def extract_csv(classification_csv,
config,
output_dir=os.path.abspath('.'),
output_name='extractions',
order=False,
verbose=False):
config = get_file_instance(config)
with config as config_in:
config_yaml = yaml.load(config_in, Loader=yaml.SafeLoader)
extractor_config = config_yaml['extractor_config']
workflow_id = config_yaml['workflow_id']
version = config_yaml['workflow_version']
blank_extracted_data = OrderedDict([('classification_id', []),
('user_name', []), ('user_id', []),
('workflow_id', []), ('task', []),
('created_at', []), ('subject_id', []),
('extractor', []), ('data', [])])
extracted_data = {}
classification_csv = get_file_instance(classification_csv)
with classification_csv as classification_csv_in:
classifications = pandas.read_csv(classification_csv_in,
encoding='utf-8',
dtype={'workflow_version': str})
wdx = classifications.workflow_id == workflow_id
assert (
wdx.sum() >
0), 'There are no classifications matching the configured workflow ID'
if '.' in version:
vdx = classifications.workflow_version == version
else:
vdx = classifications.workflow_version.apply(
get_major_version) == version
assert (
vdx.sum() > 0
), 'There are no classificaitons matching the configured version number'
assert (
(vdx & wdx).sum() > 0
), 'There are no classifications matching the combined workflow ID and version number'
widgets = [
'Extracting: ',
progressbar.Percentage(), ' ',
progressbar.Bar(), ' ',
progressbar.ETA()
]
pbar = progressbar.ProgressBar(widgets=widgets,
max_value=(wdx & vdx).sum())
counter = 0
pbar.start()
for cdx, classification in classifications[wdx & vdx].iterrows():
classification_by_task = annotation_by_task(
{'annotations': json.loads(classification.annotations)})
for extractor_name, keywords in extractor_config.items():
extractor_key = extractor_name
if 'shape_extractor' in extractor_name:
extractor_key = 'shape_extractor'
for keyword in keywords:
if extractor_key in extractors.extractors:
try:
extract = extractors.extractors[extractor_key](
copy.deepcopy(classification_by_task), **keyword)
except:
if verbose:
print()
print('Incorrectly formatted annotation')
print(classification)
print(extractor_key)
print(classification_by_task)
print(keyword)
continue
if isinstance(extract, list):
for e in extract:
extracted_data.setdefault(
extractor_name,
copy.deepcopy(blank_extracted_data))
extracted_data[extractor_name][
'classification_id'].append(
classification.classification_id)
extracted_data[extractor_name]['user_name'].append(
classification.user_name)
extracted_data[extractor_name]['user_id'].append(
classification.user_id)
extracted_data[extractor_name][
'workflow_id'].append(
classification.workflow_id)
extracted_data[extractor_name]['task'].append(
keyword['task'])
extracted_data[extractor_name][
'created_at'].append(classification.created_at)
extracted_data[extractor_name][
'subject_id'].append(
classification.subject_ids)
extracted_data[extractor_name]['extractor'].append(
extractor_name)
extracted_data[extractor_name]['data'].append(e)
else:
extracted_data.setdefault(
extractor_name,
copy.deepcopy(blank_extracted_data))
extracted_data[extractor_name][
'classification_id'].append(
classification.classification_id)
extracted_data[extractor_name]['user_name'].append(
classification.user_name)
extracted_data[extractor_name]['user_id'].append(
classification.user_id)
extracted_data[extractor_name]['workflow_id'].append(
classification.workflow_id)
extracted_data[extractor_name]['task'].append(
keyword['task'])
extracted_data[extractor_name]['created_at'].append(
classification.created_at)
extracted_data[extractor_name]['subject_id'].append(
classification.subject_ids)
extracted_data[extractor_name]['extractor'].append(
extractor_name)
extracted_data[extractor_name]['data'].append(extract)
counter += 1
pbar.update(counter)
pbar.finish()
# create one flat csv file for each extractor used
output_base_name, output_ext = os.path.splitext(output_name)
output_files = []
for extractor_name, data in extracted_data.items():
output_path = os.path.join(
output_dir, '{0}_{1}.csv'.format(extractor_name, output_base_name))
output_files.append(output_path)
flat_extract = flatten_data(data)
if order:
flat_extract = order_columns(flat_extract, front=['choice'])
flat_extract.to_csv(output_path, index=False, encoding='utf-8')
return output_files
def reduce_csv(
extracted_csv,
reducer_config,
filter='first',
output_name='reductions',
output_dir=CURRENT_PATH,
order=False,
stream=False,
cpu_count=1
):
extracted_csv = get_file_instance(extracted_csv)
with extracted_csv as extracted_csv_in:
extracted = pandas.read_csv(
extracted_csv_in,
infer_datetime_format=True,
parse_dates=['created_at'],
encoding='utf-8'
)
extracted.sort_values(['subject_id', 'created_at'], inplace=True)
resume = False
subjects = extracted.subject_id.unique()
tasks = extracted.task.unique()
workflow_id = extracted.workflow_id.iloc[0]
reducer_config = get_file_instance(reducer_config)
with reducer_config as config:
config_yaml = yaml.load(config, Loader=yaml.SafeLoader)
assert (len(config_yaml['reducer_config']) == 1), 'There must be only one reducer in the config file.'
for key, value in config_yaml['reducer_config'].items():
reducer_name = key
keywords = value
assert (reducer_name in reducers.reducers), 'The reducer in the config files does not exist.'
output_base_name, _ = os.path.splitext(output_name)
output_path = os.path.join(output_dir, '{0}_{1}.csv'.format(reducer_name, output_base_name))
if stream:
if os.path.isfile(output_path):
print('resuming from last run')
resume = True
with open(output_path, 'r', encoding='utf-8') as reduced_file:
reduced_csv = pandas.read_csv(reduced_file, encoding='utf-8')
subjects = np.setdiff1d(subjects, reduced_csv.subject_id)
reduced_data = []
sdx = 0
apply_keywords = {
'reducer_name': reducer_name,
'workflow_id': workflow_id,
'filter': filter,
'keywords': keywords
}
widgets = [
'Reducing: ',
progressbar.Percentage(),
' ', progressbar.Bar(),
' ', progressbar.ETA()
]
number_of_rows = len(subjects) * len(tasks)
pbar = progressbar.ProgressBar(widgets=widgets, max_value=number_of_rows)
def callback(reduced_data_list):
nonlocal reduced_data
nonlocal sdx
nonlocal pbar
nonlocal stream
reduced_data += reduced_data_list
if stream:
if (sdx == 0) and (not resume):
pandas.DataFrame(reduced_data).to_csv(
output_path,
mode='w',
index=False,
encoding='utf-8'
)
else:
pandas.DataFrame(reduced_data).to_csv(
output_path,
mode='a',
index=False,
header=False,
encoding='utf-8'
)
reduced_data.clear()
sdx += 1
pbar.update(sdx)
pbar.start()
if cpu_count > 1:
pool = Pool(cpu_count)
for subject in subjects:
idx = extracted.subject_id == subject
for task in tasks:
jdx = extracted.task == task
classifications = extracted[idx & jdx]
if cpu_count > 1:
pool.apply_async(
reduce_subject,
args=(
subject,
classifications,
task
),
kwds=apply_keywords,
callback=callback
)
else:
reduced_data_list = reduce_subject(
subject,
classifications,
task,
**apply_keywords
)
callback(reduced_data_list)
if cpu_count > 1:
pool.close()
pool.join()
pbar.finish()
if stream:
reduced_csv = pandas.read_csv(output_path, encoding='utf-8')
if 'data' in reduced_csv:
def eval_func(a):
# pandas uses a local namespace, make sure it has the correct imports
from collections import OrderedDict # noqa
from numpy import nan # noqa
return eval(a)
reduced_csv.data = reduced_csv.data.apply(eval_func)
flat_reduced_data = flatten_data(reduced_csv)
else:
return output_path
else:
non_flat_data = pandas.DataFrame(reduced_data)
flat_reduced_data = flatten_data(non_flat_data)
if order:
flat_reduced_data = order_columns(flat_reduced_data, front=['choice', 'total_vote_count', 'choice_count'])
flat_reduced_data.to_csv(output_path, index=False, encoding='utf-8')
return output_path
def test_flatten_data(self):
'''Test if nested `data` column flattened to many columns'''
result = csv_utils.flatten_data(nested_data)
assert_frame_equal(result, flat_data, check_like=True)
def extract_csv(
classification_csv,
config,
output_dir=CURRENT_PATH,
output_name='extractions',
order=False,
verbose=False,
cpu_count=1
):
config = get_file_instance(config)
with config as config_in:
config_yaml = yaml.load(config_in, Loader=yaml.SafeLoader)
extractor_config = config_yaml['extractor_config']
workflow_id = config_yaml['workflow_id']
version = config_yaml['workflow_version']
number_of_extractors = sum([len(value) for key, value in extractor_config.items()])
extracted_data = defaultdict(list)
classification_csv = get_file_instance(classification_csv)
with classification_csv as classification_csv_in:
classifications = pandas.read_csv(classification_csv_in, encoding='utf-8', dtype={'workflow_version': str})
wdx = classifications.workflow_id == workflow_id
assert (wdx.sum() > 0), 'There are no classifications matching the configured workflow ID'
if '.' in version:
vdx = classifications.workflow_version == version
else:
vdx = classifications.workflow_version.apply(get_major_version) == version
assert (vdx.sum() > 0), 'There are no classificaitons matching the configured version number'
assert ((vdx & wdx).sum() > 0), 'There are no classifications matching the combined workflow ID and version number'
widgets = [
'Extracting: ',
progressbar.Percentage(),
' ', progressbar.Bar(),
' ', progressbar.ETA()
]
max_pbar = (wdx & vdx).sum() * number_of_extractors
pbar = progressbar.ProgressBar(widgets=widgets, max_value=max_pbar)
counter = 0
def callback(name_with_row):
nonlocal extracted_data
nonlocal counter
nonlocal pbar
extractor_name, new_extract_row = name_with_row
if new_extract_row is not None:
extracted_data[extractor_name] += new_extract_row
counter += 1
pbar.update(counter)
pbar.start()
if cpu_count > 1:
pool = Pool(cpu_count)
for _, classification in classifications[wdx & vdx].iterrows():
classification_by_task = annotation_by_task({
'annotations': json.loads(classification.annotations),
'metadata': json.loads(classification.metadata)
})
classification_info = {
'classification_id': classification.classification_id,
'user_name': classification.user_name,
'user_id': classification.user_id,
'workflow_id': classification.workflow_id,
'created_at': classification.created_at,
'subject_ids': classification.subject_ids
}
for extractor_name, keywords in extractor_config.items():
extractor_key = extractor_name
if 'shape_extractor' in extractor_name:
extractor_key = 'shape_extractor'
for keyword in keywords:
if extractor_key in extractors.extractors:
if cpu_count > 1:
pool.apply_async(
extract_classification,
args=(
copy.deepcopy(classification_by_task),
classification_info,
extractor_key,
extractor_name,
keyword,
verbose
),
callback=callback
)
else:
name_with_row = extract_classification(
copy.deepcopy(classification_by_task),
classification_info,
extractor_key,
extractor_name,
keyword,
verbose
)
callback(name_with_row)
else:
callback((None, None))
if cpu_count > 1:
pool.close()
pool.join()
pbar.finish()
# create one flat csv file for each extractor used
output_base_name, _ = os.path.splitext(output_name)
output_files = []
for extractor_name, data in extracted_data.items():
output_path = os.path.join(output_dir, '{0}_{1}.csv'.format(extractor_name, output_base_name))
output_files.append(output_path)
non_flat_extract = pandas.DataFrame(data)
flat_extract = flatten_data(non_flat_extract)
if order:
flat_extract = order_columns(flat_extract, front=['choice'])
flat_extract.to_csv(output_path, index=False, encoding='utf-8')
return output_files
def reduce_csv(extracted_csv,
reducer_config,
filter='first',
output_name='reductions',
output_dir=os.path.abspath('.'),
order=False,
stream=False):
extracted_csv = get_file_instance(extracted_csv)
with extracted_csv as extracted_csv_in:
extracted = pandas.read_csv(extracted_csv_in,
infer_datetime_format=True,
parse_dates=['created_at'],
encoding='utf-8')
extracted.sort_values(['subject_id', 'created_at'], inplace=True)
resume = False
subjects = extracted.subject_id.unique()
tasks = extracted.task.unique()
workflow_id = extracted.workflow_id.iloc[0]
reducer_config = get_file_instance(reducer_config)
with reducer_config as config:
config_yaml = yaml.load(config, Loader=yaml.SafeLoader)
assert (len(config_yaml['reducer_config']) == 1
), 'There must be only one reducer in the config file.'
for key, value in config_yaml['reducer_config'].items():
reducer_name = key
keywords = value
assert (reducer_name in reducers.reducers
), 'The reducer in the config files does not exist.'
output_base_name, output_ext = os.path.splitext(output_name)
output_path = os.path.join(
output_dir, '{0}_{1}.csv'.format(reducer_name, output_base_name))
if stream:
if os.path.isfile(output_path):
print('resuming from last run')
resume = True
with open(output_path, 'r', encoding='utf-8') as reduced_file:
reduced_csv = pandas.read_csv(reduced_file, encoding='utf-8')
subjects = np.setdiff1d(subjects, reduced_csv.subject_id)
blank_reduced_data = OrderedDict([('subject_id', []), ('workflow_id', []),
('task', []), ('reducer', []),
('data', [])])
reduced_data = copy.deepcopy(blank_reduced_data)
widgets = [
'Reducing: ',
progressbar.Percentage(), ' ',
progressbar.Bar(), ' ',
progressbar.ETA()
]
pbar = progressbar.ProgressBar(widgets=widgets, max_value=len(subjects))
pbar.start()
for sdx, subject in enumerate(subjects):
idx = extracted.subject_id == subject
for task in tasks:
jdx = extracted.task == task
classifications = extracted[idx & jdx]
classifications = classifications.drop_duplicates()
if filter in FILTER_TYPES:
classifications = classifications.groupby(
['user_name'],
group_keys=False).apply(FILTER_TYPES[filter])
data = [unflatten_data(c) for cdx, c in classifications.iterrows()]
reduction = reducers.reducers[reducer_name](data, **keywords)
if isinstance(reduction, list):
for r in reduction:
reduced_data['subject_id'].append(subject)
reduced_data['workflow_id'].append(workflow_id)
reduced_data['task'].append(task)
reduced_data['reducer'].append(reducer_name)
reduced_data['data'].append(r)
else:
reduced_data['subject_id'].append(subject)
reduced_data['workflow_id'].append(workflow_id)
reduced_data['task'].append(task)
reduced_data['reducer'].append(reducer_name)
reduced_data['data'].append(reduction)
if stream:
if (sdx == 0) and (not resume):
pandas.DataFrame(reduced_data).to_csv(output_path,
mode='w',
index=False,
encoding='utf-8')
else:
pandas.DataFrame(reduced_data).to_csv(output_path,
mode='a',
index=False,
header=False,
encoding='utf-8')
reduced_data = copy.deepcopy(blank_reduced_data)
pbar.update(sdx + 1)
pbar.finish()
if stream:
reduced_csv = pandas.read_csv(output_path, encoding='utf-8')
if 'data' in reduced_csv:
reduced_csv.data = reduced_csv.data.apply(eval)
flat_reduced_data = flatten_data(reduced_csv)
else:
return output_path
else:
flat_reduced_data = flatten_data(reduced_data)
if order:
flat_reduced_data = order_columns(
flat_reduced_data,
front=['choice', 'total_vote_count', 'choice_count'])
flat_reduced_data.to_csv(output_path, index=False, encoding='utf-8')
return output_path