def _delete_orientation(self):
"""
Delete orientation metadata. Garbage orientation metadata can lead to
severe mis-registration trouble.
"""
# prepare for smart caching
if self.scratch is None:
self.scratch = self.output_dir
if self.caching:
cache_dir = os.path.join(self.scratch, 'cache_dir')
if not os.path.exists(cache_dir):
os.makedirs(cache_dir)
mem = Memory(cachedir=cache_dir, verbose=5)
else:
mem = Memory(None, verbose=0)
# deleteorient for func
for attr in ['n_sessions', 'session_output_dirs']:
if getattr(self, attr) is None:
warnings.warn("'%s' attribute of is None! Skipping" % attr)
break
else:
self.func = [mem.cache(delete_orientation)(
self.func[sess], self.session_output_dirs[sess])
for sess in range(self.n_sessions)]
# deleteorient for anat
if self.anat is not None:
self.anat = mem.cache(delete_orientation)(
self.anat, self.anat_output_dir)
def set_option(physical_cache_path: str = None) -> None:
"""
Set global options to the package.
:param physical_cache_path: Caching across Python interpreter sessions
can save a lot of time. This option allows on-disk caching when the
path is specified.
Specifying an empty string ("") switches disk-caching off explicitly.
Use "." to specify the current working directory instead.
Kaggle kernels "seemed to like" on disk caching as long as I didn't
try to commit the notebook. Then things ended up with a Code: 0 error
failing the publishing attempt. It may be the most convenient there to
comment out a set_option("") while experimenting, and to uncomment it
just before committing the kernel.
Leaving it at the default of None leaves the options unchanged.
(There's likely more to come.)
:return: None
"""
if physical_cache_path is not None:
global _mem
if physical_cache_path != "":
_mem = Memory(physical_cache_path, verbose=0)
digit_correlations._cached_get_digit_correlation_data = \
_mem.cache(digit_correlations._uncached_get_digit_correlation_data)
digit_entropy_distribution.cached_generate_sample = \
_mem.cache(digit_entropy_distribution._uncached_generate_sample)
else:
_mem = None
digit_correlations.cached = \
digit_correlations._lru_cached_get_digit_correlation_data
digit_entropy_distribution.cached_generate_sample = \
digit_entropy_distribution._lru_cached_generate_sample
def load_adni_longitudinal_rs_fmri(dirname='ADNI_longitudinal_rs_fmri',
prefix='wr*.nii'):
""" Returns paths of ADNI rs-fMRI
"""
# get file paths and description
images, subject_paths, description = _get_subjects_and_description(
base_dir=dirname, prefix='I[0-9]*')
images = np.array(images)
# get func files
func_files = list(map(lambda x: _glob_subject_img(
x, suffix='func/' + prefix, first_img=True),
subject_paths))
func_files = np.array(func_files)
# get motion files
# motions = None
motions = list(map(lambda x: _glob_subject_img(
x, suffix='func/' + 'rp_*.txt', first_img=True), subject_paths))
# get phenotype from csv
dx = pd.read_csv(os.path.join(_get_data_base_dir('ADNI_csv'),
'DXSUM_PDXCONV_ADNIALL.csv'))
roster = pd.read_csv(os.path.join(_get_data_base_dir('ADNI_csv'),
'ROSTER.csv'))
df = description[description['Image_ID'].isin(images)]
df = df.sort_values(by='Image_ID')
dx_group = np.array(df['DX_Group'])
subjects = np.array(df['Subject_ID'])
exams = np.array(df['EXAM_DATE'])
exams = [date(int(e[:4]), int(e[5:7]), int(e[8:])) for e in exams]
# caching dataframe extraction functions
CACHE_DIR = _get_cache_base_dir()
cache_dir = os.path.join(CACHE_DIR, 'joblib', 'load_data_cache')
if not os.path.isdir(cache_dir):
os.makedirs(cache_dir)
memory = Memory(cachedir=cache_dir, verbose=0)
def _get_ridsfmri(subjects):
return [_ptid_to_rid(s, roster) for s in subjects]
rids = np.array(memory.cache(_get_ridsfmri)(subjects))
def _get_examdatesfmri(rids):
return [_get_dx(rids[i], dx, exams[i], viscode=None, return_code=True)
for i in range(len(rids))]
exam_dates = np.array(memory.cache(_get_examdatesfmri)(rids))
def _get_viscodesfmri(rids):
return [_get_vcodes(rids[i], str(exam_dates[i]), dx)
for i in range(len(rids))]
viscodes = np.array(memory.cache(_get_viscodesfmri)(rids))
vcodes, vcodes2 = viscodes[:, 0], viscodes[:, 1]
return Bunch(func=func_files, dx_group=dx_group, exam_codes=vcodes,
exam_dates=exam_dates, exam_codes2=vcodes2,
motion=motions,
subjects=subjects, images=images)
def main():
idir = 'data/tweet_sent'
location = './cache'
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
bert_name = 'bert-base-uncased'
memory = Memory(location, verbose=0)
train_dataloader, validation_dataloader = memory.cache(prepare_input_data)(idir, bert_name)
model = memory.cache(train_model)(train_dataloader, bert_name, device)
model.eval()
estimate_model(model, validation_dataloader)
def _niigz2nii(self):
"""
Convert .nii.gz to .nii (crucial for SPM).
"""
cache_dir = os.path.join(self.output_dir, 'cache_dir')
mem = Memory(cache_dir, verbose=100)
self.func = mem.cache(do_niigz2nii)(self.func,
output_dir=self.output_dir)
if not self.anat is None:
self.anat = mem.cache(do_niigz2nii)(self.anat,
output_dir=self.output_dir)
def load_adni_longitudinal_hippocampus_volume():
""" Returns longitudinal hippocampus measures
"""
BASE_DIR = _get_data_base_dir('ADNI_csv')
roster = pd.read_csv(os.path.join(BASE_DIR, 'ROSTER.csv'))
dx = pd.read_csv(os.path.join(BASE_DIR, 'DXSUM_PDXCONV_ADNIALL.csv'))
fs = pd.read_csv(os.path.join(BASE_DIR, 'UCSFFSX51_05_20_15.csv'))
# extract hippocampus numerical values
column_idx = np.arange(131, 147)
cols = ['ST' + str(c) + 'HS' for c in column_idx]
hipp = fs[cols].values
idx_num = np.array([~np.isnan(h).all() for h in hipp])
hipp = hipp[idx_num, :]
# extract roster id
rids = fs['RID'].values[idx_num]
# caching dataframe extraction functions
CACHE_DIR = _get_cache_base_dir()
cache_dir = os.path.join(CACHE_DIR, 'joblib', 'load_data_cache')
if not os.path.isdir(cache_dir):
os.makedirs(cache_dir)
memory = Memory(cachedir=cache_dir, verbose=0)
# get subject id
def _getptidshippo(rids):
return [_rid_to_ptid(rid, roster) for rid in rids]
ptids = memory.cache(_getptidshippo)(rids)
# extract exam date
exams = fs['EXAMDATE'].values[idx_num]
vcodes = fs['VISCODE'].values[idx_num]
vcodes2 = fs['VISCODE2'].values[idx_num]
exams = list(map(
lambda e: date(int(e[:4]), int(e[5:7]), int(e[8:])), exams))
exams = np.array(exams)
# extract diagnosis
def _getdxhippo(rids, exams):
return np.array(list(map(_get_dx, rids, [dx]*len(rids), exams)))
dx_ind = memory.cache(_getdxhippo)(rids, exams)
dx_group = DX_LIST[dx_ind]
return Bunch(dx_group=np.array(dx_group), subjects=np.array(ptids),
hipp=np.array(hipp), exam_dates=np.array(exams),
exam_codes=np.array(vcodes), exam_codes2=np.array(vcodes2))
def test_multioutput(self):
cache = Memory(location=tempfile.gettempdir())
cached_func = cache.cache(sklearn.datasets.make_regression)
X, Y = cached_func(n_samples=250,
n_features=20,
n_informative=9,
n_targets=4,
bias=0.5,
effective_rank=10,
tail_strength=0.4,
noise=0.3,
shuffle=True,
coef=False,
random_state=1)
X_train = X[:200, :]
Y_train = Y[:200, :]
X_test = X[200:, :]
Y_test = Y[200:, :]
data = {
'X_train': X_train,
'Y_train': Y_train,
'X_test': X_test,
'Y_test': Y_test
}
dataset_properties = {'multioutput': True}
cs = SimpleRegressionPipeline(dataset_properties=dataset_properties).\
get_hyperparameter_search_space()
self._test_configurations(cs,
data=data,
dataset_properties=dataset_properties)
def __init__(
self,
gmm_ubm,
feature=None, cache=False
):
super(SpeakerIdentification, self).__init__()
self.gmm_ubm = gmm_ubm
# default features for speaker identification are MFCC
# 13 coefs + delta coefs + delta delta coefs
# + delta energy + delta delta energy
if feature is None:
from pyannote.feature.yaafe import YaafeMFCC
feature = YaafeMFCC(
e=False, De=True, DDe=True,
coefs=13, D=True, DD=True
)
self.feature = feature
if cache:
# initialize cache
from joblib import Memory
from tempfile import mkdtemp
memory = Memory(cachedir=mkdtemp(), verbose=0)
# cache feature extraction method
self.get_features = memory.cache(self.get_features)
def pipeline(input_csv, output_csv, n_cores=1, cache="/tmp"):
"""
Find first commit hash of appearing identifier in file.
:param input_csv: Path to input csv.
:param output_csv: Path to store result csv.
:param n_cores: How many cores to use.
:param cache: Cache location. If empty - no caching
"""
if cache:
memory = Memory(cache, verbose=0)
parallel_comp = memory.cache(func=_parallel_comp)
else:
parallel_comp = _parallel_comp
df = pd.read_csv(input_csv, header=None)
df.columns = COLUMNS
args = [
tuple(getattr(line, col) for col in COLUMNS)
for i, line in df.iterrows()
]
res = Parallel(n_jobs=n_cores)(tqdm(
(delayed(parallel_comp)(arg) for arg in args), total=len(args) - 1))
new_args = [arg + (h, filename) for arg, (h, filename) in zip(args, res)]
to_df = defaultdict(list)
for arg in new_args:
for col in NEW_COLUMNS:
to_df[col].append(arg[NEW_COL2IND[col]])
new_df = pd.DataFrame.from_dict(to_df)
new_df = new_df[NEW_COLUMNS]
new_df.to_csv(output_csv, index=False, header=False, compression="gzip")
def parse_all(basedir, mem: Memory = None, with_paths=False, from_cache=False):
_parse = parse if mem is None else mem.cache(parse, ignore=['from_cache'])
for f in Path(basedir).rglob("*.html"):
text = _parse(f, from_cache)
if text is not None:
yield (f, text) if with_paths else text
def test_multilabel(self):
cache = Memory(location=tempfile.gettempdir())
cached_func = cache.cache(
sklearn.datasets.make_multilabel_classification
)
X, Y = cached_func(
n_samples=150,
n_features=20,
n_classes=5,
n_labels=2,
length=50,
allow_unlabeled=True,
sparse=False,
return_indicator=True,
return_distributions=False,
random_state=1
)
X_train = X[:100, :]
Y_train = Y[:100, :]
X_test = X[101:, :]
Y_test = Y[101:, ]
data = {'X_train': X_train, 'Y_train': Y_train,
'X_test': X_test, 'Y_test': Y_test}
dataset_properties = {'multilabel': True}
cs = SimpleClassificationPipeline(dataset_properties=dataset_properties).\
get_hyperparameter_search_space()
self._test_configurations(configurations_space=cs, data=data)
def test_store_standard_types(capsys, tmpdir, compress, arg):
"""Test that standard types can be cached in s3fs store."""
def func(arg):
"""Dummy function."""
print("executing function")
return arg
register_s3fs_store_backend()
mem = Memory(location=tmpdir.strpath,
backend='s3',
verbose=0,
compress=compress,
backend_options=dict(bucket="test"))
assert mem.store_backend.location == os.path.join("s3:/", tmpdir.strpath,
"joblib")
cached_func = mem.cache(func)
result = cached_func(arg)
assert result == arg
out, err = capsys.readouterr()
assert out == "executing function\n"
assert not err
# Second call should also return the cached result
result2 = cached_func(arg)
assert result2 == arg
out, err = capsys.readouterr()
assert not out
assert not err
def test_clear_cache(capsys, tmpdir):
"""Check clearing the cache."""
def func(arg):
"""Dummy function."""
print("executing function")
return arg
register_s3fs_store_backend()
mem = Memory(location=tmpdir.strpath,
backend='s3',
verbose=0,
backend_options=dict(bucket="test"))
cached_func = mem.cache(func)
cached_func("test")
out, _ = capsys.readouterr()
assert out == "executing function\n"
mem.clear()
cached_func("test")
out, _ = capsys.readouterr()
assert out == "executing function\n"
mem.clear()
print(mem.store_backend.location)
assert not os.listdir(mem.store_backend.location)
def cacheSetup(): global rollDiceCached global memory location = './.cache' memory = Memory(location, verbose=0) rng = np.random.RandomState(42) rollDiceCached = memory.cache(rollDice)
def extract_group_components(subject_components, variances,
ccs_threshold=None, n_group_components=None,
cachedir=None):
# Use asarray to cast to a non memmapped array
subject_components = np.asarray(subject_components)
if len(subject_components) == 1:
# We are in a single subject case
return subject_components[0, :n_group_components].T, \
variances[0][:n_group_components]
# The group components (concatenated subject components)
group_components = subject_components.T
group_components = np.reshape(group_components,
(group_components.shape[0], -1))
# Save memory
del subject_components
# Inter-subject CCA
memory = Memory(cachedir=cachedir, mmap_mode='r')
svd = memory.cache(linalg.svd)
cca_maps, ccs, _ = svd(group_components, full_matrices=False)
# Save memory
del group_components
if n_group_components is None:
n_group_components = np.argmin(ccs > ccs_threshold)
cca_maps = cca_maps[:, :n_group_components]
ccs = ccs[:n_group_components]
return cca_maps, ccs
def test_get_cache_items(tmpdir):
"""Test cache items listing."""
def func(arg):
"""Dummy function."""
return arg
register_hdfs_store_backend()
mem = Memory(location=tmpdir.strpath,
host=__namenode__,
backend='hdfs',
user='******',
verbose=100,
compress=False)
assert not mem.store.get_cache_items()
cached_func = mem.cache(func)
for arg in ["test1", "test2", "test3"]:
cached_func(arg)
# get_cache_items always returns an empty list for the moment
assert len(mem.store.get_cache_items()) == 3
mem.clear()
assert not mem.store.get_cache_items()
def __init__(
self,
hmm=None,
n_components=16, covariance_type='diag',
min_duration=0.250,
feature=None, cache=False
):
super(SpeechActivityDetection, self).__init__()
self.hmm = hmm
self.hmm.min_duration = min_duration
# default features for speech activity detection
# are MFCC (12 coefficients + delta coefficient + delta energy)
if feature is None:
from pyannote.feature.yaafe import YaafeMFCC
feature = YaafeMFCC(e=False, coefs=12, De=True, D=True)
self.feature = feature
if cache:
# initialize cache
from joblib import Memory
from tempfile import mkdtemp
memory = Memory(cachedir=mkdtemp(), verbose=0)
# cache feature extraction method
self.get_features = memory.cache(self.get_features)
def parallel_func(fcn, n_jobs=-1, verbose=None, total=None, mesg=None,
cache_dir=None, **kwargs):
"""Get an instance of parallel and delayed function.
This function is inspired by MNE's one.
Parameters
----------
func : callable
A function.
n_jobs : int
Number of jobs to run in parallel.
total : int | None
If int, use a progress bar to display the progress of dispatched
jobs. This should only be used when directly iterating, not when
using ``split_list`` or :func:`np.array_split`.
If None (default), do not add a progress bar.
mesg : string | None
Message to display on the progress bar
cache_dir : string | None
If path to an existing directory, the function is going to cache the
computations
kwargs : dict | {}
Additional arguments are sent to the joblibe.Parallel function.
Returns
-------
parallel: instance of joblib.Parallel or list
The parallel object.
my_func: callable
``func`` if not parallel or delayed(func).
"""
from frites.config import CONFIG
# set_log_level(verbose)
# manually merge inputs inside the default config
for k, v in CONFIG["JOBLIB_CFG"].copy().items():
kwargs[k] = v
# verbosity level of joblib
kwargs['verbose'] = 1 if verbose in ['debug', True] else 0
# caching option
if isinstance(cache_dir, str) and os.path.isdir(cache_dir):
logger.info(f'Caching computations to {cache_dir}')
memory = Memory(cache_dir, verbose=kwargs['verbose'])
fcn = memory.cache(fcn)
# parallel functions
para_fcn = delayed(fcn)
parallel = Parallel(n_jobs, **kwargs)
if total is not None:
def parallel_progress(op_iter):
return parallel(ProgressBar(iterable=op_iter, max_value=total,
mesg=mesg))
parallel_out = parallel_progress
else:
parallel_out = parallel
return parallel_out, para_fcn
def run_dmri_pipeline(subject_session, do_topup=True, do_edc=True):
subject, session = subject_session
data_dir = os.path.join(source_dir, subject, session, 'dwi')
write_dir = os.path.join(derivatives_dir, subject, session)
dwi_dir = os.path.join(write_dir, 'dwi')
# Apply topup to the images
input_imgs = sorted(glob.glob('%s/sub*.nii.gz' % data_dir))
dc_imgs = sorted(glob.glob(os.path.join(dwi_dir, 'dcsub*run*.nii.gz')))
mem = Memory('/neurospin/tmp/bthirion/cache_dir')
if len(dc_imgs) < len(input_imgs):
se_maps = [
os.path.join(source_dir, subject, session, 'fmap',
'%s_%s_dir-ap_epi.nii.gz' % (subject, session)),
os.path.join(source_dir, subject, session, 'fmap',
'%s_%s_dir-pa_epi.nii.gz' % (subject, session))]
if do_topup:
fsl_topup(se_maps, input_imgs, mem, write_dir, 'dwi')
# Then proceeed with Eddy current correction
# get the images
dc_imgs = sorted(glob.glob(os.path.join(dwi_dir, 'dc*run*.nii.gz')))
dc_img = os.path.join(dwi_dir, 'dc%s_%s_dwi.nii.gz' % (subject, session))
concat_images(dc_imgs, dc_img)
# get the bvals/bvec
file_bvals = sorted(glob.glob('%s/sub*.bval' % data_dir))
bvals = np.concatenate([np.loadtxt(fbval) for fbval in sorted(file_bvals)])
bvals_file = os.path.join(dwi_dir, 'dc%s_%s_dwi.bval' % (subject, session))
np.savetxt(bvals_file, bvals)
file_bvecs = sorted(glob.glob('%s/sub*.bvec' % data_dir))
bvecs = np.hstack([np.loadtxt(fbvec) for fbvec in sorted(file_bvecs)])
bvecs_file = os.path.join(dwi_dir, 'dc%s_%s_dwi.bvec' % (subject, session))
np.savetxt(bvecs_file, bvecs)
# Get eddy-preprocessed images
# eddy_img = nib.load(glob.glob(os.path.join(dwi_dir, 'eddc*.nii*'))[-1])
# Get eddy-preprocessed images
eddy_img = mem.cache(eddy_current_correction)(
dc_img, bvals_file, bvecs_file, dwi_dir, mem)
# load the data
gtab = gradient_table(bvals, bvecs, b0_threshold=10)
# Create a brain mask
from dipy.segment.mask import median_otsu
b0_mask, mask = median_otsu(eddy_img.get_data(), 2, 1)
if subject == 'sub-13':
from nilearn.masking import compute_epi_mask
from nilearn.image import index_img
imgs_ = [index_img(eddy_img, i)
for i in range(len(bvals)) if bvals[i] < 50]
mask_img = compute_epi_mask(imgs_, upper_cutoff=.8)
mask_img.to_filename('/tmp/mask.nii.gz')
mask = mask_img.get_data()
# do the tractography
streamlines = tractography(eddy_img, gtab, mask, dwi_dir)
return streamlines
def main():
np.random.seed(10)
n = 100
m = 100
sampling = 0.1
eps = 1
x_sampler = Sampler(mean=np.array([[1.], [2], [3]]),
cov=np.array([[[.1]], [[.1]], [[.1]]]),
p=np.ones(3) / 3)
y_sampler = Sampler(mean=np.array([[0.], [3], [5]]),
cov=np.array([[[.1]], [[.1]], [[.4]]]),
p=np.ones(3) / 3)
x = x_sampler(n)
y = y_sampler(m)
mem = Memory(expanduser('~/cache'), verbose=0)
fref, gref, refrecords = mem.cache(sinkhorn)(x,
y,
n_iter=1000,
sampling=1.,
eps=eps)
for pin_potential in [False]:
for avg_step_size in ['1/sqrt(t)']:
f, g, records = mem.cache(sinkhorn)(x,
y,
n_iter=int(1e6),
record_every=1000,
step_size='1/sqrt(t)',
avg_step_size=avg_step_size,
sampling=sampling,
pin_potential=pin_potential,
eps=eps)
norm = var_norm(refrecords['f'][-1][None, :] - records['f'],
axis=1)
plt.plot(range(len(norm)),
norm,
label=f'pin_pot{pin_potential} avg{avg_step_size}')
plt.xscale('log')
# plt.yscale('log')
plt.legend()
plt.show()
def multiprocess_with_cache(inputs):
"""
Compute multiprocess with one memory mapped cache.
"""
location = "./cachedir"
memory = Memory(location, verbose=0)
f_memoized = memory.cache(f)
return Parallel(n_jobs=-1)(delayed(f_memoized)(x) for x in inputs)
def run(aspect, word2vec, trained_model, gpu, out, test, batchsize,
sparsity_coef, coherent_coef, dependent, order):
"""
Train "Rationalizing Neural Predictions" for one specified aspect.
Please refer README.md for details.
"""
memory = Memory(cachedir='.', verbose=1)
if os.path.splitext(test)[-1] == '.json':
w2v, vocab, _, _, test_dataset = \
memory.cache(prepare_data)(None, word2vec, aspect, annotation=test)
elif os.path.splitext(test)[-1] == '.gz':
w2v, vocab, test_dataset, _, _ = \
memory.cache(prepare_data)(test, word2vec, aspect)
else:
raise ValueError(
"Input 'test' must be either json file or gzipped text file with"
" appropriate extension."
)
encoder = rationale.models.Encoder(
w2v.shape[1], order, 200, 2, dropout=0.1
)
generator_cls = (rationale.models.GeneratorDependent
if dependent else rationale.models.Generator)
# Original impl. uses two layers to model bi-directional LSTM
generator = generator_cls(w2v.shape[1], order, 200, dropout=0.1)
model = rationale.models.RationalizedRegressor(
generator, encoder, w2v.shape[0], w2v.shape[1], initialEmb=w2v,
dropout_emb=0.1,
sparsity_coef=sparsity_coef, coherent_coef=coherent_coef
)
# Resume from a snapshot
chainer.serializers.load_npz(trained_model, model)
if gpu >= 0:
# Make a specified GPU current
chainer.cuda.get_device_from_id(gpu).use()
model.to_gpu() # Copy the model to the GPU
inv_vocab = {v: k for k, v in vocab.items()}
results = rationale.prediction.test(model, test_dataset, inv_vocab,
device=gpu, batchsize=batchsize)
with open(os.path.join(out, 'output.json'), 'w') as fout:
json.dump(results, fout)
def load_adni_longitudinal_csf_biomarker():
""" Returns longitudinal csf measures
"""
BASE_DIR = _get_data_base_dir('ADNI_csv')
roster = pd.read_csv(os.path.join(BASE_DIR, 'ROSTER.csv'))
dx = pd.read_csv(os.path.join(BASE_DIR, 'DXSUM_PDXCONV_ADNIALL.csv'))
csf_files = ['UPENNBIOMK.csv', 'UPENNBIOMK2.csv', 'UPENNBIOMK3.csv',
'UPENNBIOMK4_09_06_12.csv', 'UPENNBIOMK5_10_31_13.csv',
'UPENNBIOMK6_07_02_13.csv', 'UPENNBIOMK7.csv',
'UPENNBIOMK8.csv']
cols = ['RID', 'VISCODE', 'ABETA', 'PTAU', 'TAU']
# 3,4,5,7,8
csf = pd.DataFrame()
for csf_file in csf_files[2:]:
fs = pd.read_csv(os.path.join(BASE_DIR, csf_file))
csf = csf.append(fs[cols])
# remove nans from csf values
biom = csf[cols[2:]].values
idx = np.array([~np.isnan(v).any() for v in biom])
biom = biom[idx]
# get phenotype
vcodes = csf['VISCODE'].values[idx]
rids = csf['RID'].values[idx]
# caching dataframe extraction functions
CACHE_DIR = _get_cache_base_dir()
cache_dir = os.path.join(CACHE_DIR, 'joblib', 'load_data_cache')
if not os.path.isdir(cache_dir):
os.makedirs(cache_dir)
memory = Memory(cachedir=cache_dir, verbose=0)
def _getptidscsf(rids):
return list(map(lambda x: _rid_to_ptid(x, roster), rids))
ptids = memory.cache(_getptidscsf)(rids)
# get diagnosis
def _getdxcsf(rids, vcodes):
return list(map(lambda x, y: DX_LIST[_get_dx(x, dx, viscode=y)],
rids, vcodes))
dx_group = memory.cache(_getdxcsf)(rids, vcodes)
return Bunch(dx_group=np.array(dx_group), subjects=np.array(ptids),
csf=np.array(biom), exam_codes=np.array(vcodes),
exam_codes2=np.array(vcodes))
def construct_and_attach_filename_data(self):
synsets = self.synset_list
num_per_synset = self.data['num_per_synset']
seed = self.data['seed']
folder = self.local_home('PrecomputedDicts')
mem = Memory(folder)
compute_filename_dict = mem.cache(self.compute_filename_dict)
filenames, filenames_dict = compute_filename_dict(synsets, num_per_synset, seed)
self.filenames_dict = filenames_dict
def construct_and_attach_filename_data(self):
synsets = self.synset_list
num_per_synset = self.data['num_per_synset']
seed = self.data['seed']
folder = self.local_home('PrecomputedDicts')
mem = Memory(folder)
compute_filename_dict = mem.cache(self.compute_filename_dict)
filenames, filenames_dict = compute_filename_dict(synsets, num_per_synset, seed)
self.filenames_dict = filenames_dict
def __init__(self, data_source='yahoo'):
''' Creates the object and sets up caching.
:param data_source: A data souce such as `yahoo` or `google`.
'''
self.data_source = data_source
# caching
memory = Memory(cachedir='.')
self.get = memory.cache(self.get)
def get_multilabel(self):
cache = Memory(location=tempfile.gettempdir())
cached_func = cache.cache(make_multilabel_classification)
return cached_func(n_samples=100,
n_features=10,
n_classes=5,
n_labels=5,
return_indicator=True,
random_state=1)
def add_caching_to_funcs(obj, funcNames):
mem = Memory('../.add_caching_to_funcs', verbose=11)
if obj is None or funcNames is None:
return
if isScalar(funcNames):
funcNames = [funcNames]
for name in funcNames:
func = getattr(obj, name, None)
if func is not None:
setattr(obj, name, mem.cache(func))
def _forward(self, im, indices):
memory = Memory(location=self.memory, verbose=0)
_apply_transform_cached = memory.cache(_apply_xform)
logger.info('Applying forward transformations in pipeline')
for xform in self.xforms:
im = _apply_transform_cached(xform, im, indices, False)
logger.info('All forward transformations applied')
return im
def add_caching_to_funcs(obj, funcNames):
mem = Memory('../.add_caching_to_funcs', verbose=11)
if obj is None or funcNames is None:
return
if isScalar(funcNames):
funcNames = [funcNames]
for name in funcNames:
func = getattr(obj, name, None)
if func is not None:
setattr(obj, name, mem.cache(func))
def __init__(self):
self.name = self.__class__.__name__
try:
from joblib import Memory
mem = Memory(cachedir=self.home('cache'), verbose=False)
self._get_meta = mem.cache(self._get_meta)
except ImportError:
pass
def load_adni_longitudinal_mmse_score():
""" Returns longitudinal mmse scores
"""
BASE_DIR = _get_data_base_dir('ADNI_csv')
roster = pd.read_csv(os.path.join(BASE_DIR, 'ROSTER.csv'))
dx = pd.read_csv(os.path.join(BASE_DIR, 'DXSUM_PDXCONV_ADNIALL.csv'))
fs = pd.read_csv(os.path.join(BASE_DIR, 'MMSE.csv'))
# extract nans free mmse
mmse = fs['MMSCORE'].values
idx_num = fs['MMSCORE'].notnull().values
mmse = mmse[idx_num]
# extract roster id
rids = fs['RID'].values[idx_num]
# caching dataframe extraction functions
CACHE_DIR = _get_cache_base_dir()
cache_dir = os.path.join(CACHE_DIR, 'joblib', 'load_data_cache')
if not os.path.isdir(cache_dir):
os.makedirs(cache_dir)
memory = Memory(cachedir=cache_dir, verbose=0)
def _getptidsmmse(rids):
return [_rid_to_ptid(rid, roster) for rid in rids]
# get subject id
ptids = memory.cache(_getptidsmmse)(rids)
# extract visit code (don't use EXAMDATE ; null for GO/2)
vcodes = fs['VISCODE'].values
vcodes = vcodes[idx_num]
vcodes2 = fs['VISCODE2'].values
vcodes2 = vcodes2[idx_num]
def _getdxmmse(rids, vcodes2):
return list(map(
lambda x, y: DX_LIST[_get_dx(x, dx, viscode=y)], rids, vcodes2))
# get diagnosis
dx_group = memory.cache(_getdxmmse)(rids, vcodes2)
return Bunch(dx_group=np.array(dx_group), subjects=np.array(ptids),
mmse=mmse, exam_codes=vcodes, exam_codes2=vcodes2)
def _adjoint(self, im, indices):
memory = Memory(location=self.memory, verbose=0)
_apply_transform_cached = memory.cache(_apply_xform)
logger.info('Applying adjoint transformations in pipeline')
for xform in self.xforms[::-1]:
im = _apply_transform_cached(xform, im, indices, True)
logger.info('All adjoint transformations applied')
return im
def _run_suject_level1_glm(subject_data_dir, subject_output_dir,
**kwargs):
"""
Just another wrapper.
"""
mem = Memory(os.path.join(subject_output_dir, "cache_dir"))
return mem.cache(run_suject_level1_glm)(subject_data_dir,
subject_output_dir,
**kwargs)
def __init__(self, meta=None):
if meta is not None:
self._meta = meta
self.name = self.__class__.__name__
try:
from joblib import Memory
mem = Memory(cachedir=self.home('cache'))
self._get_meta = mem.cache(self._get_meta)
except ImportError:
pass
def load_features_cached(cache_dir,
participant_label,
feature_base_dir,
clinical_feature_file,
modality='clinical'):
memory = Memory(cache_dir / "load", verbose=0)
allowed_modalities = [
"clinical", "structural", "structGlobScort", "functional", "fullcon"
]
modality_parts = modality.split("+")
for m in modality_parts:
if m not in allowed_modalities:
raise ValueError(m, modality)
features, feature_files = memory.cache(stack_subjects_features)(
participant_label, feature_base_dir, clinical_feature_file, modality)
stacked_features = memory.cache(concat_modality_features)(features,
modality)
return stacked_features
def single(n_players, n_actions, n_matrices, _seed, conditioning, skewness,
l1_penalty, gaussian_noise, stochastic_noise, _run):
mem = Memory(location=expanduser('~/cache'))
H = mem.cache(make_positive_matrices)(n_players, n_actions, n_matrices,
conditioning, skewness,
stochastic_noise, _seed)
game = MatrixGame(H, l1_penalty=l1_penalty, gaussian_noise=gaussian_noise)
values, policies = compute_nash(game)
_run.info['policies'] = policies.tolist()
_run.info['values'] = values.tolist()
def getagreement(tpl,datadir,task_type='all'):
"""Get agreement values for annotators in the :data:'tpl' list
Args:
tpl (list): combination group of annotators
datadir (str): Cache data directory used by joblib
Returns:
namedtuple defined as ``Agree = collections.namedtuple('Agree', ['kappa', 'alpha','avg_ao'], verbose=True)``
"""
mem = Memory(cachedir=datadir)
readjson=mem.cache(json2taskdata.readjson,mmap_mode='r')
create_task_data= mem.cache(json2taskdata.create_task_data)
count_occurrances=mem.cache(json2taskdata.count_occurrances)
count_labels=mem.cache(json2taskdata.count_labels)
annotators=set()
lectask=[]
#-------------------------------------------------------------------------------
# for each annotator in group tpl
#-------------------------------------------------------------------------------
for stditem in tpl:
aname=stditem.split('.')[0][3:][-2:]
annotators.add(aname)
lecdict=readjson(stditem)
newlectask= create_task_data(lecdict,task_type=task_type,annotator=aname)
label_data=json2taskdata.create_labels_list(newlectask)
abscount=count_occurrances(str(label_data))
yaml.dump(abscount,open(os.path.join( datadir,'abscount-'+aname+'.yaml'),'w'))
setcount=count_labels(newlectask)
yaml.dump(setcount,open(os.path.join( datadir,'setcount-'+aname+'.yaml'),'w'))
lectask=lectask+newlectask
task=AnnotationTask(data=lectask,distance=nltk.metrics.distance.masi_distance_mod)
return {frozenset(annotators): Agree(task.kappa(),task.alpha(),task.avg_Ao())}
def make_dictionary(X,
n_components=20,
alpha=5.,
write_dir='/tmp/',
contrasts=[],
method='multitask',
l1_ratio=.5,
n_subjects=13):
"""Create dictionary + encoding"""
from sklearn.decomposition import dict_learning_online, sparse_encode
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import MultiTaskLasso, MultiTaskElasticNet
mem = Memory(write_dir, verbose=0)
dictionary = mem.cache(initial_dictionary)(n_components, X)
np.savez(os.path.join(write_dir, 'dictionary.npz'),
loadings=dictionary,
contrasts=contrasts)
if method == 'online':
components, dictionary = dict_learning_online(X.T,
n_components,
alpha=alpha,
dict_init=dictionary,
batch_size=200,
method='cd',
return_code=True,
shuffle=True,
n_jobs=1,
positive_code=True)
np.savez(os.path.join(write_dir, 'dictionary.npz'),
loadings=dictionary,
contrasts=contrasts)
elif method == 'sparse':
components = sparse_encode(X.T,
dictionary,
alpha=alpha,
max_iter=10,
n_jobs=1,
check_input=True,
verbose=0,
positive=True)
elif method == 'multitask':
# too many hard-typed parameters !!!
n_voxels = X.shape[1] // n_subjects
components = np.zeros((X.shape[1], n_components))
clf = MultiTaskLasso(alpha=alpha)
clf = MultiTaskElasticNet(alpha=alpha, l1_ratio=l1_ratio)
for i in range(n_voxels):
x = X[:, i:i + n_subjects * n_voxels:n_voxels]
components[i: i + n_subjects * n_voxels: n_voxels] =\
clf.fit(dictionary.T, x).coef_
return dictionary, components
def init_phd_lab(cls, directory: phd_lab_directory) -> None:
if cls.phd_lab_directory is not None:
raise RuntimeError("Cannot initialize PhdLabWrapper "
"a second time!")
cls.phd_lab_directory = directory
memcache = os.path.join(cls.phd_lab_directory, 'memcache')
if not os.path.exists(memcache):
os.makedirs(memcache)
memory = Memory(".memcache", verbose=True)
global fit_with_cache
fit_with_cache = memory.cache(fit_with_cache)
def main():
## subsdir=r'E:\elan projects\L2\submissions\extracted'
## dstdir=os.path.join(subsdir,r'passed')
## copypassedfiles(dstdir,subsdir)
dstdir=r'E:\elan projects\L2\resubmission\full'
import glob
jsonflist=glob.glob(dstdir+'\\'+r'*.379.json')
mem = Memory(cachedir=dstdir)
json2agreementmatrix_cached=mem.cache(json2agreementmatrix)
c=json2agreementmatrix_cached(jsonflist,task_type='all')
print c
def __init__(self, use_cache=True, cachedir=None):
"""Inits TpsSolverFactory
Args:
use_cache: whether to cache solver matrices in file
cache_dir: cached directory. if not specified, the .cache directory in parent directory of top-level package is used.
"""
if use_cache:
if cachedir is None:
# .cache directory in parent directory of top-level package
cachedir = os.path.join(__import__(__name__.split('.')[0]).__path__[0], os.path.pardir, ".cache")
memory = Memory(cachedir=cachedir, verbose=0)
self.get_solver_mats = memory.cache(self.get_solver_mats)
def _load_data(root_dir="/",
data_set="ds107",
cache_dir="/volatile/storage/workspace/parietal_retreat/" +
"covariance_learn/cache/",
n_jobs=1):
from joblib import Memory
mem = Memory(cachedir=cache_dir)
load_data_ = mem.cache(setup_data_paths.run)
df = setup_data_paths.get_all_paths(root_dir=root_dir, data_set=data_set)
# region_signals = joblib.load(os.path.join(root_dir, dump_file))
region_signals = load_data_(root_dir=root_dir, data_set=data_set,
n_jobs=n_jobs,
dump_dir=os.path.join(cache_dir, data_set))
return df, region_signals
def _delete_orientation(self):
"""
Delete orientation metadata. Garbage orientation metadata can lead to
severe mis-registration trouble.
"""
# prepare for smart caching
cache_dir = os.path.join(self.output_dir, 'cache_dir')
if not os.path.exists(cache_dir):
os.makedirs(cache_dir)
mem = Memory(cachedir=cache_dir, verbose=5)
# deleteorient for func
self.func = [mem.cache(delete_orientation)(
self.func[j],
self.tmp_output_dir,
output_tag=self.session_id[j])
for j in xrange(len(self.session_id))]
# deleteorient for anat
if not self.anat is None:
self.anat = mem.cache(delete_orientation)(
self.anat, self.tmp_output_dir)
def __init__(self, caching=False):
"""Create a new CompatIdFetcher object.
Args:
caching: Whether to cache setup from run to run. See
PrebuiltCompatibilityTest.CACHING for details.
"""
self.compat_ids = None
if caching:
# This import occurs here rather than at the top of the file because we
# don't want to force developers to install joblib. The caching argument
# is only set to True if PrebuiltCompatibilityTest.CACHING is hand-edited
# (for testing purposes).
# pylint: disable=import-error
from joblib import Memory
memory = Memory(cachedir=tempfile.gettempdir(), verbose=0)
self.FetchCompatIds = memory.cache(self.FetchCompatIds)
def ica_step(group_maps, group_variance, cachedir=None):
memory = Memory(cachedir=cachedir, mmap_mode='r')
# We do a spatial ICA: the arrays are transposed in the following,
# axis1 = component, and axis2 is voxel number.
_, ica_maps = memory.cache(fastica)(group_maps.T, whiten=False)
# Project the ICAs on the group maps to give a 'cross-subject
# reproducibility' score.
proj = np.dot(ica_maps, group_maps)
reproducibility_score = (np.abs(proj)*group_variance).sum(axis=-1)
order = np.argsort(reproducibility_score)[::-1]
ica_maps = ica_maps[order, :]
return ica_maps.T
class _DiskCache(object):
cached_methods = methods
def __init__(self, *args, **kwargs):
from tempfile import mkdtemp
from joblib import Memory
self.cachedir = cachedir or mkdtemp()
self.memory = Memory(cachedir=self.cachedir)
for method in self.cached_methods:
setattr(self, method, self.memory.cache(getattr(self, method)))
if not os.path.isdir(self.cachedir):
raise OSError("Non-existent directory: ", self.cachedir)
super(_DiskCache, self).__init__(*args, **kwargs)
def __init__(self, meta=None, seed=0, ntrain=15, ntest=15, num_splits=10):
self.seed = seed
self.ntrain = ntrain
self.ntest = ntest
self.num_splits = num_splits
if meta is not None:
self._meta = meta
self.name = self.__class__.__name__
try:
from joblib import Memory
mem = Memory(cachedir=self.home('cache'))
self._get_meta = mem.cache(self._get_meta)
except ImportError:
pass
def __init__(self, root,
filter_species_ids=None,
required_attributes=None,
transform=None,
is_training=False,
cachedir=CACHE_DIR):
super(GogglesDataset, self).__init__()
mem = Memory(cachedir)
metadata_loader = mem.cache(self._load_metadata)
self.is_training = is_training
self._data_dir = root
required_species, \
self.attributes, \
self._image_data = metadata_loader(root) # _load_metadata(root) cached
if filter_species_ids is not None:
assert type(filter_species_ids) is list
filter_species_ids = set(filter_species_ids)
required_species = list(filter(lambda s: s.id in filter_species_ids, required_species))
self._image_data = list(filter(lambda d: d.species.id in filter_species_ids, self._image_data))
self._species_labels = {species: label for label, species in enumerate(required_species)}
if is_training is not None:
self._image_data = list(filter(
lambda d: d.is_for_training == is_training,
self._image_data))
if required_attributes is not None:
assert type(required_attributes) is list
self.attributes = required_attributes
elif filter_species_ids is not None:
attributes = set()
for species in required_species:
attributes = attributes.union(species.attributes)
self.attributes = list(sorted(attributes, key=lambda a: a.id))
self.num_attributes = len(self.attributes)
if transform is not None:
self._transform = transform
else:
self._transform = transforms.Compose([transforms.ToTensor()])
def __init__(self, meta=None, seed=0, ntrain=10, ntest=10, num_splits=5):
self.seed = seed
self.ntrain = ntrain
self.ntest = ntest
self.num_splits = num_splits
self.names = ["Face", "Body", "Object"]
if meta is not None:
self._meta = meta
self.name = self.__class__.__name__
try:
from joblib import Memory
mem = Memory(cachedir=self.home("cache"))
self._get_meta = mem.cache(self._get_meta)
except ImportError:
pass
def test_cached(self):
try:
from joblib import Memory
mem = Memory(self.cache_dir)
dep_tree = {
'a': 5,
'b': 6,
'c': mem.cache(slow_func),
}
data = Pipeline(dep_tree)
t0 = time.time()
data.resolve()
delta = time.time() - t0
t0 = time.time()
data.resolve()
delta = time.time() - t0
assert delta < .1
except:
pass
def compute_confidence_par(allLearners, dada):
lab_confidence = np.zeros([dada.shape[0], len(allLearners)])
tic = time.time()
#import ipdb;ipdb.set_trace()
print 'producing weighted outputs IN PARALLEL'
mem = Memory(cachedir='tmp')
classif_RBF2 = mem.cache(confidence_par)
c = l_c[0]
r = Parallel(n_jobs=N_JOBS)(delayed(confidence_par)(allLearners,ii,dada) for ii in enumerate(allLearners))
res, iis = zip(*r)
for t,y in enumerate(iis):
lab_confidence[:,y] = res[t]
print "time taken to produce confidence:", round(time.time() - tic,2), "seconds"
#import ipdb;ipdb.set_trace()
return lab_confidence
def __init__(
self,
segmentation=None,
duration=1., step=0.1, gap=0., threshold=0.,
feature=None, cache=False
):
super(SpeechTurnSegmentation, self).__init__()
if segmentation is None:
self.segmentation = SegmentationGaussianDivergence(
duration=duration, step=step, gap=gap,
threshold=threshold
)
else:
self.segmentation = segmentation
# default features for segmentation
# are MFCC (energy + 12 coefficients)
if feature is None:
from pyannote.feature.yaafe import YaafeMFCC
feature = YaafeMFCC(
e=True, De=False, DDe=False,
coefs=12, D=False, DD=False
)
self.feature = feature
if cache:
# initialize cache
from joblib import Memory
from tempfile import mkdtemp
memory = Memory(cachedir=mkdtemp(), verbose=0)
# cache feature extraction method
self.get_features = memory.cache(self.get_features)
def svm_cla_sklearn_feat_sel(features_train, features_test, labels_train, labels_test):
from sklearn.feature_selection import SelectPercentile, SelectKBest, f_classif, RFECV
from sklearn.cross_validation import StratifiedKFold
from sklearn.metrics import zero_one_loss
features_train = sp.array(features_train, dtype = 'uint8')
features_test = sp.array(features_test, dtype = 'uint8')
print "zscore features"
tic = time.time()
features_train, mean_f, std_f = features_preprocessing(features_train)
features_test, mean_f, std_f = features_preprocessing(features_test, mean_f, std_f)
print "time taken to zscore data is:", round(time.time() - tic) , "seconds"
featSize = np.shape(features_train)
selector = LinearSVC(C=0.0007, penalty="l1", dual=False).fit(features_train, labels_train)
print 'Starting with %d samp, %d feats, keeping %d' % (featSize[0], featSize[1], (np.shape(selector.transform(features_train)))[1])
print 'classifying'
features_train = selector.transform(features_train)
features_test = selector.transform(features_test)
#import ipdb; ipdb.set_trace()
mem = Memory(cachedir='tmp')
classif_RBF2 = mem.cache(classif_RBF)
c = l_c[0]
Parallel(n_jobs=8)(delayed(classif_RBF2)(features_train, features_test, labels_train, labels_test, g, c) for g in l_g)
#import ipdb; ipdb.set_trace()
print "Starting CONTROL classification for c = ", c
tic = time.time()
clf = SVC(C=c)
clf.fit(features_train, labels_train) #[:1960][:]
score = clf.score(features_test, labels_test) #[:13841][:]
print "selected CONTROL score for c = ", c, "is: ", score
print "time taken:", time.time() - tic, "seconds"
test_func = mem.cache(test_func)
Parallel(n_jobs=1)(delayed(test_func)(i) for i in [a, a, a])
Parallel(n_jobs=2)(delayed(test_func)(i) for i in [a, a, a])
### Can use with latest version on github
from joblib import Parallel, delayed
import numpy as np
a = np.memmap('/tmp/memmaped', dtype=np.float32, mode='w+', shape=(3, 5))
b = np.memmap('/tmp/memmaped', dtype=np.float32, mode='r', shape=(3, 5))
Parallel(n_jobs=2)(delayed(np.mean)(x) for x in np.array_split(b, 3))
cachedir2 = mkdtemp()
memory2 = Memory(cachedir=cachedir2, mmap_mode='r')
square = memory2.cache(np.square)
a = np.vander(np.arange(3)).astype(np.float)
square(a)
import joblib
import numpy as np
testarray = {}
for i in xrange(5):
testarray[i] = convert2memmap(np.array(range(500*100)))
filepath = "/tmp/test.joblib"
res = joblib.dump(testarray, filepath)
testarray = joblib.load(filepath, mmap_mode="r+")
for key in testarray:
def cfunc(*fargs, **fkwargs):
return Memory.cache(self, func, *args, **kwargs).__call__(*fargs, **fkwargs)