def _generate_scraper_outputs_df(self, use_dump=False):
def abs_url(url, source_url):
if url.startswith(
('../', './', '/')) or not urllib.parse.urlparse(url).scheme:
full_url = urllib.parse.urljoin(source_url, url)
return full_url
else:
return url
if self.deduplicated_list_path is None:
files = traverse_output()
else:
try:
with open(self.deduplicated_list_path, 'r') as fp:
files = [pathlib.Path(line.rstrip()) for line in fp]
except:
files = traverse_output()
df_dump = str(
pathlib.Path(
os.path.join(os.getenv('ED_OUTPUT_PATH'), 'out_df.csv')))
if use_dump:
df = pd.read_csv(df_dump)
else:
dfs = []
for fp in files:
with open(fp, 'r') as json_file:
try:
j = json.load(json_file)
# if it's marked for removal by the sanitizer, skip it
if j.get('_clean_data', dict()).get('_remove_dataset'):
logger.debug(f"Ignoring {j.get('source_url')}")
continue
j = [{
'url':
abs_url(r['url'], r['source_url']),
'source_url':
r['source_url'],
'publisher':
str(j['publisher']),
'size':
r.get('headers', dict()).get('content-length', 0),
'scraper':
fp.parent.name
} for r in j['resources']
if r['source_url'].find('/print/') == -1]
dfs.append(pd.read_json(json.dumps(j)))
except Exception as e:
logger.warning(
f'Could not parse file {json_file} as JSON! {e}')
df = pd.concat(dfs, ignore_index=True)
df.to_csv(df_dump, index=False)
return df
def __init__(self, name=None):
if name is None:
self.file_list = traverse_output()
else:
self.file_list = traverse_output(name)
# Deduplicate using a Python dict's keys uniqueness
self.urls_dict = dict()
self._make_list('source_url')
def transform(name=None, input_file=None):
if input_file is None:
file_list = h.traverse_output(name)
else:
try:
with open(input_file, 'r') as fp:
file_list = [line.rstrip() for line in fp]
except:
logger.warning(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list = h.traverse_output(name)
# loop through filepath in file list
for file_path in file_list:
# read the json data in each filepath
data = h.read_file(file_path)
if not data: # if data is None
continue
# mark as private datasets that have certain keywords in their data
data = _mark_private(data,
search_words=[
'conference', 'awards', 'user guide',
'applications'
])
# mark of removal datasets that have certain keywords
data = _remove_dataset(
data, search_words=['photo', 'foto', 'photos', 'fotos'])
# REMOVE UNWANTED STRING FROM THE VALUE OF A DATASET'S KEY
# 1. remove 'table [0-9].' from beginning of dataset title
data = _strip_unwanted_string(data,
r'^table [0-9a-z]+(-?[a-z])?\.',
dict_key='title')
# set the 'level of data' for the dataset
data = _set_dataset_level_of_data(data)
# assign the dataset to groups
# according to https://www2.ed.gov/rschstat/catalog/index.html
data = _set_dataset_groups(data)
# remove the old format for collections / sourcs
data = _remove_old_sources_collections(data)
# write modified dataset back to file
h.write_file(file_path, data)
def _generate_scraper_outputs_df(self, use_dump=False):
def abs_url(url, source_url):
if url.startswith(
('../', './', '/')) or not urllib.parse.urlparse(url).scheme:
full_url = urllib.parse.urljoin(source_url, url)
return full_url
else:
return url
if self.deduplicated_list_path is None:
files = traverse_output()
else:
try:
with open(self.deduplicated_list_path, 'r') as fp:
files = [pathlib.Path(line.rstrip()) for line in fp]
except:
files = traverse_output()
df_dump = str(
pathlib.Path(
os.path.join(os.getenv('ED_OUTPUT_PATH'), 'out_df.csv')))
if use_dump:
df = pd.read_csv(df_dump)
else:
dfs = []
for fp in files:
# TODO refactor these rules or the files structure
if 'data.json' in str(fp):
continue
with open(fp, 'r') as json_file:
try:
j = json.load(json_file)
j = [{
'url': abs_url(r['url'], r['source_url']),
'source_url': r['source_url'],
'scraper': fp.parent.name
} for r in j['resources']
if r['source_url'].find('/print/') == -1]
dfs.append(pd.read_json(json.dumps(j)))
except:
logger.warning(
f'Could not parse file {json_file} as JSON!')
df = pd.concat(dfs, ignore_index=True)
df.to_csv(df_dump, index=False)
return df
def transform(name=None, input_file=None):
"""
function is responsible for transofrming raw datasets into Sources
"""
if input_file is None: # no input file specified
file_list = h.traverse_output(
name) # run through all the files in 'name' directory
else:
try:
with open(input_file, 'r') as fp:
file_list = [line.rstrip() for line in fp]
except:
logger.warning(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list = h.traverse_output(name)
sources_list = [
] # holds the list of sources acquired from 'name' scraper directory
# loop through filepath in file list
for file_path in file_list:
# read the json data in each filepath
data = h.read_file(file_path)
if not data: # if data is None
continue
# retrieve source from dataset
source = extract_source_from(dataset=data, use_key='collection')
if not source: # source could not be retrieved
continue
# add source to list
sources_list.append(source)
# get a list of non-duplicate Sources
sources_list = get_distinct_sources_from(sources_list,
min_occurence_counter=2)
# get the path were the gotten Sources will be saved to on local disk
file_output_path = f'{CURRENT_TRANSFORMER_OUTPUT_DIR}/{(name or "all")}.sources.json'
# write to file the Sources gotten from 'name' scraped output
h.write_file(file_output_path, sources_list)
# write file the Sources gotten from 'name' scraped out to S3 bucket
h.upload_to_s3_if_configured(file_output_path,
f'{(name or "all")}.sources.json')
def transform(name, input_file=None):
if input_file is None:
file_list = traverse_output(name)
else:
try:
with open(input_file, 'r') as fp:
file_list = [line.rstrip() for line in fp]
except:
logger.warn(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list = traverse_output(name)
logger.debug(f'{len(file_list)} files to transform.')
catalog = Catalog()
catalog.catalog_id = "datopian_data_json_" + name
datasets_number = 0
resources_number = 0
for file_path in file_list:
data = read_file(file_path)
if not data:
continue
dataset = _transform_scraped_dataset(data, name)
catalog.datasets.append(dataset)
datasets_number += 1
resources_number += len(dataset.distribution)
logger.debug('{} datasets transformed.'.format(datasets_number))
logger.debug('{} resources transformed.'.format(resources_number))
output_path = h.get_output_path('datajson')
file_path = os.path.join(output_path, f'{name}.data.json')
with open(file_path, 'w') as output:
output.write(catalog.dump())
logger.debug(f'Output file: {file_path}')
h.upload_to_s3_if_configured(file_path, f'{name}.data.json')
def list_datasets_per_scraper(self, ordered=True):
"""Generate page count per domain
PARAMETERS
- ordered: whether the resulting DataFrame or
Excel sheet result be sorted/ordered. If True, order by 'page count'
"""
filenames = []
try:
with open(self.deduplicated_list_path, 'r') as fp:
filenames = fp.readlines()
except:
logger.warning(
'Warning! Cannot read deduplication results. Please run deduplicate transformer first'
)
filenames = traverse_output()
scraper_counts = {}
for filename in filenames:
scraper_name = str(filename).rstrip().split('/')[-2]
scraper_counts[scraper_name] = (
scraper_counts.get(scraper_name, 0) + 1)
df = pd.DataFrame(columns=['scraper', 'dataset count'])
df['scraper'] = list(scraper_counts.keys())
df['dataset count'] = list(scraper_counts.values())
if ordered:
df.sort_values(by='dataset count',
axis='index',
ascending=False,
inplace=True,
ignore_index=True)
self._add_to_spreadsheet(sheet_name='DATASET COUNT PER SCRAPER',
result=df)
return df
def transform(name=None,
input_file=None,
use_raw_datasets=False) -> pd.DataFrame:
""" function transforms the datajson/datasets into
a dataframe/csv containig data to be used for RAG analyses on
the efficacy of the scraping toolkit to get viable/usable structured data from
the unstructured data source.
The function by default operates on/utilises datajson i.e.
the json that is ready to be ingested by the ckan harvester;
However, setting 'use_raw_datasets' to True means the function will
operate on the raw, parsed data which was scraped from the data source.
PARAMETERS
- name: if provided must correspond to the name of a scraper.
if 'use_raw_datasets' is False, file with the format '<name>.data.json'
will be located in the datajson subdirectory of 'ED_OUTPUT_PATH/transformers'
and read.
if 'use_raw_datasets' is True, dataset files contained in the 'name'
scrapers subdirectory of the 'ED_OUTPUT_PATH/scrapers' will be read
input_file: if provided mut be a file with list of datajson or dataset files
to read.
If no parameters are provided, which is the default behaviour;
then all datajson files contained in datajson subdirectory of
'ED_OUTPUT_PATH/transformers' will be read.
function returns the DataFrame containing the transformed datajson/dataset files
"""
file_list = [] # holds the list of files which contain datajson/dataset
datasets_list = [] # holds the data jsons gotten from files
if use_raw_datasets == True: # work on raw datasets
if not input_file: # no input file provided
# loop over directory structure
if name:
# loop over <name> scraper output e.g nces
file_list = h.traverse_output(name)
# datasets = list of all <name> files
else:
# loop over everything
file_list = h.traverse_output(None)
# datasets = list of all JSON files
else: # input file provided
# read input_file, which is a list of files
with open(input_file, 'r') as fp:
try:
file_list = [line.rstrip() for line in fp]
except Exception:
logger.warning(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list = h.traverse_output(None)
else: # work with processed/transformed datajson
if not input_file: # no input file provided
if name: # name of processed datajson is provided so get the file path
file_list.append(
Path(h.get_output_path('datajson'), f'{name}.data.json'))
else: # name of processed datajson not provided
file_list.extend(
Path(h.get_output_path('datajson')).glob('*.json'))
else: # input file provided
# read input_file, which is a list of files
with open(input_file, 'r') as fp:
try:
file_list = [line.rstrip() for line in fp]
except Exception:
logger.warning(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list.extend(
Path(h.get_output_path('datajson')).glob('*.json'))
if use_raw_datasets == True: # work on raw datasets
# read the contents in file_list
for file_path in file_list:
# read json from the file using helper
data = h.read_file(file_path)
# compute the weight score of the dataset
compute_score(data, append_score=True, use_raw_datasets=True)
datasets_list.append(data)
else: # work with processed json data
# read the contents in the file_list
for file_path in file_list:
# read json from file using helper function
data = h.read_file(file_path)
for dataset_dict in data.get(
'dataset',
[]): # loop through the datasets contained in data
# compute the weighted score of the dataset
compute_score(dataset_dict,
append_score=True,
use_raw_datasets=False)
datasets_list.append(dataset_dict)
if use_raw_datasets == True: # work on raw datasets
# map the datasets to pandas format
dataset_rows_list = map(lambda dataset: [dataset.get('publisher'),\
dataset.get('source_url'), \
dataset.get('_weighted_score'), \
dataset.get('_weighted_score_ratio')],
datasets_list)
else: # work on processed datajson
# map the dataset to pandas format
dataset_rows_list = map(lambda dataset: [dataset.get('publisher')['name'],\
dataset.get('scraped_from'), \
dataset.get('_weighted_score'), \
dataset.get('_weighted_score_ratio')],
datasets_list)
# create the pandas df
weighted_datasets_scores_df = pd.DataFrame(dataset_rows_list,
columns=[
'publisher', 'source url',
'weighted score',
'weighted score ratio'
])
# create a df that incorporates domain info
weighted_datasets_scores_df2 = pd.DataFrame(columns=['domain'])
weighted_datasets_scores_df2['domain'] = weighted_datasets_scores_df.\
apply(lambda row: urllib.parse.\
urlparse(row['source url']).hostname.\
replace('www2.', 'www.').replace('www.', ''), axis=1)
weighted_datasets_scores_df2['publisher'] = weighted_datasets_scores_df[
'publisher']
weighted_datasets_scores_df2['source url'] = weighted_datasets_scores_df[
'source url']
weighted_datasets_scores_df2[
'weighted score'] = weighted_datasets_scores_df['weighted score']
weighted_datasets_scores_df2[
'weighted score ratio'] = weighted_datasets_scores_df[
'weighted score ratio']
# create the output csv file name
output_dated_dir = os.path.join(
OUTPUT_DIR, f'{dt.now().year}-{dt.now().month}-{dt.now().day}')
Path(output_dated_dir).mkdir(parents=True, exist_ok=True)
if use_raw_datasets == True: # use raw datasets
output_filename = "datasets_weighted_scores_{}_raw.csv".format(
name or "all")
else: # use processed datajson
output_filename = "datasets_weighted_scores_{}.csv".format(name
or "all")
# create the fullpath weer file will be written
fullpath = os.path.join(OUTPUT_DIR, output_filename)
# write the dataframe to csv
weighted_datasets_scores_df2.to_csv(fullpath, index=False)
weighted_datasets_scores_df2.to_csv(os.path.join(output_dated_dir,
output_filename),
index=False)
# write the csv to S3 bucket
h.upload_to_s3_if_configured(fullpath, f'{output_filename}')
return weighted_datasets_scores_df2
def transform(name=None, input_file=None):
if input_file is None:
file_list = h.traverse_output(name)
else:
try:
with open(input_file, 'r') as fp:
file_list = [line.rstrip() for line in fp]
except:
logger.warning(f'Cannot read from list of output files at {input_file}, falling back to all collected data!')
file_list = h.traverse_output(name)
# loop through filepath in file list
for file_path in file_list:
# read the json data in each filepath
data = h.read_file(file_path)
if not data: # if data is None
continue
# skip the dataset that has only txt resources
if _dataset_only_has_txt_resources(data):
clean_data = {}
clean_data['_remove_dataset'] = True # mark dataset for removal
data['_clean_data'] = clean_data # update dataset
# Remove datasets with no resources or no relevant resources
if not len(_filter_resources_list(data['resources'])) or not len(data['resources']):
clean_data = {}
clean_data['_remove_dataset'] = True # mark dataset for removal
data['_clean_data'] = clean_data # update dataset
# Special hacks for ed.gov data
if name == 'edgov':
clean_data = {}
clean_data['_remove_dataset'] = False # unmark dataset for removal
# # Get the publisher name
# try:
# publisher_name = data['publisher'].get('name')
# except:
# publisher_name = data['publisher']
# Check for "bad" URLs and remove them
bad_subdomains = ['dashboard', 'rems']
if any([f'{bs}.ed.gov' in data['source_url'] for bs in bad_subdomains]):
clean_data['_remove_dataset'] = True # mark dataset for removal
data['_clean_data'] = clean_data # update dataset
# OESE hack. Remove datasets outside oese.ed.gov domain
publisher = data.get('publisher')
publisher_name = ""
if type(publisher) == dict:
publisher_name = publisher.get('name')
elif type(publisher) == str:
publisher_name = publisher
if publisher_name in ['oese',
'Office of Elementary and Secondary Education',
'Office of Elementary and Secondary Education (OESE)']:
if _dataset_outside_oese_domain(data):
clean_data = {}
clean_data['_remove_dataset'] = True # mark dataset for removal
data['_clean_data'] = clean_data # update dataset
# Remove duplicate identifiers generated by duplicate URLs in IES/NCES
if publisher_name in ['ies',
'Institute of Education Sciences (IES)',
'National Center for Education Statistics (NCES)',
'nces']:
if data.get('source_url').endswith('current=yes'):
clean_data = data['_clean_data']
clean_data['_remove_dataset'] = True # mark dataset for removal
data['_clean_data'] = clean_data # update dataset
# Filter resources
data = _filter_dataset_resources(data)
# mark as private datasets that have certain keywords in their data
data = _mark_private(data, search_words=['conference', 'awards',
'user guide', 'applications'])
# mark of removal datasets that have certain keywords
data = _remove_dataset(data, search_words=['photo', 'foto', 'photos', 'fotos'])
# REMOVE UNWANTED STRING FROM THE VALUE OF A DATASET'S KEY
# 1. remove 'table [0-9].' from beginning of dataset title
data = _strip_unwanted_string(data, r'^table [0-9a-z]+(-?[a-z])?\.',
dict_key='title')
# set the 'level of data' for the dataset
data = _set_dataset_level_of_data(data)
# assign the dataset to groups
# according to https://www2.ed.gov/rschstat/catalog/index.html
data = _set_dataset_groups(data)
# remove the old format for collections / sourcs
data = _remove_old_sources_collections(data)
# write modified dataset back to file
h.write_file(file_path, data)
def transform(name, input_file=None):
if input_file is None:
file_list = traverse_output(name)
else:
try:
with open(input_file, 'r') as fp:
file_list = [line.rstrip() for line in fp]
except:
logger.warning(
f'Cannot read from list of output files at {input_file}, falling back to all collected data!'
)
file_list = traverse_output(name)
logger.debug(f'{len(file_list)} files to transform.')
catalog = Catalog()
catalog.catalog_id = "datopian_data_json_" + (name or 'all')
# keep track/stata for item transformed
datasets_number = 0
resources_number = 0
sources_number = 0
collections_number = 0
# loop through the list of filepaths to be transformed
for file_path in file_list:
data = read_file(file_path)
if not data:
continue
dataset = _transform_scraped_dataset(data, name)
if not dataset: # no dataset was returned (i.e. dataset probably marked for removal)
continue
catalog.datasets.append(dataset)
datasets_number += 1
resources_number += len(dataset.distribution)
# TODO WORK FROM BELOW HERE
# get the list of Sources for this catalog
catalog_sources = list()
try:
# read the list of preprocessed (but still 'raw') Sources from file
catalog_sources = read_file(
f"{h.get_output_path('sources')}/{(name or 'all')}.sources.json")
# transform the list of preprocessed Sources to a list of Source objects acceptable for the catalog object
catalog_sources = _transform_preprocessed_sources(catalog_sources)
except:
logger.warning(
f'"sources transformer" output file ({(name or "all")}.sources.json) not found. This datajson output will have no "source" field'
)
# add the list of Source objects to the catalog
catalog.sources = catalog_sources or []
# update the number fo transformed Sources
sources_number = len(catalog_sources or [])
# get the list of Collections for this catalog
catalog_collections = list()
try:
# read the list of preprocessed (but still 'raw') Collections from file
catalog_collections = read_file(
f"{h.get_output_path('collections')}/{(name or 'all')}.collections.json"
)
# transform the list of preprocessed Collections to a list of Collection objects acceptable for the catalog object
catalog_collections = _transform_preprocessed_collections(
catalog_collections)
except:
logger.warning(
f'"sources transformer" output file ({(name or "all")}.collections.json) not found. This datajson output will have no "collection" field'
)
# add the list of Collection objects to the catalog
catalog.collections = catalog_collections or []
# update the number fo transformed Collections
collections_number = len(catalog_collections or [])
# validate the catalog object
if not catalog.validate_catalog(pls_fix=True):
logger.error(f"catalog validation Failed! Ending transform process")
return
logger.debug('{} Sources transformed.'.format(sources_number))
logger.debug('{} Collections transformed.'.format(collections_number))
logger.debug('{} datasets transformed.'.format(datasets_number))
logger.debug('{} resources transformed.'.format(resources_number))
output_path = h.get_output_path('datajson')
file_path = os.path.join(output_path, f'{(name or "all")}.data.json')
with open(file_path, 'w') as output:
output.write(catalog.dump())
logger.debug(f'Output file: {file_path}')
h.upload_to_s3_if_configured(file_path, f'{(name or "all")}.data.json')