Module tagmaps.classes.prepare_data
Module for preparing base data and calculating overall statistics.
Returns
PreparedStats- Statistics prepared for Tag Maps clustering
self.cleaned_post_dict- Cleaned list of posts
Expand source code
# -*- coding: utf-8 -*-
"""Module for preparing base data and calculating
overall statistics.
Returns:
PreparedStats: Statistics prepared for Tag Maps clustering
self.cleaned_post_dict: Cleaned list of posts
"""
from __future__ import absolute_import
import collections
import csv
import logging
import math
from collections import defaultdict
from dataclasses import dataclass
from pathlib import Path
from typing import Counter, DefaultDict, Dict, List, Optional, Set, Union
from _csv import QUOTE_MINIMAL
from tagmaps.classes.shared_structure import (EMOJI, LOCATIONS, POST_FIELDS, TAGS, TOPICS,
AnalysisBounds, CleanedPost, ClusterTypes,
ItemCounter, PostStructure)
from tagmaps.classes.utils import Utils
@dataclass
class ItemStats:
top_items: List[ItemCounter]
total_unique: int
total_without_longtail: Optional[int] = None
@dataclass
class PreparedStats:
top_items_list: List[ItemCounter]
total_unique_items: int
total_item_count: int
max_items: int
class PrepareData():
"""Main Class for building summary statistics.
- will process individual cleaned post data into dict/set structures
- will filter data, cleaned output can be stored
- will generate statistics
"""
def __init__(
self, cluster_types: List[str], max_items: int,
output_folder: Path, remove_long_tail: bool,
limit_bottom_user_count: int, topic_modeling: bool):
"""Initializes Prepare Data structure"""
# global settings
self.cluster_types = cluster_types
# make sure statistics for locations are always calculated
# because these needed for the cluster process
if LOCATIONS not in self.cluster_types:
self.cluster_types.append(LOCATIONS)
self.max_items = max_items
self.output_folder = output_folder
self.remove_long_tail = remove_long_tail
self.limit_bottom_user_count = limit_bottom_user_count
self.topic_modeling = topic_modeling
# global vars
self.count_glob = 0
self.bounds = AnalysisBounds()
self.log = logging.getLogger("tagmaps")
# The following dict stores, per cls_type,
# the total number of times items appeared
# these are used to measure total counts
self.total_item_counter: Dict[str, Counter[str]] = \
dict.fromkeys(self.cluster_types, collections.Counter())
# for cls_type in self.cluster_types:
# self.total_item_counter[cls_type] =
# cleaned stats dict for each ClusterType
self.cleaned_stats: Dict[str, PreparedStats] = {}
# Hashsets:
self.items_per_userloc: Dict[
str, DefaultDict[str, Set[str]]] = {}
for cls_type in [EMOJI, TAGS]:
# items per user_location [EMOJI, TAGS, TOPICS]
self.items_per_userloc[cls_type] = defaultdict(set)
# and LOCATIONS per user
self.locations_per_user = defaultdict(set)
# dict to store names for loc ids
self.locid_locname_dict: Dict[str, str] = dict() # nopep8
if self.topic_modeling:
self.user_topiclist_dict = defaultdict(set)
self.user_post_ids_dict = defaultdict(set)
self.userpost_first_thumb_dict = defaultdict(str)
# list of distinct terms per user-location
self.userlocation_terms_dict = defaultdict(set)
# first item for each UPL, required
# for some attributes to generate CleanedPost
self.userlocations_firstpost_dict: \
Dict[str, Union[PostStructure, CleanedPost]] = dict()
# The following dicts store, per cls_type,
# distinct items on a per user basis, e.g.
# self.useritem_counts_global[TAGS][USER] = {term1, term2, term3}
self.useritem_counts_global: Dict[
str, DefaultDict[str, Set[str]]] = dict()
for cls_type in self.cluster_types:
self.useritem_counts_global[cls_type] = defaultdict(set)
def add_record(
self, lbsn_post: Union[PostStructure, CleanedPost]):
"""Method will merge all tags/emoji/terms
of a single user for each location (Metric 'UPL') to
produce a cleaned version of input data
- further information is derived from the first
post for each user-location
- the result is a cleaned output containing
reduced information that is necessary for tag maps
- get cleaned output with get_prepared_data()
"""
self.count_glob += 1
self._update_toplists(lbsn_post)
# create userid_loc_id, this is used as the base
# for clustering data (metric UPL)
post_locid_userid: str = f'{lbsn_post.loc_id}::{lbsn_post.user_guid}'
if lbsn_post.loc_id and (
lbsn_post.loc_name and lbsn_post.loc_id
not in self.locid_locname_dict):
# add locname to dict
self.locid_locname_dict[
lbsn_post.loc_id] = lbsn_post.loc_name
if (lbsn_post.user_guid not in
self.locations_per_user or
lbsn_post.loc_id not in
self.locations_per_user[lbsn_post.user_guid]):
# Bit wise or and assignment in one step.
# -> assign locID to UserDict list
# if not already contained
self.locations_per_user[lbsn_post.user_guid] |= \
{lbsn_post.loc_id}
# self.stats.count_loc += 1
self.userlocations_firstpost_dict[
post_locid_userid] = lbsn_post
# union tags/emoji per userid/unique location
if TAGS in self.cluster_types:
self.items_per_userloc[TAGS][post_locid_userid] \
|= lbsn_post.hashtags
if EMOJI in self.cluster_types:
self.items_per_userloc[EMOJI][post_locid_userid] \
|= lbsn_post.emoji
if isinstance(lbsn_post, PostStructure):
# get cleaned wordlist
cleaned_terms_body = set(self._get_cleaned_wordlist(
lbsn_post.post_body))
cleaned_terms_title = set(self._get_cleaned_wordlist(
lbsn_post.post_title))
cleaned_terms = cleaned_terms_body.union(cleaned_terms_title)
else:
# words already cleaned
cleaned_terms = lbsn_post.post_body
# union words per userid/unique location
self.userlocation_terms_dict[
post_locid_userid] |= cleaned_terms
def get_cleaned_post_dict(
self, input_path=None) -> Optional[Dict[str, CleanedPost]]:
"""Output wrapper
- calls loop user locations method
- optionally initializes output to file
"""
if input_path is None:
# load from ingested data
cleaned_post_dict = self._compile_cleaned_data()
else:
# load from file store
cleaned_post_dict = self._load_cleaned_data(input_path)
return cleaned_post_dict
def _load_cleaned_data(self, input_path):
"""Get cleaned Post Dict from intermediate
data stored in file"""
input_file = Path.cwd() / input_path
if not input_file.exists():
raise ValueError(f"File does not exist: {input_file}")
cleaned_post_dict = self._read_cleaned_data(input_file)
return cleaned_post_dict
def write_cleaned_data(
self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] = None,
panon: bool = None):
"""Write cleaned data to intermediate file"""
self.log.info(
f'Writing cleaned intermediate '
f'data to file (Output_cleaned.csv)..')
if cleaned_post_dict is None:
cleaned_post_dict = self.get_cleaned_post_dict()
if panon is None:
panon = True
if panon:
panon_set = self._get_panon_sets()
else:
panon_set = None
with open(self.output_folder / 'Output_cleaned.csv', 'w',
encoding='utf8') as csvfile:
# get headerline from class structure
headerline = ','.join(POST_FIELDS)
csvfile.write(f'{headerline}\n')
# values will be written with CSV writer module
datawriter = csv.writer(
csvfile, delimiter=',', lineterminator='\n',
quotechar='"', quoting=csv.QUOTE_NONNUMERIC)
for cleaned_post in cleaned_post_dict.values():
self._write_location_tocsv(
datawriter, cleaned_post,
panon_set)
self.log.info(' done.')
def _get_panon_sets(self):
"""Prepare panon by generating dict of sets with popular terms
"""
panon_set = dict()
for cls_type in self.cluster_types:
max_items = self.cleaned_stats[cls_type].max_items
panon_set[cls_type] = {
item.name for item in self.cleaned_stats[cls_type].top_items_list[:max_items]}
return panon_set
def get_panonymized_posts(
self,
cleaned_post_dict: Optional[Dict[str, CleanedPost]]
) -> Dict[str, CleanedPost]:
"""Returns a new cleaned post dict with reduced information detail
based on global information patterns
This is not a true anonymization. Returned items have specifically
the highly identifyable information removed (specific tags/terms used by few
users), which make it harder to identify original users from resulting data.
"""
panon_cleaned_post_dict = defaultdict(CleanedPost)
panon_set = self._get_panon_sets()
for upl, cleaned_post in cleaned_post_dict.items():
upl_panon = PrepareData._panonymize_cleaned_post(
cleaned_post, panon_set)
panon_cleaned_post_dict[upl] = upl_panon
return panon_cleaned_post_dict
def get_item_stats(self) -> Dict[str, PreparedStats]:
"""After data is loaded, this collects data and stats
for distribution of tags, emoji and locations
- prepare data for tag maps clustering
- store to self.data_prepared
"""
if not self.cleaned_stats:
self._init_item_stats()
return self.cleaned_stats
def _init_item_stats(self):
"""Init stats for all cls_types"""
for cls_type in ClusterTypes:
self.cleaned_stats[cls_type] = self._prepare_item_stats(
cls_type)
def _prepare_item_stats(self, cls_type) -> PreparedStats:
"""Calculate overall tag and emoji statistics
- write results (optionally) to file
- stats for user excluded types are initialized empty
"""
if not cls_type in self.cluster_types:
# return empty stats if excluded by user
return PreparedStats([], 0, 0, 0)
# top lists and unique
item_stats = self._get_top_list(cls_type)
top_items_list = item_stats.top_items
total_unique_items = item_stats.total_unique
itemcount_without_longtail = item_stats.total_without_longtail
# top counts
total_item_count = PrepareData._get_total_count(
top_items_list, self.total_item_counter[cls_type])
# assign stats to structure
# update max_item from optionally long tail removal
if itemcount_without_longtail and cls_type in [EMOJI, TAGS]:
max_items = itemcount_without_longtail
else:
max_items = self.max_items
item_stats = PreparedStats(
top_items_list, total_unique_items, total_item_count,
max_items)
return item_stats
def _update_toplists(self, lbsn_post):
"""Calculate toplists for emoji, tags and locations
- adds tag/emojicount of this post to overall
tag/emojicount for this user,
- initialize counter for user if not already done
"""
for cls_type in [EMOJI, TAGS, TOPICS]:
if cls_type not in self.cluster_types:
continue
if cls_type in [TAGS, TOPICS]:
# to do: TOPIC implementation
item_list = lbsn_post.hashtags
else:
item_list = lbsn_post.emoji
if not item_list:
continue
self.useritem_counts_global[cls_type][
lbsn_post.user_guid].update(
item_list)
self.total_item_counter[cls_type].update(item_list)
# locations
if lbsn_post.loc_id:
# update single item
self.useritem_counts_global[LOCATIONS][lbsn_post.user_guid].add(
lbsn_post.loc_id)
self.total_item_counter[LOCATIONS][lbsn_post.loc_id] += 1
@staticmethod
def _write_toplist(
top_list, list_type, max_items, output_folder,
locid_name_dict=None):
"""Write toplists to fileget_locname
e.g.:
tag, usercount
toptag1, 1231
toptag2, 560
...
"""
if not top_list:
return
# only ever write top 1000 to file
max_to_write = min(1000, max_items)
top_list = top_list[:max_to_write]
# reformat list as lines to be written
if list_type == LOCATIONS and locid_name_dict:
# construct line string and
# get name for locid, if possible
top_list_rf = []
for item in top_list:
loc_name = Utils.get_locname(item.name, locid_name_dict)
coords = item.name.split(":")
ucount = item.ucount
line = (f'{loc_name.replace(",","-")},{coords[0]},{coords[1]},'
f'{ucount}')
top_list_rf.append(line)
top_list = top_list_rf
else:
# construct line string
top_list = [f'{v.name},{v.ucount}' for v in top_list]
# overwrite, if exists:
with open(output_folder / f'Output_top{list_type}.txt',
'w', encoding='utf8') as out_file:
if list_type == LOCATIONS:
out_file.write(f'{list_type},lat,lng,usercount\n')
else:
out_file.write(f'{list_type},usercount\n')
for line in top_list:
out_file.write(f'{line}\n')
def write_toplists(self):
"""Writes toplists (tags, emoji, locations) to file"""
for cls_type in self.cluster_types:
top_list = self.cleaned_stats[cls_type].top_items_list
max_items = self.cleaned_stats[cls_type].max_items
PrepareData._write_toplist(
top_list, cls_type, max_items,
self.output_folder, self.locid_locname_dict)
def _get_top_list(
self, cls_type: str = TAGS) -> ItemStats:
"""Get Top Tags on a per user basis, i.e.
- the global number of distinct users who used each distinct tag
- this ignores duplicate use of
- calculation is based on dict userdict_itemcounters_global,
with counters of tags for each user
Returns:
- list of top tags up to tmax [1000]
- count of total unique tags
"""
overall_usercount_per_item = collections.Counter()
for item_hash in self.useritem_counts_global[cls_type].values():
# taghash contains unique values (= strings) for each user,
# thus summing up these taghashes counts each user
# only once per tag (or emoji)
overall_usercount_per_item.update(item_hash)
total_unique = len(overall_usercount_per_item)
# get all items for "locations"
# but clip list for tags and emoji
if cls_type in (TAGS, EMOJI):
max_items = self.max_items
else:
max_items = None
top_items_list = overall_usercount_per_item.most_common(max_items)
# convert list of item counts into list of namedtuple
top_items_list = [
ItemCounter(*ic_tuple) for ic_tuple in top_items_list]
total_without_longtail = 0
if self.remove_long_tail is True:
total_without_longtail = self._remove_long_tail(
top_items_list, cls_type)
return ItemStats(top_items_list, total_unique, total_without_longtail)
@staticmethod
def _get_total_count(top_list, top_counter):
"""Calculate Total Tags for selected
Arguments:
top_list (Long Tail Stat)
top_counter (Reference to counter object)
"""
total_count = 0
for item in top_list:
count = top_counter.get(item.name)
if count:
total_count += count
return total_count
def _remove_long_tail(self,
top_list: List[ItemCounter],
listtype: str
) -> Optional[int]:
"""Removes all items from list that are used by less
than x number of users,
where x is given as input arg limit_bottom_user_count
Note: since list is a mutable object, method
will modify top_tags_list
"""
if listtype == LOCATIONS:
# keep all locations
return len(top_list)
if listtype == EMOJI:
# emoji use a smaller area than tags on the map
# therefore we can keep more emoji
# (e.g..: use 2 instead of 5)
bottomuser_count = math.trunc(
self.limit_bottom_user_count/2)
else:
bottomuser_count = self.limit_bottom_user_count
index_min = next(
(i for i, (t1, t2) in enumerate(
top_list) if t2 < bottomuser_count
), None)
if not index_min:
return
len_before = len(top_list)
# delete based on slicing
del top_list[index_min:]
len_after = len(top_list)
if len_before == len_after:
# if no change, return
return len_after
self.log.info(
f'Long tail removal: Filtered {len_before - len_after} '
f'{listtype} that were used by less than '
f'{bottomuser_count} users.')
return len_after
def _compile_cleaned_data(self) -> Optional[Dict[str, CleanedPost]]:
"""Will produce final cleaned list
of items to be processed by clustering.
- optionally writes entries to file, if handler exists
"""
cleaned_post_dict: Dict[str, CleanedPost] = dict()
for user_guid, locationhash in self.locations_per_user.items():
# loop all distinct user locations
for location in locationhash:
locid_userid = f'{location}::{user_guid}'
post_latlng = location.split(':')
first_post = self.userlocations_firstpost_dict.get(
locid_userid, None)
if first_post is None:
return
# create tuple with cleaned photo data
cleaned_post = self._compile_cleaned_post(
first_post, locid_userid, post_latlng, user_guid)
if cleaned_post is None:
continue
if self.topic_modeling:
self._update_topic_models(
cleaned_post, user_guid)
cleaned_post_dict[cleaned_post.guid] = cleaned_post
# update boundary
self.bounds.upd_latlng_bounds(
cleaned_post.lat, cleaned_post.lng)
return cleaned_post_dict
def _read_cleaned_data(self, cdata: Path):
"""Create cleaned post dict from intermediate data file store"""
cleaned_post_dict: Dict[str, CleanedPost] = dict()
with open(cdata, 'r', newline='', encoding='utf8') as f_handle:
cpost_reader = csv.DictReader(
f_handle,
delimiter=',',
quotechar='"',
quoting=QUOTE_MINIMAL)
for cpost in cpost_reader:
# row_num += 1
cleaned_post = PrepareData._parse_cleaned_post(cpost)
cleaned_post_dict[cleaned_post.guid] = cleaned_post
# update statistics from cleaned post
self.add_record(cleaned_post)
# update boundary
self.bounds.upd_latlng_bounds(
cleaned_post.lat, cleaned_post.lng)
return cleaned_post_dict
def write_topic_models(self):
"""Initialize two lists for topic modeling output
- hashed (anonymized) output (*add salt)
- original output
"""
headerline = "topics,post_ids,user_ids\n"
with open(
self.output_folder / 'Output_usertopics_anonymized.csv',
'w', encoding='utf8') as csvfile_anon, open(
self.output_folder / 'Output_usertopics.csv',
'w', encoding='utf8') as csvfile:
dw_list = list()
for cfile in (csvfile, csvfile_anon):
cfile.write(headerline)
datawriter = csv.writer(cfile, delimiter=',',
lineterminator='\n', quotechar='"',
quoting=csv.QUOTE_NONNUMERIC)
dw_list.append(datawriter)
self._write_topic_rows(dw_list)
def _write_topic_rows(self, dw_list):
"""Write Topic models to file"""
datawriter = dw_list[0]
dw_anon = dw_list[1]
def _join_encode(keys):
joined_keys = ",".join(keys)
joined_encoded_keys = ",".join(
[Utils.encode_string(post_id) for post_id in keys])
return joined_keys, joined_encoded_keys
for user_key, topics in self.user_topiclist_dict.items():
joined_topics = " ".join(topics)
post_keys = self.user_post_ids_dict.get(user_key, None)
joined_keys, joined_encoded_keys = _join_encode(post_keys)
dw_anon.writerow([joined_topics,
"{" + joined_encoded_keys + "}",
Utils.encode_string(user_key)])
datawriter.writerow([joined_topics,
"{" + joined_keys + "}",
str(user_key)])
def _update_topic_models(self,
cleaned_post_location,
user_key):
"""If Topic Modeling enabled, update
required dictionaries with merged words from
title, tags and post_body
"""
if cleaned_post_location.hashtags:
self.user_topiclist_dict[user_key] |= \
cleaned_post_location.hashtags
# also use descriptions for Topic Modeling
self.user_topiclist_dict[user_key] |= \
cleaned_post_location.post_body
# Bit wise or and assignment in one step.
# -> assign PhotoGuid to UserDict list
# if not already contained
self.user_post_ids_dict[user_key] |= {
cleaned_post_location.guid}
# UserPhotoFirstThumb_dict[user_key] = photo[5]
@staticmethod
def _parse_cleaned_post(cpost: Dict[str, str]) -> CleanedPost:
"""Process single cleaned post from (file) dict stream"""
# process column with concatenate items (";item1;item2")
split_string_dict = dict()
for split_col in ["post_body", "hashtags", "emoji"]:
item_str = cpost.get(split_col)
if item_str:
items = set(item_str.split(";"))
split_string_dict[split_col] = items
locid = cpost.get("loc_id")
if not locid:
raise ValueError("Field loc_id empty.")
cleaned_post = CleanedPost(
origin_id=int(cpost["origin_id"]),
lat=float(cpost["lat"]),
lng=float(cpost["lng"]),
guid=cpost["guid"],
user_guid=cpost["user_guid"],
post_body=split_string_dict.get("post_body", set()),
post_create_date=cpost.get("post_create_date"),
post_publish_date=cpost.get("post_publish_date"),
post_views_count=int(cpost.get("post_views_count", 0)),
post_like_count=int(cpost.get("post_like_count", 0)),
emoji=split_string_dict.get("emoji", set()),
hashtags=split_string_dict.get("hashtags", set()),
loc_id=locid,
loc_name=cpost.get("loc_name")
)
return cleaned_post
def _compile_cleaned_post(self, first_post, locid_userid,
post_latlng, user_key) -> Optional[CleanedPost]:
"""Merge cleaned post from all posts of a certain user
at a specific location. This is producing the final CleanedPost.
- some information is not needed, those post attributes
are simply skipped (e.g. location name)
- some information must not be merged, this can be directly copied
from the first post at a location/user (e.g. origin_id - will always be
the same for a particular user, post_create_date, post_publish_date)
- some information (e.g. hashtags) need merge with removing dupliates:
use prepared dictionaries
- some important information is type-checked (longitude, latitude)
Keyword arguments:
first_post -- first post of a user_guid at a location
locid_userid -- user_guid and loc_id in merged format
(f'{location}::{user_key}')
post_latlng -- tuple with lat/lng coordinates
user_key -- user_guid
Note:
("",) means: substitute empty tuple as default
"""
merged_wordlist = PrepareData._get_merged(
self.userlocation_terms_dict, locid_userid)
merged_emojilist = PrepareData._get_merged(
self.items_per_userloc[EMOJI], locid_userid)
merged_taglist = PrepareData._get_merged(
self.items_per_userloc[TAGS], locid_userid)
try:
lat = float(post_latlng[0])
lng = float(post_latlng[1])
except ValueError:
# skip posts with no
# or malformed lat/lng
return
cleaned_post = CleanedPost(
origin_id=first_post.origin_id,
lat=lat,
lng=lng,
guid=first_post.guid,
user_guid=user_key,
post_body=merged_wordlist,
post_create_date=first_post.post_create_date,
post_publish_date=first_post.post_publish_date,
post_views_count=first_post.post_views_count,
post_like_count=first_post.post_like_count,
emoji=merged_emojilist,
hashtags=merged_taglist,
loc_id=first_post.loc_id,
loc_name=first_post.loc_name
)
return cleaned_post
@staticmethod
def _get_merged(ref_dict: Dict, locid_userid: str) -> Set[str]:
"""Gets set of words for userlocid from ref dictionary
Note: since using defaultdict,
keys not found will return empty set()
"""
value = ref_dict[locid_userid]
return value
def _write_location_tocsv(self, datawriter,
cleaned_post_location: CleanedPost,
panon_set=None) -> None:
"""Writes a single record of cleaned posts to CSV list
- write intermediate cleaned post data to file for later use
Arguments
datawriter - open file file_handle to
output file
cleaned_post_location - cleaned post of type CleanedPost
(namedtuple)
panonymize - This will limit written item-lists
(emoji, tags, body-content) to
the terms that exist in identified
toplists. The result is a pseudo-
anonymized post that only contains
the less identifiable popular terms
that are used by many users.
"""
if panon_set:
cleaned_post_location = self._panonymize_cleaned_post(
cleaned_post_location, panon_set)
ploc_list = PrepareData._cleaned_ploc_tolist(
cleaned_post_location)
datawriter.writerow(ploc_list)
@staticmethod
def _panonymize_cleaned_post(
upl: CleanedPost,
panon_set: Dict[str, Set[str]]) -> CleanedPost:
"""Returns a new cleaned post with reduced information detail
based on global information patterns"""
panon_post = CleanedPost(
origin_id=upl.origin_id,
lat=upl.lat,
lng=upl.lng,
guid=upl.guid,
user_guid=upl.user_guid,
post_body=PrepareData._filter_private_terms(
upl.post_body, panon_set.get(TAGS)),
post_create_date=PrepareData._agg_date(upl.post_create_date),
post_publish_date=PrepareData._agg_date(upl.post_publish_date),
post_views_count=upl.post_views_count,
post_like_count=upl.post_like_count,
emoji=PrepareData._filter_private_terms(
upl.emoji, panon_set.get(EMOJI)),
hashtags=PrepareData._filter_private_terms(
upl.hashtags, panon_set.get(TAGS)),
loc_id=upl.loc_id,
loc_name=upl.loc_name
)
return panon_post
@staticmethod
def _agg_date(
str_date: Optional[str]) -> str:
"""Remove time info from string, e.g.
2010-05-07 16:00:54
to 2010-05-07
"""
if str_date:
str_date_hr = f'{str_date[:10]}'
return str_date_hr
return ""
@staticmethod
def _filter_private_terms(
str_list: Optional[Set[str]] = None,
top_terms_set: Optional[Set[str]] = None) -> Set[str]:
if str_list and top_terms_set:
filtered_set = {term for term in str_list if term in top_terms_set}
return filtered_set
# if none or empty, return empty set
return set()
@staticmethod
def _cleaned_ploc_tolist(cleaned_post_location: CleanedPost) -> List[str]:
"""Converts a cleaned post structure to list for CSV write"""
attr_list = list()
for attr in cleaned_post_location:
if isinstance(attr, set):
attr_list.append(";".join(attr))
else:
attr_list.append(attr)
return attr_list
def _get_cleaned_wordlist(self, post_body_string):
cleaned_post_body = Utils.remove_special_chars(post_body_string)
cleaned_wordlist = PrepareData._get_wordlist(cleaned_post_body)
return cleaned_wordlist
@staticmethod
def _get_wordlist(cleaned_post_body):
"""split by space-characterm, filter by length"""
wordlist = [word for word in cleaned_post_body.lower().split(
' ') if len(word) > 2]
return wordlist
def global_stats_report(self, cleaned=None):
"""Report global stats after data has been read"""
if cleaned is None:
cleaned = True
self.log.info(
f'Total user count (UC): '
f'{len(self.locations_per_user)}')
upl = sum(len(v) for v in self.locations_per_user.values())
self.log.info(
f'Total user post locations (UPL): '
f'{upl}')
if not cleaned:
return
if not self.cleaned_stats:
self._init_item_stats()
self.log.info(
f'Total (cleaned) post count (PC): '
f'{self.count_glob:02d}')
self.log.info(
f'Total (cleaned) tag count (PTC): '
f'{self.cleaned_stats[TAGS].total_item_count}')
self.log.info(
f'Total (cleaned) emoji count (PEC): '
f'{self.cleaned_stats[EMOJI].total_item_count}')Classes
class ItemStats (top_items: List[ItemCounter], total_unique: int, total_without_longtail: Optional[int] = None)-
ItemStats(top_items: List[tagmaps.classes.shared_structure.ItemCounter], total_unique: int, total_without_longtail: Optional[int] = None)
Expand source code
@dataclass class ItemStats: top_items: List[ItemCounter] total_unique: int total_without_longtail: Optional[int] = NoneClass variables
var top_items : List[ItemCounter]var total_unique : intvar total_without_longtail : Optional[int]
class PrepareData (cluster_types: List[str], max_items: int, output_folder: pathlib.Path, remove_long_tail: bool, limit_bottom_user_count: int, topic_modeling: bool)-
Main Class for building summary statistics.
- will process individual cleaned post data into dict/set structures
- will filter data, cleaned output can be stored
- will generate statistics
Initializes Prepare Data structure
Expand source code
class PrepareData(): """Main Class for building summary statistics. - will process individual cleaned post data into dict/set structures - will filter data, cleaned output can be stored - will generate statistics """ def __init__( self, cluster_types: List[str], max_items: int, output_folder: Path, remove_long_tail: bool, limit_bottom_user_count: int, topic_modeling: bool): """Initializes Prepare Data structure""" # global settings self.cluster_types = cluster_types # make sure statistics for locations are always calculated # because these needed for the cluster process if LOCATIONS not in self.cluster_types: self.cluster_types.append(LOCATIONS) self.max_items = max_items self.output_folder = output_folder self.remove_long_tail = remove_long_tail self.limit_bottom_user_count = limit_bottom_user_count self.topic_modeling = topic_modeling # global vars self.count_glob = 0 self.bounds = AnalysisBounds() self.log = logging.getLogger("tagmaps") # The following dict stores, per cls_type, # the total number of times items appeared # these are used to measure total counts self.total_item_counter: Dict[str, Counter[str]] = \ dict.fromkeys(self.cluster_types, collections.Counter()) # for cls_type in self.cluster_types: # self.total_item_counter[cls_type] = # cleaned stats dict for each ClusterType self.cleaned_stats: Dict[str, PreparedStats] = {} # Hashsets: self.items_per_userloc: Dict[ str, DefaultDict[str, Set[str]]] = {} for cls_type in [EMOJI, TAGS]: # items per user_location [EMOJI, TAGS, TOPICS] self.items_per_userloc[cls_type] = defaultdict(set) # and LOCATIONS per user self.locations_per_user = defaultdict(set) # dict to store names for loc ids self.locid_locname_dict: Dict[str, str] = dict() # nopep8 if self.topic_modeling: self.user_topiclist_dict = defaultdict(set) self.user_post_ids_dict = defaultdict(set) self.userpost_first_thumb_dict = defaultdict(str) # list of distinct terms per user-location self.userlocation_terms_dict = defaultdict(set) # first item for each UPL, required # for some attributes to generate CleanedPost self.userlocations_firstpost_dict: \ Dict[str, Union[PostStructure, CleanedPost]] = dict() # The following dicts store, per cls_type, # distinct items on a per user basis, e.g. # self.useritem_counts_global[TAGS][USER] = {term1, term2, term3} self.useritem_counts_global: Dict[ str, DefaultDict[str, Set[str]]] = dict() for cls_type in self.cluster_types: self.useritem_counts_global[cls_type] = defaultdict(set) def add_record( self, lbsn_post: Union[PostStructure, CleanedPost]): """Method will merge all tags/emoji/terms of a single user for each location (Metric 'UPL') to produce a cleaned version of input data - further information is derived from the first post for each user-location - the result is a cleaned output containing reduced information that is necessary for tag maps - get cleaned output with get_prepared_data() """ self.count_glob += 1 self._update_toplists(lbsn_post) # create userid_loc_id, this is used as the base # for clustering data (metric UPL) post_locid_userid: str = f'{lbsn_post.loc_id}::{lbsn_post.user_guid}' if lbsn_post.loc_id and ( lbsn_post.loc_name and lbsn_post.loc_id not in self.locid_locname_dict): # add locname to dict self.locid_locname_dict[ lbsn_post.loc_id] = lbsn_post.loc_name if (lbsn_post.user_guid not in self.locations_per_user or lbsn_post.loc_id not in self.locations_per_user[lbsn_post.user_guid]): # Bit wise or and assignment in one step. # -> assign locID to UserDict list # if not already contained self.locations_per_user[lbsn_post.user_guid] |= \ {lbsn_post.loc_id} # self.stats.count_loc += 1 self.userlocations_firstpost_dict[ post_locid_userid] = lbsn_post # union tags/emoji per userid/unique location if TAGS in self.cluster_types: self.items_per_userloc[TAGS][post_locid_userid] \ |= lbsn_post.hashtags if EMOJI in self.cluster_types: self.items_per_userloc[EMOJI][post_locid_userid] \ |= lbsn_post.emoji if isinstance(lbsn_post, PostStructure): # get cleaned wordlist cleaned_terms_body = set(self._get_cleaned_wordlist( lbsn_post.post_body)) cleaned_terms_title = set(self._get_cleaned_wordlist( lbsn_post.post_title)) cleaned_terms = cleaned_terms_body.union(cleaned_terms_title) else: # words already cleaned cleaned_terms = lbsn_post.post_body # union words per userid/unique location self.userlocation_terms_dict[ post_locid_userid] |= cleaned_terms def get_cleaned_post_dict( self, input_path=None) -> Optional[Dict[str, CleanedPost]]: """Output wrapper - calls loop user locations method - optionally initializes output to file """ if input_path is None: # load from ingested data cleaned_post_dict = self._compile_cleaned_data() else: # load from file store cleaned_post_dict = self._load_cleaned_data(input_path) return cleaned_post_dict def _load_cleaned_data(self, input_path): """Get cleaned Post Dict from intermediate data stored in file""" input_file = Path.cwd() / input_path if not input_file.exists(): raise ValueError(f"File does not exist: {input_file}") cleaned_post_dict = self._read_cleaned_data(input_file) return cleaned_post_dict def write_cleaned_data( self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] = None, panon: bool = None): """Write cleaned data to intermediate file""" self.log.info( f'Writing cleaned intermediate ' f'data to file (Output_cleaned.csv)..') if cleaned_post_dict is None: cleaned_post_dict = self.get_cleaned_post_dict() if panon is None: panon = True if panon: panon_set = self._get_panon_sets() else: panon_set = None with open(self.output_folder / 'Output_cleaned.csv', 'w', encoding='utf8') as csvfile: # get headerline from class structure headerline = ','.join(POST_FIELDS) csvfile.write(f'{headerline}\n') # values will be written with CSV writer module datawriter = csv.writer( csvfile, delimiter=',', lineterminator='\n', quotechar='"', quoting=csv.QUOTE_NONNUMERIC) for cleaned_post in cleaned_post_dict.values(): self._write_location_tocsv( datawriter, cleaned_post, panon_set) self.log.info(' done.') def _get_panon_sets(self): """Prepare panon by generating dict of sets with popular terms """ panon_set = dict() for cls_type in self.cluster_types: max_items = self.cleaned_stats[cls_type].max_items panon_set[cls_type] = { item.name for item in self.cleaned_stats[cls_type].top_items_list[:max_items]} return panon_set def get_panonymized_posts( self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] ) -> Dict[str, CleanedPost]: """Returns a new cleaned post dict with reduced information detail based on global information patterns This is not a true anonymization. Returned items have specifically the highly identifyable information removed (specific tags/terms used by few users), which make it harder to identify original users from resulting data. """ panon_cleaned_post_dict = defaultdict(CleanedPost) panon_set = self._get_panon_sets() for upl, cleaned_post in cleaned_post_dict.items(): upl_panon = PrepareData._panonymize_cleaned_post( cleaned_post, panon_set) panon_cleaned_post_dict[upl] = upl_panon return panon_cleaned_post_dict def get_item_stats(self) -> Dict[str, PreparedStats]: """After data is loaded, this collects data and stats for distribution of tags, emoji and locations - prepare data for tag maps clustering - store to self.data_prepared """ if not self.cleaned_stats: self._init_item_stats() return self.cleaned_stats def _init_item_stats(self): """Init stats for all cls_types""" for cls_type in ClusterTypes: self.cleaned_stats[cls_type] = self._prepare_item_stats( cls_type) def _prepare_item_stats(self, cls_type) -> PreparedStats: """Calculate overall tag and emoji statistics - write results (optionally) to file - stats for user excluded types are initialized empty """ if not cls_type in self.cluster_types: # return empty stats if excluded by user return PreparedStats([], 0, 0, 0) # top lists and unique item_stats = self._get_top_list(cls_type) top_items_list = item_stats.top_items total_unique_items = item_stats.total_unique itemcount_without_longtail = item_stats.total_without_longtail # top counts total_item_count = PrepareData._get_total_count( top_items_list, self.total_item_counter[cls_type]) # assign stats to structure # update max_item from optionally long tail removal if itemcount_without_longtail and cls_type in [EMOJI, TAGS]: max_items = itemcount_without_longtail else: max_items = self.max_items item_stats = PreparedStats( top_items_list, total_unique_items, total_item_count, max_items) return item_stats def _update_toplists(self, lbsn_post): """Calculate toplists for emoji, tags and locations - adds tag/emojicount of this post to overall tag/emojicount for this user, - initialize counter for user if not already done """ for cls_type in [EMOJI, TAGS, TOPICS]: if cls_type not in self.cluster_types: continue if cls_type in [TAGS, TOPICS]: # to do: TOPIC implementation item_list = lbsn_post.hashtags else: item_list = lbsn_post.emoji if not item_list: continue self.useritem_counts_global[cls_type][ lbsn_post.user_guid].update( item_list) self.total_item_counter[cls_type].update(item_list) # locations if lbsn_post.loc_id: # update single item self.useritem_counts_global[LOCATIONS][lbsn_post.user_guid].add( lbsn_post.loc_id) self.total_item_counter[LOCATIONS][lbsn_post.loc_id] += 1 @staticmethod def _write_toplist( top_list, list_type, max_items, output_folder, locid_name_dict=None): """Write toplists to fileget_locname e.g.: tag, usercount toptag1, 1231 toptag2, 560 ... """ if not top_list: return # only ever write top 1000 to file max_to_write = min(1000, max_items) top_list = top_list[:max_to_write] # reformat list as lines to be written if list_type == LOCATIONS and locid_name_dict: # construct line string and # get name for locid, if possible top_list_rf = [] for item in top_list: loc_name = Utils.get_locname(item.name, locid_name_dict) coords = item.name.split(":") ucount = item.ucount line = (f'{loc_name.replace(",","-")},{coords[0]},{coords[1]},' f'{ucount}') top_list_rf.append(line) top_list = top_list_rf else: # construct line string top_list = [f'{v.name},{v.ucount}' for v in top_list] # overwrite, if exists: with open(output_folder / f'Output_top{list_type}.txt', 'w', encoding='utf8') as out_file: if list_type == LOCATIONS: out_file.write(f'{list_type},lat,lng,usercount\n') else: out_file.write(f'{list_type},usercount\n') for line in top_list: out_file.write(f'{line}\n') def write_toplists(self): """Writes toplists (tags, emoji, locations) to file""" for cls_type in self.cluster_types: top_list = self.cleaned_stats[cls_type].top_items_list max_items = self.cleaned_stats[cls_type].max_items PrepareData._write_toplist( top_list, cls_type, max_items, self.output_folder, self.locid_locname_dict) def _get_top_list( self, cls_type: str = TAGS) -> ItemStats: """Get Top Tags on a per user basis, i.e. - the global number of distinct users who used each distinct tag - this ignores duplicate use of - calculation is based on dict userdict_itemcounters_global, with counters of tags for each user Returns: - list of top tags up to tmax [1000] - count of total unique tags """ overall_usercount_per_item = collections.Counter() for item_hash in self.useritem_counts_global[cls_type].values(): # taghash contains unique values (= strings) for each user, # thus summing up these taghashes counts each user # only once per tag (or emoji) overall_usercount_per_item.update(item_hash) total_unique = len(overall_usercount_per_item) # get all items for "locations" # but clip list for tags and emoji if cls_type in (TAGS, EMOJI): max_items = self.max_items else: max_items = None top_items_list = overall_usercount_per_item.most_common(max_items) # convert list of item counts into list of namedtuple top_items_list = [ ItemCounter(*ic_tuple) for ic_tuple in top_items_list] total_without_longtail = 0 if self.remove_long_tail is True: total_without_longtail = self._remove_long_tail( top_items_list, cls_type) return ItemStats(top_items_list, total_unique, total_without_longtail) @staticmethod def _get_total_count(top_list, top_counter): """Calculate Total Tags for selected Arguments: top_list (Long Tail Stat) top_counter (Reference to counter object) """ total_count = 0 for item in top_list: count = top_counter.get(item.name) if count: total_count += count return total_count def _remove_long_tail(self, top_list: List[ItemCounter], listtype: str ) -> Optional[int]: """Removes all items from list that are used by less than x number of users, where x is given as input arg limit_bottom_user_count Note: since list is a mutable object, method will modify top_tags_list """ if listtype == LOCATIONS: # keep all locations return len(top_list) if listtype == EMOJI: # emoji use a smaller area than tags on the map # therefore we can keep more emoji # (e.g..: use 2 instead of 5) bottomuser_count = math.trunc( self.limit_bottom_user_count/2) else: bottomuser_count = self.limit_bottom_user_count index_min = next( (i for i, (t1, t2) in enumerate( top_list) if t2 < bottomuser_count ), None) if not index_min: return len_before = len(top_list) # delete based on slicing del top_list[index_min:] len_after = len(top_list) if len_before == len_after: # if no change, return return len_after self.log.info( f'Long tail removal: Filtered {len_before - len_after} ' f'{listtype} that were used by less than ' f'{bottomuser_count} users.') return len_after def _compile_cleaned_data(self) -> Optional[Dict[str, CleanedPost]]: """Will produce final cleaned list of items to be processed by clustering. - optionally writes entries to file, if handler exists """ cleaned_post_dict: Dict[str, CleanedPost] = dict() for user_guid, locationhash in self.locations_per_user.items(): # loop all distinct user locations for location in locationhash: locid_userid = f'{location}::{user_guid}' post_latlng = location.split(':') first_post = self.userlocations_firstpost_dict.get( locid_userid, None) if first_post is None: return # create tuple with cleaned photo data cleaned_post = self._compile_cleaned_post( first_post, locid_userid, post_latlng, user_guid) if cleaned_post is None: continue if self.topic_modeling: self._update_topic_models( cleaned_post, user_guid) cleaned_post_dict[cleaned_post.guid] = cleaned_post # update boundary self.bounds.upd_latlng_bounds( cleaned_post.lat, cleaned_post.lng) return cleaned_post_dict def _read_cleaned_data(self, cdata: Path): """Create cleaned post dict from intermediate data file store""" cleaned_post_dict: Dict[str, CleanedPost] = dict() with open(cdata, 'r', newline='', encoding='utf8') as f_handle: cpost_reader = csv.DictReader( f_handle, delimiter=',', quotechar='"', quoting=QUOTE_MINIMAL) for cpost in cpost_reader: # row_num += 1 cleaned_post = PrepareData._parse_cleaned_post(cpost) cleaned_post_dict[cleaned_post.guid] = cleaned_post # update statistics from cleaned post self.add_record(cleaned_post) # update boundary self.bounds.upd_latlng_bounds( cleaned_post.lat, cleaned_post.lng) return cleaned_post_dict def write_topic_models(self): """Initialize two lists for topic modeling output - hashed (anonymized) output (*add salt) - original output """ headerline = "topics,post_ids,user_ids\n" with open( self.output_folder / 'Output_usertopics_anonymized.csv', 'w', encoding='utf8') as csvfile_anon, open( self.output_folder / 'Output_usertopics.csv', 'w', encoding='utf8') as csvfile: dw_list = list() for cfile in (csvfile, csvfile_anon): cfile.write(headerline) datawriter = csv.writer(cfile, delimiter=',', lineterminator='\n', quotechar='"', quoting=csv.QUOTE_NONNUMERIC) dw_list.append(datawriter) self._write_topic_rows(dw_list) def _write_topic_rows(self, dw_list): """Write Topic models to file""" datawriter = dw_list[0] dw_anon = dw_list[1] def _join_encode(keys): joined_keys = ",".join(keys) joined_encoded_keys = ",".join( [Utils.encode_string(post_id) for post_id in keys]) return joined_keys, joined_encoded_keys for user_key, topics in self.user_topiclist_dict.items(): joined_topics = " ".join(topics) post_keys = self.user_post_ids_dict.get(user_key, None) joined_keys, joined_encoded_keys = _join_encode(post_keys) dw_anon.writerow([joined_topics, "{" + joined_encoded_keys + "}", Utils.encode_string(user_key)]) datawriter.writerow([joined_topics, "{" + joined_keys + "}", str(user_key)]) def _update_topic_models(self, cleaned_post_location, user_key): """If Topic Modeling enabled, update required dictionaries with merged words from title, tags and post_body """ if cleaned_post_location.hashtags: self.user_topiclist_dict[user_key] |= \ cleaned_post_location.hashtags # also use descriptions for Topic Modeling self.user_topiclist_dict[user_key] |= \ cleaned_post_location.post_body # Bit wise or and assignment in one step. # -> assign PhotoGuid to UserDict list # if not already contained self.user_post_ids_dict[user_key] |= { cleaned_post_location.guid} # UserPhotoFirstThumb_dict[user_key] = photo[5] @staticmethod def _parse_cleaned_post(cpost: Dict[str, str]) -> CleanedPost: """Process single cleaned post from (file) dict stream""" # process column with concatenate items (";item1;item2") split_string_dict = dict() for split_col in ["post_body", "hashtags", "emoji"]: item_str = cpost.get(split_col) if item_str: items = set(item_str.split(";")) split_string_dict[split_col] = items locid = cpost.get("loc_id") if not locid: raise ValueError("Field loc_id empty.") cleaned_post = CleanedPost( origin_id=int(cpost["origin_id"]), lat=float(cpost["lat"]), lng=float(cpost["lng"]), guid=cpost["guid"], user_guid=cpost["user_guid"], post_body=split_string_dict.get("post_body", set()), post_create_date=cpost.get("post_create_date"), post_publish_date=cpost.get("post_publish_date"), post_views_count=int(cpost.get("post_views_count", 0)), post_like_count=int(cpost.get("post_like_count", 0)), emoji=split_string_dict.get("emoji", set()), hashtags=split_string_dict.get("hashtags", set()), loc_id=locid, loc_name=cpost.get("loc_name") ) return cleaned_post def _compile_cleaned_post(self, first_post, locid_userid, post_latlng, user_key) -> Optional[CleanedPost]: """Merge cleaned post from all posts of a certain user at a specific location. This is producing the final CleanedPost. - some information is not needed, those post attributes are simply skipped (e.g. location name) - some information must not be merged, this can be directly copied from the first post at a location/user (e.g. origin_id - will always be the same for a particular user, post_create_date, post_publish_date) - some information (e.g. hashtags) need merge with removing dupliates: use prepared dictionaries - some important information is type-checked (longitude, latitude) Keyword arguments: first_post -- first post of a user_guid at a location locid_userid -- user_guid and loc_id in merged format (f'{location}::{user_key}') post_latlng -- tuple with lat/lng coordinates user_key -- user_guid Note: ("",) means: substitute empty tuple as default """ merged_wordlist = PrepareData._get_merged( self.userlocation_terms_dict, locid_userid) merged_emojilist = PrepareData._get_merged( self.items_per_userloc[EMOJI], locid_userid) merged_taglist = PrepareData._get_merged( self.items_per_userloc[TAGS], locid_userid) try: lat = float(post_latlng[0]) lng = float(post_latlng[1]) except ValueError: # skip posts with no # or malformed lat/lng return cleaned_post = CleanedPost( origin_id=first_post.origin_id, lat=lat, lng=lng, guid=first_post.guid, user_guid=user_key, post_body=merged_wordlist, post_create_date=first_post.post_create_date, post_publish_date=first_post.post_publish_date, post_views_count=first_post.post_views_count, post_like_count=first_post.post_like_count, emoji=merged_emojilist, hashtags=merged_taglist, loc_id=first_post.loc_id, loc_name=first_post.loc_name ) return cleaned_post @staticmethod def _get_merged(ref_dict: Dict, locid_userid: str) -> Set[str]: """Gets set of words for userlocid from ref dictionary Note: since using defaultdict, keys not found will return empty set() """ value = ref_dict[locid_userid] return value def _write_location_tocsv(self, datawriter, cleaned_post_location: CleanedPost, panon_set=None) -> None: """Writes a single record of cleaned posts to CSV list - write intermediate cleaned post data to file for later use Arguments datawriter - open file file_handle to output file cleaned_post_location - cleaned post of type CleanedPost (namedtuple) panonymize - This will limit written item-lists (emoji, tags, body-content) to the terms that exist in identified toplists. The result is a pseudo- anonymized post that only contains the less identifiable popular terms that are used by many users. """ if panon_set: cleaned_post_location = self._panonymize_cleaned_post( cleaned_post_location, panon_set) ploc_list = PrepareData._cleaned_ploc_tolist( cleaned_post_location) datawriter.writerow(ploc_list) @staticmethod def _panonymize_cleaned_post( upl: CleanedPost, panon_set: Dict[str, Set[str]]) -> CleanedPost: """Returns a new cleaned post with reduced information detail based on global information patterns""" panon_post = CleanedPost( origin_id=upl.origin_id, lat=upl.lat, lng=upl.lng, guid=upl.guid, user_guid=upl.user_guid, post_body=PrepareData._filter_private_terms( upl.post_body, panon_set.get(TAGS)), post_create_date=PrepareData._agg_date(upl.post_create_date), post_publish_date=PrepareData._agg_date(upl.post_publish_date), post_views_count=upl.post_views_count, post_like_count=upl.post_like_count, emoji=PrepareData._filter_private_terms( upl.emoji, panon_set.get(EMOJI)), hashtags=PrepareData._filter_private_terms( upl.hashtags, panon_set.get(TAGS)), loc_id=upl.loc_id, loc_name=upl.loc_name ) return panon_post @staticmethod def _agg_date( str_date: Optional[str]) -> str: """Remove time info from string, e.g. 2010-05-07 16:00:54 to 2010-05-07 """ if str_date: str_date_hr = f'{str_date[:10]}' return str_date_hr return "" @staticmethod def _filter_private_terms( str_list: Optional[Set[str]] = None, top_terms_set: Optional[Set[str]] = None) -> Set[str]: if str_list and top_terms_set: filtered_set = {term for term in str_list if term in top_terms_set} return filtered_set # if none or empty, return empty set return set() @staticmethod def _cleaned_ploc_tolist(cleaned_post_location: CleanedPost) -> List[str]: """Converts a cleaned post structure to list for CSV write""" attr_list = list() for attr in cleaned_post_location: if isinstance(attr, set): attr_list.append(";".join(attr)) else: attr_list.append(attr) return attr_list def _get_cleaned_wordlist(self, post_body_string): cleaned_post_body = Utils.remove_special_chars(post_body_string) cleaned_wordlist = PrepareData._get_wordlist(cleaned_post_body) return cleaned_wordlist @staticmethod def _get_wordlist(cleaned_post_body): """split by space-characterm, filter by length""" wordlist = [word for word in cleaned_post_body.lower().split( ' ') if len(word) > 2] return wordlist def global_stats_report(self, cleaned=None): """Report global stats after data has been read""" if cleaned is None: cleaned = True self.log.info( f'Total user count (UC): ' f'{len(self.locations_per_user)}') upl = sum(len(v) for v in self.locations_per_user.values()) self.log.info( f'Total user post locations (UPL): ' f'{upl}') if not cleaned: return if not self.cleaned_stats: self._init_item_stats() self.log.info( f'Total (cleaned) post count (PC): ' f'{self.count_glob:02d}') self.log.info( f'Total (cleaned) tag count (PTC): ' f'{self.cleaned_stats[TAGS].total_item_count}') self.log.info( f'Total (cleaned) emoji count (PEC): ' f'{self.cleaned_stats[EMOJI].total_item_count}')Methods
def add_record(self, lbsn_post: Union[PostStructure, CleanedPost])-
Method will merge all tags/emoji/terms of a single user for each location (Metric 'UPL') to produce a cleaned version of input data
- further information is derived from the first post for each user-location
- the result is a cleaned output containing reduced information that is necessary for tag maps
- get cleaned output with get_prepared_data()
Expand source code
def add_record( self, lbsn_post: Union[PostStructure, CleanedPost]): """Method will merge all tags/emoji/terms of a single user for each location (Metric 'UPL') to produce a cleaned version of input data - further information is derived from the first post for each user-location - the result is a cleaned output containing reduced information that is necessary for tag maps - get cleaned output with get_prepared_data() """ self.count_glob += 1 self._update_toplists(lbsn_post) # create userid_loc_id, this is used as the base # for clustering data (metric UPL) post_locid_userid: str = f'{lbsn_post.loc_id}::{lbsn_post.user_guid}' if lbsn_post.loc_id and ( lbsn_post.loc_name and lbsn_post.loc_id not in self.locid_locname_dict): # add locname to dict self.locid_locname_dict[ lbsn_post.loc_id] = lbsn_post.loc_name if (lbsn_post.user_guid not in self.locations_per_user or lbsn_post.loc_id not in self.locations_per_user[lbsn_post.user_guid]): # Bit wise or and assignment in one step. # -> assign locID to UserDict list # if not already contained self.locations_per_user[lbsn_post.user_guid] |= \ {lbsn_post.loc_id} # self.stats.count_loc += 1 self.userlocations_firstpost_dict[ post_locid_userid] = lbsn_post # union tags/emoji per userid/unique location if TAGS in self.cluster_types: self.items_per_userloc[TAGS][post_locid_userid] \ |= lbsn_post.hashtags if EMOJI in self.cluster_types: self.items_per_userloc[EMOJI][post_locid_userid] \ |= lbsn_post.emoji if isinstance(lbsn_post, PostStructure): # get cleaned wordlist cleaned_terms_body = set(self._get_cleaned_wordlist( lbsn_post.post_body)) cleaned_terms_title = set(self._get_cleaned_wordlist( lbsn_post.post_title)) cleaned_terms = cleaned_terms_body.union(cleaned_terms_title) else: # words already cleaned cleaned_terms = lbsn_post.post_body # union words per userid/unique location self.userlocation_terms_dict[ post_locid_userid] |= cleaned_terms def get_cleaned_post_dict(self, input_path=None) ‑> Optional[Dict[str, CleanedPost]]-
Output wrapper
- calls loop user locations method
- optionally initializes output to file
Expand source code
def get_cleaned_post_dict( self, input_path=None) -> Optional[Dict[str, CleanedPost]]: """Output wrapper - calls loop user locations method - optionally initializes output to file """ if input_path is None: # load from ingested data cleaned_post_dict = self._compile_cleaned_data() else: # load from file store cleaned_post_dict = self._load_cleaned_data(input_path) return cleaned_post_dict def get_item_stats(self) ‑> Dict[str, PreparedStats]-
After data is loaded, this collects data and stats for distribution of tags, emoji and locations
- prepare data for tag maps clustering
- store to self.data_prepared
Expand source code
def get_item_stats(self) -> Dict[str, PreparedStats]: """After data is loaded, this collects data and stats for distribution of tags, emoji and locations - prepare data for tag maps clustering - store to self.data_prepared """ if not self.cleaned_stats: self._init_item_stats() return self.cleaned_stats def get_panonymized_posts(self, cleaned_post_dict: Optional[Dict[str, CleanedPost]]) ‑> Dict[str, CleanedPost]-
Returns a new cleaned post dict with reduced information detail based on global information patterns
This is not a true anonymization. Returned items have specifically the highly identifyable information removed (specific tags/terms used by few users), which make it harder to identify original users from resulting data.
Expand source code
def get_panonymized_posts( self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] ) -> Dict[str, CleanedPost]: """Returns a new cleaned post dict with reduced information detail based on global information patterns This is not a true anonymization. Returned items have specifically the highly identifyable information removed (specific tags/terms used by few users), which make it harder to identify original users from resulting data. """ panon_cleaned_post_dict = defaultdict(CleanedPost) panon_set = self._get_panon_sets() for upl, cleaned_post in cleaned_post_dict.items(): upl_panon = PrepareData._panonymize_cleaned_post( cleaned_post, panon_set) panon_cleaned_post_dict[upl] = upl_panon return panon_cleaned_post_dict def global_stats_report(self, cleaned=None)-
Report global stats after data has been read
Expand source code
def global_stats_report(self, cleaned=None): """Report global stats after data has been read""" if cleaned is None: cleaned = True self.log.info( f'Total user count (UC): ' f'{len(self.locations_per_user)}') upl = sum(len(v) for v in self.locations_per_user.values()) self.log.info( f'Total user post locations (UPL): ' f'{upl}') if not cleaned: return if not self.cleaned_stats: self._init_item_stats() self.log.info( f'Total (cleaned) post count (PC): ' f'{self.count_glob:02d}') self.log.info( f'Total (cleaned) tag count (PTC): ' f'{self.cleaned_stats[TAGS].total_item_count}') self.log.info( f'Total (cleaned) emoji count (PEC): ' f'{self.cleaned_stats[EMOJI].total_item_count}') def write_cleaned_data(self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] = None, panon: bool = None)-
Write cleaned data to intermediate file
Expand source code
def write_cleaned_data( self, cleaned_post_dict: Optional[Dict[str, CleanedPost]] = None, panon: bool = None): """Write cleaned data to intermediate file""" self.log.info( f'Writing cleaned intermediate ' f'data to file (Output_cleaned.csv)..') if cleaned_post_dict is None: cleaned_post_dict = self.get_cleaned_post_dict() if panon is None: panon = True if panon: panon_set = self._get_panon_sets() else: panon_set = None with open(self.output_folder / 'Output_cleaned.csv', 'w', encoding='utf8') as csvfile: # get headerline from class structure headerline = ','.join(POST_FIELDS) csvfile.write(f'{headerline}\n') # values will be written with CSV writer module datawriter = csv.writer( csvfile, delimiter=',', lineterminator='\n', quotechar='"', quoting=csv.QUOTE_NONNUMERIC) for cleaned_post in cleaned_post_dict.values(): self._write_location_tocsv( datawriter, cleaned_post, panon_set) self.log.info(' done.') def write_topic_models(self)-
Initialize two lists for topic modeling output
- hashed (anonymized) output (*add salt)
- original output
Expand source code
def write_topic_models(self): """Initialize two lists for topic modeling output - hashed (anonymized) output (*add salt) - original output """ headerline = "topics,post_ids,user_ids\n" with open( self.output_folder / 'Output_usertopics_anonymized.csv', 'w', encoding='utf8') as csvfile_anon, open( self.output_folder / 'Output_usertopics.csv', 'w', encoding='utf8') as csvfile: dw_list = list() for cfile in (csvfile, csvfile_anon): cfile.write(headerline) datawriter = csv.writer(cfile, delimiter=',', lineterminator='\n', quotechar='"', quoting=csv.QUOTE_NONNUMERIC) dw_list.append(datawriter) self._write_topic_rows(dw_list) def write_toplists(self)-
Writes toplists (tags, emoji, locations) to file
Expand source code
def write_toplists(self): """Writes toplists (tags, emoji, locations) to file""" for cls_type in self.cluster_types: top_list = self.cleaned_stats[cls_type].top_items_list max_items = self.cleaned_stats[cls_type].max_items PrepareData._write_toplist( top_list, cls_type, max_items, self.output_folder, self.locid_locname_dict)
class PreparedStats (top_items_list: List[ItemCounter], total_unique_items: int, total_item_count: int, max_items: int)-
PreparedStats(top_items_list: List[tagmaps.classes.shared_structure.ItemCounter], total_unique_items: int, total_item_count: int, max_items: int)
Expand source code
@dataclass class PreparedStats: top_items_list: List[ItemCounter] total_unique_items: int total_item_count: int max_items: intClass variables
var max_items : intvar top_items_list : List[ItemCounter]var total_item_count : intvar total_unique_items : int