, model='ml.sbm_aicher',
kernel='poisson',
)
wsbm_t = ExpGroup([wsbm_1, wsbm_2, wsbm_3])
wmmsb = ExpTensor(
base_graph,
model="iwmmsb_scvb3",
chunk='stratify',
delta='auto',
sampling_coverage=0.5,
zeros_set_prob=1 / 2,
zeros_set_len=50,
chi_a=1,
tau_a=1024,
kappa_a=0.5,
chi_b=1,
tau_b=1024,
kappa_b=0.5,
tol=0.001,
#fig_xaxis = ('_observed_pt', 'visited edges'),
)
mmsb = ExpTensor(wmmsb, model="immsb_scvb3")
epm = ExpTensor(wmmsb, model="epm")
aistats_design_wmmsb = ExpGroup(
[wmmsb],
corpus=net_final,
training_ratio=[1, 5, 10, 20, 30, 50, 100], # subsample the edges
_refdir="ai19_1",
)
aistats_design_mmsb = ExpGroup(
[mmsb],
corpus=net_final,
training_ratio=[1, 5, 10, 20, 30, 50, 100], # subsample the edges
_refdir="ai19_1",
)
aistats_design_mm = ExpGroup([aistats_design_wmmsb, aistats_design_mmsb])
aistats_design_wsbm = ExpGroup(
[wsbm],
corpus=net_final,
training_ratio=[1, 5, 10, 20, 30, 50, 100], # subsample the edges
_refdir="ai19_1",
)
aistats_design_peixoto = ExpGroup(
[sbm_peixoto, wsbm_peixoto],
corpus=net_final,
training_ratio=[1, 5, 10, 20, 30, 50, 100], # subsample the edges
_refdir="ai19_1",
)
aistats_design_final = ExpGroup(
[wmmsb, mmsb, wsbm_t, sbm_peixoto, wsbm_peixoto],
corpus=net_final,
training_ratio=[1, 5, 10, 20, 30, 50, 100], # subsample the edges
_refdir="ai19_1",
)
aistats_design_final_2 = ExpGroup(
[wmmsb, mmsb, wsbm_t, sbm_peixoto, wsbm_peixoto, epm],
corpus=net_final,
K=[20, 30, 50],
training_ratio=[100], # subsample the edges
_refdir="ai19_1",
)
#
#
# Post expe Fix
#
#
aistats_compute_zcp_w_tmp = ExpGroup(
[wmmsb],
corpus=net_final,
K=[20, 30, 50],
training_ratio=[100], # subsample the edges
_refdir="ai19_1",
)
aistats_compute_zcp_a_tmp = ExpGroup(
[wsbm_3],
corpus=net_final,
K=[20, 30, 50],
training_ratio=[100], # subsample the edges
_refdir="ai19_1",
)
aistats_compute_wsim4 = ExpGroup(
[wsbm_t, sbm_peixoto, wsbm_peixoto],
corpus=net_final,
K=[10],
training_ratio=[100], # subsample the edges
_refdir="ai19_1",
)
aistats_doh = ExpGroup(
[epm],
corpus=net_final,
K=[20],
training_ratio=[100], # subsample the edges
_refdir="ai19_1",
)