def _batch_import_urls(cls, tweets):
with_urls = [t for t in tweets if 'entities' in t and 'urls' in t['entities']]
urls = list(chain(*[t['entities']['urls'] for t in with_urls]))
logging.debug('URLs: {0} in {1} tweets'.format(len(urls), len(with_urls)))
cls._save_urls(urls)
url_map = dict(map(lambda x: (x.url, x), Url.objects.filter(url__in=pluck('expanded_url', urls))))
with_media_urls = [t for t in tweets if 'entities' in t and 'media' in t['entities'] and t['entities']['media']]
media_urls = list(chain(*[t['entities']['media'] for t in with_media_urls]))
logging.debug('Media URLs: {0} in {1} tweets'.format(len(media_urls), len(with_media_urls)))
cls._save_media_urls(media_urls)
media_url_map = dict(map(lambda x: (x.internal_id, x), Media.objects.filter(internal_id__in=pluck('id_str', media_urls))))
tweet_map = Tweet.objects.in_bulk(pluck('__pk__', tweets))
logging.debug('URLs saved: {0}'.format(len(url_map)))
logging.debug('Media URLs saved: {0}'.format(len(media_url_map)))
for tweet in tweets:
tweet_urls = [url_map[url['expanded_url']] for url in tweet['entities']['urls'] if url['expanded_url'] in url_map]
if tweet_urls:
tweet_map[tweet['__pk__']].links = tweet_urls
if 'media' in tweet['entities'] and tweet['entities']['media']:
print(tweet['entities']['media'])
tweet_media_urls = [media_url_map[media_url['id_str']] for media_url in tweet['entities']['media'] if media_url['id_str'] in media_url_map]
if tweet_media_urls:
tweet_map[tweet['__pk__']].media = tweet_media_urls
def test_url(self) -> None:
"""Test the generated url and get the required parameters from the service."""
try:
resp = self.session.get(self.base_url, {"f": "json"}).json()
try:
self.units = resp["units"].replace("esri", "").lower()
except KeyError:
self.units = None
self._max_nrecords = int(resp["maxRecordCount"])
self.query_formats = resp["supportedQueryFormats"].replace(
" ", "").lower().split(",")
self.valid_fields = list(
set(
utils.traverse_json(resp, ["fields", "name"]) +
utils.traverse_json(resp, ["fields", "alias"]) + ["*"]))
try:
extent = resp["extent"] if "extent" in resp else resp[
"fullExtent"]
bounds = (extent["xmin"], extent["ymin"], extent["xmax"],
extent["ymax"])
crs = extent["spatialReference"]["latestWkid"]
self.extent = utils.MatchCRS.bounds(bounds, crs, DEF_CRS)
except KeyError:
self.extent = None
try:
self.feature_types = dict(
zip((tlz.pluck("id", resp["types"])),
tlz.pluck("name", resp["types"])))
except KeyError:
self.feature_types = None
except KeyError:
raise ServerError(self.base_url)
def _batch_import_keywords(cls, tweets):
kws = set()
for kw in pluck('keywords', tweets):
kws.update(kw)
keyword_map = dict(map(lambda x: (x.name, x), Keyword.retrieve(kws)))
tweet_map = Tweet.objects.in_bulk(pluck('__pk__', tweets))
logging.debug('Keywords: {0} in {1} tweets'.format(len(keyword_map), len(tweet_map)))
for tweet in tweets:
tweet_kws = [keyword_map[k] for k in tweet['keywords']]
tweet_map[tweet['__pk__']].keywords = tweet_kws
def _batch_import_urls(cls, tweets):
with_urls = [
t for t in tweets if 'entities' in t and 'urls' in t['entities']
]
urls = list(chain(*[t['entities']['urls'] for t in with_urls]))
logging.debug('URLs: {0} in {1} tweets'.format(len(urls),
len(with_urls)))
cls._save_urls(urls)
url_map = dict(
map(lambda x: (x.url, x),
Url.objects.filter(url__in=pluck('expanded_url', urls))))
with_media_urls = [
t for t in tweets if 'entities' in t and 'media' in t['entities']
and t['entities']['media']
]
media_urls = list(
chain(*[t['entities']['media'] for t in with_media_urls]))
logging.debug('Media URLs: {0} in {1} tweets'.format(
len(media_urls), len(with_media_urls)))
cls._save_media_urls(media_urls)
media_url_map = dict(
map(
lambda x: (x.internal_id, x),
Media.objects.filter(
internal_id__in=pluck('id_str', media_urls))))
tweet_map = Tweet.objects.in_bulk(pluck('__pk__', tweets))
logging.debug('URLs saved: {0}'.format(len(url_map)))
logging.debug('Media URLs saved: {0}'.format(len(media_url_map)))
for tweet in tweets:
tweet_urls = [
url_map[url['expanded_url']]
for url in tweet['entities']['urls']
if url['expanded_url'] in url_map
]
if tweet_urls:
tweet_map[tweet['__pk__']].links = tweet_urls
if 'media' in tweet['entities'] and tweet['entities']['media']:
print(tweet['entities']['media'])
tweet_media_urls = [
media_url_map[media_url['id_str']]
for media_url in tweet['entities']['media']
if media_url['id_str'] in media_url_map
]
if tweet_media_urls:
tweet_map[tweet['__pk__']].media = tweet_media_urls
def _batch_import_keywords(cls, tweets):
kws = set()
for kw in pluck('keywords', tweets):
kws.update(kw)
keyword_map = dict(map(lambda x: (x.name, x), Keyword.retrieve(kws)))
tweet_map = Tweet.objects.in_bulk(pluck('__pk__', tweets))
logging.debug('Keywords: {0} in {1} tweets'.format(
len(keyword_map), len(tweet_map)))
for tweet in tweets:
tweet_kws = [keyword_map[k] for k in tweet['keywords']]
tweet_map[tweet['__pk__']].keywords = tweet_kws
def get_files(self, item: str) -> Dict[str, Tuple[str, str]]:
"""Get all the available zip files in an item."""
url = "https://www.sciencebase.gov/catalog/item"
payload = {"fields": "files,downloadUri", "format": "json"}
r = self.session.get(f"{url}/{item}", payload=payload).json()
files_url = zip(tlz.pluck("name", r["files"]),
tlz.pluck("url", r["files"]))
# TODO: Add units
meta = "".join(tlz.pluck("metadataHtmlViewUri", r["files"],
default=""))
return {
f.replace("_CONUS.zip", ""): (u, meta)
for f, u in files_url if ".zip" in f
}
def _get_py(self, key):
if isinstance(key, tuple):
assert len(key) == 2
rows, cols = key
usecols = cols
ds = self.dshape.subshape[rows, cols]
usecols = None if isinstance(usecols, slice) else listpack(usecols)
else:
rows = key
ds = self.dshape.subshape[rows]
usecols = None
if isinstance(ds, DataShape) and isdimension(ds[0]):
ds = ds.subshape[0]
seq = self._iter(usecols=usecols)
if isinstance(key, tuple) and isinstance(cols, _strtypes + _inttypes):
seq = pluck(0, seq)
seq = coerce(ds, seq)
if isinstance(rows, compatibility._inttypes):
line = nth(rows, seq)
try:
return next(line).item()
except TypeError:
try:
return line.item()
except AttributeError:
return line
elif isinstance(rows, list):
return nth_list(rows, seq)
elif isinstance(rows, slice):
return it.islice(seq, rows.start, rows.stop, rows.step)
else:
raise IndexError("key %r is not valid" % rows)
def get_stats(self, per_worker=False):
individual_stats = _each(self.ctx, yappi.get_func_stats)
if per_worker:
return individual_stats
else:
stat, *rest = pluck(1, individual_stats)
# merging adapted from _add_from_YSTAT
for other in rest:
for saved_stat in other:
if saved_stat not in stat:
stat._idx_max += 1
saved_stat.index = stat._idx_max
stat.append(saved_stat)
# fix children's index values
for saved_stat in other:
for saved_child_stat in saved_stat.children:
# we know for sure child's index is pointing to a valid stat in saved_stats
# so as saved_stat is already in sync. (in above loop), we can safely assume
# that we shall point to a valid stat in current_stats with the child's full_name
saved_child_stat.index = stat[
saved_child_stat.full_name].index
# merge stats
for saved_stat in other:
saved_stat_in_curr = stat[saved_stat.full_name]
saved_stat_in_curr += saved_stat
return stat
def _check_contents(self, dset, duplication=1):
files = sorted(dset.icollect(),
key=lambda file: int(
basename(file[0]).replace('test_file_', '').replace(
'.tmp', '')))
expected = b''.join(interleave([self.contents] * duplication))
self.assertEqual(b''.join(pluck(1, files)), expected)
def _get_py(self, key):
if isinstance(key, tuple):
assert len(key) == 2
rows, cols = key
usecols = cols
ds = self.dshape.subshape[rows, cols]
usecols = None if isinstance(usecols, slice) else listpack(usecols)
else:
rows = key
ds = self.dshape.subshape[rows]
usecols = None
if isinstance(ds, DataShape) and isdimension(ds[0]):
ds = ds.subshape[0]
seq = self._iter(usecols=usecols)
if isinstance(key, tuple) and isinstance(cols, _strtypes + _inttypes):
seq = pluck(0, seq)
seq = coerce(ds, seq)
if isinstance(rows, compatibility._inttypes):
line = nth(rows, seq)
try:
return next(line).item()
except TypeError:
try:
return line.item()
except AttributeError:
return line
elif isinstance(rows, list):
return nth_list(rows, seq)
elif isinstance(rows, slice):
return it.islice(seq, rows.start, rows.stop, rows.step)
else:
raise IndexError("key %r is not valid" % rows)
def call_function(func, func_token, args, kwargs, pure=None, nout=None):
dask_key_name = kwargs.pop('dask_key_name', None)
pure = kwargs.pop('pure', pure)
if dask_key_name is None:
name = '%s-%s' % (funcname(func),
tokenize(func_token, *args, pure=pure, **kwargs))
else:
name = dask_key_name
dsk = sharedict.ShareDict()
args_dasks = list(map(to_task_dask, args))
for arg, d in args_dasks:
if isinstance(d, sharedict.ShareDict):
dsk.update_with_key(d)
elif isinstance(arg, (str, tuple)):
dsk.update_with_key(d, key=arg)
else:
dsk.update(d)
args = tuple(pluck(0, args_dasks))
if kwargs:
dask_kwargs, dsk2 = to_task_dask(kwargs)
dsk.update(dsk2)
task = (apply, func, list(args), dask_kwargs)
else:
task = (func,) + args
dsk.update_with_key({name: task}, key=name)
nout = nout if nout is not None else None
return Delayed(name, dsk, length=nout)
def batch_import(self, tweets, parse_json=False):
logging.info('Received {0} tweets'.format(len(tweets)))
tweets = list(filter(self.accept_tweet, map(self.pre_func, tweets)))
logging.info('After Filtering: {0} tweets'.format(len(tweets)))
user_fn = lambda x: pluck('user', x)
tweet_fn = lambda x: x
retweet_sources = [t['retweeted_status'] for t in tweets if 'retweeted_status' in t]
quote_sources = [t['quoted_status'] for t in tweets if 'quoted_status' in t]
imported_users = Importer._batch_import(User, tweets, user_fn)
imported_users.extend(Importer._batch_import(User, retweet_sources, user_fn))
imported_users.extend(Importer._batch_import(User, quote_sources, user_fn))
imported = Importer._batch_import(Tweet, tweets, tweet_fn)
imported.extend(Importer._batch_import(Tweet, [t['retweeted_status'] for t in tweets if 'retweeted_status' in t], tweet_fn))
imported.extend(Importer._batch_import(Tweet, [t['quoted_status'] for t in tweets if 'quoted_status' in t], tweet_fn))
Importer._batch_import_retweets(list(filter(lambda x: 'retweeted_status' in x, tweets)))
Importer._batch_import_retweets(list(filter(lambda x: 'quoted_status' in x, tweets)), retweet_key='quoted_status')
Importer._batch_import_keywords(list(filter(lambda x: x['__created__'], tweets)))
Importer._batch_import_keywords(list(filter(lambda x: x['__created__'], retweet_sources)))
Importer._batch_import_keywords(list(filter(lambda x: x['__created__'], quote_sources)))
Importer._batch_import_urls(list(filter(lambda x: x['__created__'], tweets)))
Importer._batch_import_urls(list(filter(lambda x: x['__created__'], retweet_sources)))
Importer._batch_import_urls(list(filter(lambda x: x['__created__'], quote_sources)))
self.post_func(imported_users, imported)
logging.debug('Imported Tweets: {0}'.format(len(imported)))
return imported_users, imported
def build_graph(estimator, cv, scorer, candidate_params, X, y=None,
groups=None, fit_params=None, iid=True, refit=True,
error_score='raise', return_train_score=True, cache_cv=True):
X, y, groups = to_indexable(X, y, groups)
cv = check_cv(cv, y, is_classifier(estimator))
# "pairwise" estimators require a different graph for CV splitting
is_pairwise = getattr(estimator, '_pairwise', False)
dsk = {}
X_name, y_name, groups_name = to_keys(dsk, X, y, groups)
n_splits = compute_n_splits(cv, X, y, groups)
if fit_params:
# A mapping of {name: (name, graph-key)}
param_values = to_indexable(*fit_params.values(), allow_scalars=True)
fit_params = {k: (k, v) for (k, v) in
zip(fit_params, to_keys(dsk, *param_values))}
else:
fit_params = {}
fields, tokens, params = normalize_params(candidate_params)
main_token = tokenize(normalize_estimator(estimator), fields, params,
X_name, y_name, groups_name, fit_params, cv,
error_score == 'raise', return_train_score)
cv_name = 'cv-split-' + main_token
dsk[cv_name] = (cv_split, cv, X_name, y_name, groups_name,
is_pairwise, cache_cv)
if iid:
weights = 'cv-n-samples-' + main_token
dsk[weights] = (cv_n_samples, cv_name)
else:
weights = None
scores = do_fit_and_score(dsk, main_token, estimator, cv_name, fields,
tokens, params, X_name, y_name, fit_params,
n_splits, error_score, scorer,
return_train_score)
cv_results = 'cv-results-' + main_token
candidate_params_name = 'cv-parameters-' + main_token
dsk[candidate_params_name] = (decompress_params, fields, params)
dsk[cv_results] = (create_cv_results, scores, candidate_params_name,
n_splits, error_score, weights)
keys = [cv_results]
if refit:
best_params = 'best-params-' + main_token
dsk[best_params] = (get_best_params, candidate_params_name, cv_results)
best_estimator = 'best-estimator-' + main_token
if fit_params:
fit_params = (dict, (zip, list(fit_params.keys()),
list(pluck(1, fit_params.values()))))
dsk[best_estimator] = (fit_best, clone(estimator), best_params,
X_name, y_name, fit_params)
keys.append(best_estimator)
return dsk, keys, n_splits
def call_function(func, func_token, args, kwargs, pure=None, nout=None):
dask_key_name = kwargs.pop('dask_key_name', None)
pure = kwargs.pop('pure', pure)
if dask_key_name is None:
name = '%s-%s' % (funcname(func),
tokenize(func_token, *args, pure=pure, **kwargs))
else:
name = dask_key_name
dsk = sharedict.ShareDict()
args_dasks = list(map(to_task_dask, args))
for arg, d in args_dasks:
if isinstance(d, sharedict.ShareDict):
dsk.update_with_key(d)
elif isinstance(arg, (str, tuple)):
dsk.update_with_key(d, key=arg)
else:
dsk.update(d)
args = tuple(pluck(0, args_dasks))
if kwargs:
dask_kwargs, dsk2 = to_task_dask(kwargs)
dsk.update(dsk2)
task = (apply, func, list(args), dask_kwargs)
else:
task = (func,) + args
dsk.update_with_key({name: task}, key=name)
nout = nout if nout is not None else None
return Delayed(name, dsk, length=nout)
def _remote(self):
contents = pluck(1, super()._remote())
if self.dset.encoding is not None:
contents = map(
partial(_decode, self.dset.encoding, self.dset.errors),
contents)
lines = map(partial(_splitlines, True), contents)
return chain.from_iterable(lines)
def take_snapshot(self, per_worker=False):
snapshots = _each(self.ctx, tracemalloc.take_snapshot)
if per_worker:
return snapshots
snapshot_merged, *snapshots = pluck(1, snapshots)
traces_merged = snapshot_merged.traces._traces
for s in snapshots:
traces_merged.extend(s.traces._traces)
return snapshot_merged
def _compress_letter_and_initabbr(words):
if not words:
return words
result = []
for word_len, words_grp in groupby(
words, lambda x: len(x[0])): # compress single letter words
if word_len == 1:
result.append(("".join(pluck(0, words_grp)), None))
else:
result.extend(words_grp)
if (result and 2 <= len(result[0]) <= len(result) - 1
and # remove if first word is an abbrevation of following words, e.g. "ABC Aa Bb Cc Company"
all(char == word[0]
for char, word in zip(result[0][0], pluck(0, result[1:])))):
result = result[1:]
return result
def batch_import(self, tweets, parse_json=False):
logging.info('Received {0} tweets'.format(len(tweets)))
tweets = list(filter(self.accept_tweet, map(self.pre_func, tweets)))
logging.info('After Filtering: {0} tweets'.format(len(tweets)))
user_fn = lambda x: pluck('user', x)
tweet_fn = lambda x: x
retweet_sources = [
t['retweeted_status'] for t in tweets if 'retweeted_status' in t
]
quote_sources = [
t['quoted_status'] for t in tweets if 'quoted_status' in t
]
imported_users = Importer._batch_import(User, tweets, user_fn)
imported_users.extend(
Importer._batch_import(User, retweet_sources, user_fn))
imported_users.extend(
Importer._batch_import(User, quote_sources, user_fn))
imported = Importer._batch_import(Tweet, tweets, tweet_fn)
imported.extend(
Importer._batch_import(Tweet, [
t['retweeted_status']
for t in tweets if 'retweeted_status' in t
], tweet_fn))
imported.extend(
Importer._batch_import(
Tweet,
[t['quoted_status']
for t in tweets if 'quoted_status' in t], tweet_fn))
Importer._batch_import_retweets(
list(filter(lambda x: 'retweeted_status' in x, tweets)))
Importer._batch_import_retweets(list(
filter(lambda x: 'quoted_status' in x, tweets)),
retweet_key='quoted_status')
Importer._batch_import_keywords(
list(filter(lambda x: x['__created__'], tweets)))
Importer._batch_import_keywords(
list(filter(lambda x: x['__created__'], retweet_sources)))
Importer._batch_import_keywords(
list(filter(lambda x: x['__created__'], quote_sources)))
Importer._batch_import_urls(
list(filter(lambda x: x['__created__'], tweets)))
Importer._batch_import_urls(
list(filter(lambda x: x['__created__'], retweet_sources)))
Importer._batch_import_urls(
list(filter(lambda x: x['__created__'], quote_sources)))
self.post_func(imported_users, imported)
logging.debug('Imported Tweets: {0}'.format(len(imported)))
return imported_users, imported
def toolz_max(self, col, rows):
"""
Max of values for a given column in a list of dictionaries
Args:
col (str): Column to process
rows (list): Records
"""
return max(list(toolz.pluck(col, rows)))
def toolz_count(self, col, rows):
"""
Unique values for a given column in a list of dictionaries
Args:
col (str): Column to process
rows (list): Records
"""
return len(set(list(toolz.pluck(col, rows))))
def _group_ids_by_index(index, tokens):
id_groups = []
def new_group():
o = []
id_groups.append(o)
return o.append
_id_groups = defaultdict(new_group)
for n, t in enumerate(pluck(index, tokens)):
_id_groups[t](n)
return id_groups
def _group_ids_by_index(index, tokens):
id_groups = []
def new_group():
o = []
id_groups.append(o)
return o.append
_id_groups = defaultdict(new_group)
for n, t in enumerate(pluck(index, tokens)):
_id_groups[t](n)
return id_groups
def increment_rt_counts(tweet_pks):
"""
:param tweet_pks: dictionary {tweet_pk: rt_count}
:return:
"""
items = sorted(tweet_pks.items(), key=lambda x: x[1], reverse=True)
grouped = groupby(lambda x: x[1], items)
for incr, pairs in grouped.items():
if incr > 0:
pks = pluck(0, pairs)
TweetFeatures.objects.filter(tweet_id__in=pks).update(count_rts=F('count_rts') + incr)
def pre_compute(expr, seq):
try:
if isinstance(seq, Iterator):
first = next(seq)
seq = concat([[first], seq])
else:
first = next(iter(seq))
except StopIteration:
return []
if isinstance(first, dict):
return pluck(expr.fields, seq)
else:
return seq
def toolz_sum(self, col, rows, dtype=None):
"""
Sum a given column in a list of dictionaries
Args:
col (str): Column to process
rows (list): Records
"""
values = list(toolz.pluck(col, rows))
if dtype is not None:
values = [dtype(v) for v in values]
return sum(values)
def toolz_avg(self, col, rows):
"""
Unique values for a given column in a list of dictionaries
Args:
col (str): Column to process
rows (list): Records
"""
count = len(rows)
total = sum(list(toolz.pluck(col, rows)))
return total / count if count > 0 else None
def _batch_import_retweets(cls, tweets, retweet_key='retweeted_status'):
rts = []
user_map = User.objects.in_bulk(pluck('__pk__', pluck('user', tweets)))
tweet_ids = list(pluck('__pk__', tweets))
tweet_ids.extend(pluck('__pk__', pluck(retweet_key, tweets)))
tweet_map = Tweet.objects.in_bulk(tweet_ids)
rt_counts = defaultdict(int)
for current in tweets:
rts.append(ReTweet(user=user_map[current['user']['__pk__']],
tweet_instance=tweet_map[current['__pk__']],
source_tweet=tweet_map[current[retweet_key]['__pk__']],
datetime=tweet_map[current['__pk__']].datetime))
rt_counts[current['__pk__']] += 1
ReTweet.objects.bulk_create(rts)
logging.debug('RT Increments: {0}'.format(rt_counts))
increment_rt_counts(rt_counts)
def increment_rt_counts(tweet_pks):
"""
:param tweet_pks: dictionary {tweet_pk: rt_count}
:return:
"""
items = sorted(tweet_pks.items(), key=lambda x: x[1], reverse=True)
grouped = groupby(lambda x: x[1], items)
for incr, pairs in grouped.items():
if incr > 0:
pks = pluck(0, pairs)
TweetFeatures.objects.filter(tweet_id__in=pks).update(
count_rts=F('count_rts') + incr)
def _into_iter_mongodb(l, coll, columns=None, schema=None):
""" Into helper function
Return both a lazy sequence of tuples and a list of column names
"""
seq = coll.find()
if not columns and schema:
columns = schema[0].names
elif not columns:
item = next(seq)
seq = concat([[item], seq])
columns = sorted(item.keys())
columns.remove('_id')
return columns, pluck(columns, seq)
def _batch_import_retweets(cls, tweets, retweet_key='retweeted_status'):
rts = []
user_map = User.objects.in_bulk(pluck('__pk__', pluck('user', tweets)))
tweet_ids = list(pluck('__pk__', tweets))
tweet_ids.extend(pluck('__pk__', pluck(retweet_key, tweets)))
tweet_map = Tweet.objects.in_bulk(tweet_ids)
rt_counts = defaultdict(int)
for current in tweets:
rts.append(
ReTweet(user=user_map[current['user']['__pk__']],
tweet_instance=tweet_map[current['__pk__']],
source_tweet=tweet_map[current[retweet_key]['__pk__']],
datetime=tweet_map[current['__pk__']].datetime))
rt_counts[current['__pk__']] += 1
ReTweet.objects.bulk_create(rts)
logging.debug('RT Increments: {0}'.format(rt_counts))
increment_rt_counts(rt_counts)
def pre_compute(expr, seq, scope=None, **kwargs):
try:
if isinstance(seq, Iterator):
first = next(seq)
seq = concat([[first], seq])
else:
first = next(iter(seq))
except StopIteration:
return []
if isinstance(first, dict):
leaf = expr._leaves()[0]
return pluck(leaf.fields, seq)
else:
return seq
def normalize_params(params):
"""Take a list of dictionaries, and tokenize/normalize."""
# Collect a set of all fields
fields = set()
for p in params:
fields.update(p)
fields = sorted(fields)
params2 = list(pluck(fields, params, MISSING))
# Non-basic types (including MISSING) are unique to their id
tokens = [tuple(x if isinstance(x, (int, float, str)) else id(x)
for x in p) for p in params2]
return fields, tokens, params2
def pre_compute(expr, seq, scope=None, **kwargs):
try:
if isinstance(seq, Iterator):
first = next(seq)
seq = concat([[first], seq])
else:
first = next(iter(seq))
except StopIteration:
return []
if isinstance(first, dict):
leaf = expr._leaves()[0]
return pluck(leaf.fields, seq)
else:
return seq
def _into_iter_mongodb(l, coll, columns=None, schema=None):
""" Into helper function
Return both a lazy sequence of tuples and a list of column names
"""
seq = coll.find()
if not columns and schema:
columns = schema[0].names
elif not columns:
item = next(seq)
seq = concat([[item], seq])
columns = sorted(item.keys())
columns.remove('_id')
return columns, pluck(columns, seq)
def trans_iter():
for kdnr, groups in groupby(map(dict, pluck(1, trans_dat.iterrows())),
key=itemgetter("kdnr")):
aggs = [
FrequencyTracker(varname, varfunc, ["timestamp"], 0)
for varname, varfunc in [
("land", itemgetter("ref_land")),
("kanal", lambda x: str(x["UTU5_EINGABE_NAME"])[:5]),
]
] + [TimestampSameCustomer()]
for row in groups:
for agg in aggs:
agg.add(row)
row.update(agg.value)
yield row
def normalize_params(params):
"""Take a list of dictionaries, and tokenize/normalize."""
# Collect a set of all fields
fields = set()
for p in params:
fields.update(p)
fields = sorted(fields)
params2 = list(pluck(fields, params, MISSING))
# Non-basic types (including MISSING) are unique to their id
tokens = [
tuple(x if isinstance(x, (int, float, str)) else id(x) for x in p)
for p in params2
]
return fields, tokens, params2
def _batch_import(base_class, cls, elements, fn):
logging.debug('Trying to import {1} from {0} elements'.format(
len(elements), cls))
internal_ids = set(pluck('id_str', fn(elements)))
existing_users = cls.objects.filter(internal_id__in=internal_ids)
existing_ids = set([u.internal_id for u in existing_users])
user_pks = dict([(u.internal_id, u.pk) for u in existing_users])
new_ids = internal_ids - existing_ids
logging.debug('Existing IDs: {0}'.format(len(existing_ids)))
logging.debug('New IDs: {0}'.format(len(new_ids)))
added_keys = set()
new_elements = []
for element in fn(elements):
if element['id_str'] in user_pks:
element['__pk__'] = user_pks[element['id_str']]
element['__created__'] = False
else:
if not element['id_str'] in added_keys:
user_model = cls()
user_model.copy_json(valfilter(lambda x: x, element))
new_elements.append(user_model)
element['__created__'] = True
element['__pk__'] = None
added_keys.add(element['id_str'])
cls.objects.bulk_create(new_elements)
new_models = list(cls.objects.filter(internal_id__in=new_ids))
logging.debug('New IDs created successfully: {0}'.format(
len(new_models)))
new_pks = dict([(u.internal_id, u.pk) for u in new_models])
for element in fn(elements):
if element['id_str'] in new_pks:
element['__pk__'] = new_pks[element['id_str']]
return new_models
def _batch_import(base_class, cls, elements, fn):
logging.debug('Trying to import {1} from {0} elements'.format(len(elements), cls))
internal_ids = set(pluck('id_str', fn(elements)))
existing_users = cls.objects.filter(internal_id__in=internal_ids)
existing_ids = set([u.internal_id for u in existing_users])
user_pks = dict([(u.internal_id, u.pk) for u in existing_users])
new_ids = internal_ids - existing_ids
logging.debug('Existing IDs: {0}'.format(len(existing_ids)))
logging.debug('New IDs: {0}'.format(len(new_ids)))
added_keys = set()
new_elements = []
for element in fn(elements):
if element['id_str'] in user_pks:
element['__pk__'] = user_pks[element['id_str']]
element['__created__'] = False
else:
if not element['id_str'] in added_keys:
user_model = cls()
user_model.copy_json(valfilter(lambda x: x, element))
new_elements.append(user_model)
element['__created__'] = True
element['__pk__'] = None
added_keys.add(element['id_str'])
cls.objects.bulk_create(new_elements)
new_models = list(cls.objects.filter(internal_id__in=new_ids))
logging.debug('New IDs created successfully: {0}'.format(len(new_models)))
new_pks = dict([(u.internal_id, u.pk) for u in new_models])
for element in fn(elements):
if element['id_str'] in new_pks:
element['__pk__'] = new_pks[element['id_str']]
return new_models
def build_graph(estimator,
cv,
scorer,
candidate_params,
X,
y=None,
groups=None,
fit_params=None,
iid=True,
refit=True,
error_score='raise',
return_train_score=True,
cache_cv=True):
X, y, groups = to_indexable(X, y, groups)
cv = check_cv(cv, y, is_classifier(estimator))
# "pairwise" estimators require a different graph for CV splitting
is_pairwise = getattr(estimator, '_pairwise', False)
dsk = {}
X_name, y_name, groups_name = to_keys(dsk, X, y, groups)
n_splits = compute_n_splits(cv, X, y, groups)
if fit_params:
# A mapping of {name: (name, graph-key)}
param_values = to_indexable(*fit_params.values(), allow_scalars=True)
fit_params = {
k: (k, v)
for (k, v) in zip(fit_params, to_keys(dsk, *param_values))
}
else:
fit_params = {}
fields, tokens, params = normalize_params(candidate_params)
main_token = tokenize(normalize_estimator(estimator), fields, params,
X_name, y_name, groups_name, fit_params, cv,
error_score == 'raise', return_train_score)
cv_name = 'cv-split-' + main_token
dsk[cv_name] = (cv_split, cv, X_name, y_name, groups_name, is_pairwise,
cache_cv)
if iid:
weights = 'cv-n-samples-' + main_token
dsk[weights] = (cv_n_samples, cv_name)
else:
weights = None
scores = do_fit_and_score(dsk, main_token, estimator, cv_name, fields,
tokens, params, X_name, y_name, fit_params,
n_splits, error_score, scorer,
return_train_score)
cv_results = 'cv-results-' + main_token
candidate_params_name = 'cv-parameters-' + main_token
dsk[candidate_params_name] = (decompress_params, fields, params)
dsk[cv_results] = (create_cv_results, scores, candidate_params_name,
n_splits, error_score, weights)
keys = [cv_results]
if refit:
best_params = 'best-params-' + main_token
dsk[best_params] = (get_best_params, candidate_params_name, cv_results)
best_estimator = 'best-estimator-' + main_token
if fit_params:
fit_params = (dict, (zip, list(fit_params.keys()),
list(pluck(1, fit_params.values()))))
dsk[best_estimator] = (fit_best, clone(estimator), best_params, X_name,
y_name, fit_params)
keys.append(best_estimator)
return dsk, keys, n_splits
def _do_fit_step(dsk, next_token, step, cv, fields, tokens, params, Xs, ys,
fit_params, n_splits, error_score, step_fields_lk,
fit_params_lk, field_to_index, step_name, none_passthrough,
is_transform):
sub_fields, sub_inds = map(list, unzip(step_fields_lk[step_name], 2))
sub_fit_params = fit_params_lk[step_name]
if step_name in field_to_index:
# The estimator may change each call
new_fits = {}
new_Xs = {}
est_index = field_to_index[step_name]
for ids in _group_ids_by_index(est_index, tokens):
# Get the estimator for this subgroup
sub_est = params[ids[0]][est_index]
if sub_est is MISSING:
sub_est = step
# If an estimator is `None`, there's nothing to do
if sub_est is None:
nones = dict.fromkeys(ids, None)
new_fits.update(nones)
if is_transform:
if none_passthrough:
new_Xs.update(zip(ids, get(ids, Xs)))
else:
new_Xs.update(nones)
else:
# Extract the proper subset of Xs, ys
sub_Xs = get(ids, Xs)
sub_ys = get(ids, ys)
# Only subset the parameters/tokens if necessary
if sub_fields:
sub_tokens = list(pluck(sub_inds, get(ids, tokens)))
sub_params = list(pluck(sub_inds, get(ids, params)))
else:
sub_tokens = sub_params = None
if is_transform:
sub_fits, sub_Xs = do_fit_transform(
dsk, next_token, sub_est, cv, sub_fields, sub_tokens,
sub_params, sub_Xs, sub_ys, sub_fit_params, n_splits,
error_score)
new_Xs.update(zip(ids, sub_Xs))
new_fits.update(zip(ids, sub_fits))
else:
sub_fits = do_fit(dsk, next_token, sub_est, cv, sub_fields,
sub_tokens, sub_params, sub_Xs, sub_ys,
sub_fit_params, n_splits, error_score)
new_fits.update(zip(ids, sub_fits))
# Extract lists of transformed Xs and fit steps
all_ids = list(range(len(Xs)))
if is_transform:
Xs = get(all_ids, new_Xs)
fits = get(all_ids, new_fits)
elif step is None:
# Nothing to do
fits = [None] * len(Xs)
if not none_passthrough:
Xs = fits
else:
# Only subset the parameters/tokens if necessary
if sub_fields:
sub_tokens = list(pluck(sub_inds, tokens))
sub_params = list(pluck(sub_inds, params))
else:
sub_tokens = sub_params = None
if is_transform:
fits, Xs = do_fit_transform(dsk, next_token, step, cv, sub_fields,
sub_tokens, sub_params, Xs, ys,
sub_fit_params, n_splits, error_score)
else:
fits = do_fit(dsk, next_token, step, cv, sub_fields, sub_tokens,
sub_params, Xs, ys, sub_fit_params, n_splits,
error_score)
return (fits, Xs) if is_transform else (fits, None)
def _do_featureunion(dsk, next_token, est, cv, fields, tokens, params, Xs, ys,
fit_params, n_splits, error_score):
if 'transformer_list' in fields:
raise NotImplementedError("Setting FeatureUnion.transformer_list "
"in a gridsearch")
(field_to_index,
step_fields_lk) = _group_subparams(est.transformer_list,
fields,
ignore=('transformer_weights'))
fit_params_lk = _group_fit_params(est.transformer_list, fit_params)
token = next_token(est)
n_samples = _do_n_samples(dsk, token, Xs, n_splits)
fit_steps = []
tr_Xs = []
for (step_name, step) in est.transformer_list:
fits, out_Xs = _do_fit_step(dsk, next_token, step, cv, fields, tokens,
params, Xs, ys, fit_params, n_splits,
error_score, step_fields_lk, fit_params_lk,
field_to_index, step_name, False, True)
fit_steps.append(fits)
tr_Xs.append(out_Xs)
# Rebuild the FeatureUnions
step_names = [n for n, _ in est.transformer_list]
if 'transformer_weights' in field_to_index:
index = field_to_index['transformer_weights']
weight_lk = {}
weight_tokens = list(pluck(index, tokens))
for i, tok in enumerate(weight_tokens):
if tok not in weight_lk:
weights = params[i][index]
if weights is MISSING:
weights = est.transformer_weights
lk = weights or {}
weight_list = [lk.get(n) for n in step_names]
weight_lk[tok] = (weights, weight_list)
weights = get(weight_tokens, weight_lk)
else:
lk = est.transformer_weights or {}
weight_list = [lk.get(n) for n in step_names]
weight_tokens = repeat(None)
weights = repeat((est.transformer_weights, weight_list))
out = []
out_append = out.append
fit_name = 'feature-union-' + token
tr_name = 'feature-union-concat-' + token
m = 0
seen = {}
for steps, Xs, wt, (w, wl), nsamp in zip(zip(*fit_steps), zip(*tr_Xs),
weight_tokens, weights,
n_samples):
if (steps, wt) in seen:
out_append(seen[steps, wt])
else:
for n in range(n_splits):
dsk[(fit_name, m,
n)] = (feature_union, step_names,
[None if s is None else s + (n, )
for s in steps], w)
dsk[(tr_name, m,
n)] = (feature_union_concat,
[None if x is None else x + (n, )
for x in Xs], nsamp + (n, ), wl)
seen[steps, wt] = m
out_append(m)
m += 1
return [(fit_name, i) for i in out], [(tr_name, i) for i in out]
sns.violinplot(x='diff', y='cond', data=freq_diffs)
#%%
clip = np.percentile(freq_diffs['diff'], [2.5, 97.5])
for cond, data in freq_diffs.groupby('cond'):
sns.kdeplot(data['diff'], clip=clip, label=cond)
plt.savefig('figures/word_freq_of_suggs.pdf')
#%%
import random
acceptability_frames = []
for group in cytoolz.partition_all(len(conditions), samples):
context = group[0]['context']
meta = list(
cytoolz.pluck(['review_idx', 'sent_idx', 'word_idx', 'true_follows'],
group))
assert len(set(meta)) == 1
meta = list(meta[0])
true_follows = meta.pop(-1)
options = [('true', group[0]['true_follows'])]
for sample in group:
for sugg in sample['suggs'].split('\n')[:1]:
options.append((sample['cond'], sugg))
random.shuffle(options)
acceptability_frames.append(
dict(meta=meta, context=group[0]['context'], options=options))
import json
json.dump(acceptability_frames, open('acceptability_frames.json', 'w'))
def pluck(self, ind):
if cytoolz.isiterable(ind): return self.__class__(itertools.imap(flist, cytoolz.pluck(ind, self)))
else: return self.__class__(cytoolz.pluck(ind, self))
def stage_data(self) -> pd.DataFrame:
"""Stage the NHDPlus Attributes database and save to nhdplus_attrs.feather."""
r = self.get_children(self.nhd_attr_item)
titles = tlz.pluck("title", r["items"])
titles = tlz.concat(
tlz.map(tlz.partial(re.findall, "Select(.*?)Attributes"), titles))
titles = tlz.map(str.strip, titles)
main_items = dict(zip(titles, tlz.pluck("id", r["items"])))
files = {}
soil = main_items.pop("Soil")
for i, item in main_items.items():
r = self.get_children(item)
titles = tlz.pluck("title", r["items"])
titles = tlz.map(
lambda s: s.split(":")[1].strip() if ":" in s else s, titles)
child_items = dict(zip(titles, tlz.pluck("id", r["items"])))
files[i] = {t: self.get_files(c) for t, c in child_items.items()}
r = self.get_children(soil)
titles = tlz.pluck("title", r["items"])
titles = tlz.map(lambda s: s.split(":")[1].strip()
if ":" in s else s, titles)
child_items = dict(zip(titles, tlz.pluck("id", r["items"])))
stat = child_items.pop("STATSGO Soil Characteristics")
ssur = child_items.pop("SSURGO Soil Characteristics")
files["Soil"] = {t: self.get_files(c) for t, c in child_items.items()}
r = self.get_children(stat)
titles = tlz.pluck("title", r["items"])
titles = tlz.map(lambda s: s.split(":")[1].split(",")[1].strip(),
titles)
child_items = dict(zip(titles, tlz.pluck("id", r["items"])))
files["STATSGO"] = {
t: self.get_files(c)
for t, c in child_items.items()
}
r = self.get_children(ssur)
titles = tlz.pluck("title", r["items"])
titles = tlz.map(lambda s: s.split(":")[1].strip(), titles)
child_items = dict(zip(titles, tlz.pluck("id", r["items"])))
files["SSURGO"] = {
t: self.get_files(c)
for t, c in child_items.items()
}
chars = []
types = {"CAT": "local", "TOT": "upstream_acc", "ACC": "div_routing"}
for t, dd in files.items():
for d, fd in dd.items():
for f, u in fd.items():
chars.append({
"name": f,
"type": types.get(f[-3:], "other"),
"theme": t,
"description": d,
"url": u[0],
"meta": u[1],
})
char_df = pd.DataFrame(chars, dtype="category")
char_df.to_feather(self.char_feather)
return char_df
def _do_fit_step(dsk, next_token, step, cv, fields, tokens, params, Xs, ys,
fit_params, n_splits, error_score, step_fields_lk,
fit_params_lk, field_to_index, step_name, none_passthrough,
is_transform):
sub_fields, sub_inds = map(list, unzip(step_fields_lk[step_name], 2))
sub_fit_params = fit_params_lk[step_name]
if step_name in field_to_index:
# The estimator may change each call
new_fits = {}
new_Xs = {}
est_index = field_to_index[step_name]
for ids in _group_ids_by_index(est_index, tokens):
# Get the estimator for this subgroup
sub_est = params[ids[0]][est_index]
if sub_est is MISSING:
sub_est = step
# If an estimator is `None`, there's nothing to do
if sub_est is None:
nones = dict.fromkeys(ids, None)
new_fits.update(nones)
if is_transform:
if none_passthrough:
new_Xs.update(zip(ids, get(ids, Xs)))
else:
new_Xs.update(nones)
else:
# Extract the proper subset of Xs, ys
sub_Xs = get(ids, Xs)
sub_ys = get(ids, ys)
# Only subset the parameters/tokens if necessary
if sub_fields:
sub_tokens = list(pluck(sub_inds, get(ids, tokens)))
sub_params = list(pluck(sub_inds, get(ids, params)))
else:
sub_tokens = sub_params = None
if is_transform:
sub_fits, sub_Xs = do_fit_transform(dsk, next_token,
sub_est, cv, sub_fields,
sub_tokens, sub_params,
sub_Xs, sub_ys,
sub_fit_params,
n_splits, error_score)
new_Xs.update(zip(ids, sub_Xs))
new_fits.update(zip(ids, sub_fits))
else:
sub_fits = do_fit(dsk, next_token, sub_est, cv,
sub_fields, sub_tokens, sub_params,
sub_Xs, sub_ys, sub_fit_params,
n_splits, error_score)
new_fits.update(zip(ids, sub_fits))
# Extract lists of transformed Xs and fit steps
all_ids = list(range(len(Xs)))
if is_transform:
Xs = get(all_ids, new_Xs)
fits = get(all_ids, new_fits)
elif step is None:
# Nothing to do
fits = [None] * len(Xs)
if not none_passthrough:
Xs = fits
else:
# Only subset the parameters/tokens if necessary
if sub_fields:
sub_tokens = list(pluck(sub_inds, tokens))
sub_params = list(pluck(sub_inds, params))
else:
sub_tokens = sub_params = None
if is_transform:
fits, Xs = do_fit_transform(dsk, next_token, step, cv,
sub_fields, sub_tokens, sub_params,
Xs, ys, sub_fit_params, n_splits,
error_score)
else:
fits = do_fit(dsk, next_token, step, cv, sub_fields,
sub_tokens, sub_params, Xs, ys, sub_fit_params,
n_splits, error_score)
return (fits, Xs) if is_transform else (fits, None)
def _do_featureunion(dsk, next_token, est, cv, fields, tokens, params, Xs, ys,
fit_params, n_splits, error_score):
if 'transformer_list' in fields:
raise NotImplementedError("Setting FeatureUnion.transformer_list "
"in a gridsearch")
(field_to_index,
step_fields_lk) = _group_subparams(est.transformer_list, fields,
ignore=('transformer_weights'))
fit_params_lk = _group_fit_params(est.transformer_list, fit_params)
token = next_token(est)
n_samples = _do_n_samples(dsk, token, Xs, n_splits)
fit_steps = []
tr_Xs = []
for (step_name, step) in est.transformer_list:
fits, out_Xs = _do_fit_step(dsk, next_token, step, cv, fields, tokens,
params, Xs, ys, fit_params, n_splits,
error_score, step_fields_lk, fit_params_lk,
field_to_index, step_name, False, True)
fit_steps.append(fits)
tr_Xs.append(out_Xs)
# Rebuild the FeatureUnions
step_names = [n for n, _ in est.transformer_list]
if 'transformer_weights' in field_to_index:
index = field_to_index['transformer_weights']
weight_lk = {}
weight_tokens = list(pluck(index, tokens))
for i, tok in enumerate(weight_tokens):
if tok not in weight_lk:
weights = params[i][index]
if weights is MISSING:
weights = est.transformer_weights
lk = weights or {}
weight_list = [lk.get(n) for n in step_names]
weight_lk[tok] = (weights, weight_list)
weights = get(weight_tokens, weight_lk)
else:
lk = est.transformer_weights or {}
weight_list = [lk.get(n) for n in step_names]
weight_tokens = repeat(None)
weights = repeat((est.transformer_weights, weight_list))
out = []
out_append = out.append
fit_name = 'feature-union-' + token
tr_name = 'feature-union-concat-' + token
m = 0
seen = {}
for steps, Xs, wt, (w, wl), nsamp in zip(zip(*fit_steps), zip(*tr_Xs),
weight_tokens, weights, n_samples):
if (steps, wt) in seen:
out_append(seen[steps, wt])
else:
for n in range(n_splits):
dsk[(fit_name, m, n)] = (feature_union, step_names,
[None if s is None else s + (n,)
for s in steps], w)
dsk[(tr_name, m, n)] = (feature_union_concat,
[None if x is None else x + (n,)
for x in Xs], nsamp + (n,), wl)
seen[steps, wt] = m
out_append(m)
m += 1
return [(fit_name, i) for i in out], [(tr_name, i) for i in out]
def setcompare(iter1, iter2):
cntr1 = Counter(iter1)
cntr2 = Counter(iter2)
only1 = cntr1.keys() - cntr2.keys()
only2 = cntr2.keys() - cntr1.keys()
both = cntr1.keys() & cntr2.keys()
cnt1 = sum(cntr1[key] for key in only1)
cnt2 = sum(cntr2[key] for key in only2)
cnt12a = sum(cntr1[key] for key in both)
cnt12b = sum(cntr2[key] for key in both)
distinct1 = len(only1)
distinct2 = len(only2)
distinct12 = len(both)
cnt_perct = "{} ({:.0%})".format
if hasattr(iter1, "name"):
name1 = f"1 {iter1.name}"
else:
name1 = "1"
if hasattr(iter2, "name"):
name2 = f"2 {iter2.name}"
else:
name2 = "2"
display_data = [
["", f"Set {name1} only", "Intersect.", f"Set {name2} only"],
[
"Count",
cnt_perct(cnt1, cnt1 / (cnt1 + cnt12a)),
"{} | {}".format(cnt12a, cnt12b),
cnt_perct(cnt2, cnt2 / (cnt2 + cnt12b)),
],
[
"Distinct count",
cnt_perct(distinct1, distinct1 / (distinct1 + distinct12)),
distinct12,
cnt_perct(distinct2, distinct2 / (distinct2 + distinct12)),
],
[
"Examples",
format_tuples(
pluck(
0,
Counter({key: cntr1[key]
for key in only1}).most_common(5)),
cntr1,
),
format_tuples(
pluck(
0,
Counter({key: cntr1[key] + cntr2[key]
for key in both}).most_common(5),
),
cntr1,
cntr2,
),
format_tuples(
pluck(
0,
Counter({key: cntr2[key]
for key in only2}).most_common(5)),
cntr2,
),
],
]
make_table(display_data)
table = apply_theme("basic_both")
for x, y in product([0, 1, 2], [1, 2, 3]):
set_cell_style(x, y, align="center")
return table
def generate_timeline(time_range, save_to_db=True, tweets=None, skip_fields=None, size=40, sideline_turns=5,
time_bucket_size=30, location_depth=1, time_zone=None, twitter_api=None,
min_length=50, exclude_replies=True, informational_only=False, exclude_retweets=False,
post_update=True, retweet_tweeps=True, skip_sensitive_content=True,
retweeted_only=False, post_to_twitter=False, update_users=True, shout_score=0.5,
n_candidates=None, target_entropy=0.99):
if n_candidates is not None and n_candidates <= 0:
n_candidates = None
# generate filtered set
characterizer = Characterizer()
if not time_zone:
time_zone = pytz.timezone(settings.TIME_ZONE)
locations = list(Location.objects.filter(depth=location_depth))
location_pks = [loc.pk for loc in locations]
location_populations = dict([(loc.pk, loc.population) for loc in locations])
if not time_zone:
time_zone = pytz.timezone(settings.TIME_ZONE)
if tweets is None:
tweets = _timeline_queryset(time_range, location_depth=location_depth, min_length=min_length, exclude_replies=exclude_replies,
informational_only=informational_only, exclude_retweets=exclude_retweets,
skip_sensitive_content=skip_sensitive_content, retweeted_only=retweeted_only)
feature_keys = ['pk', 'text', 'favourite_count', 'retweet_count',
'user__screen_name', 'datetime', 'user__name', 'user__profile_image_url', 'internal_id',
'user__internal_id', 'user__friends_count', 'user__followers_count',
'user__statuses_count', 'characterization__count_rts',
'characterization__manual_rt', 'characterization__is_reply']
if location_depth > 0:
feature_keys.append('user__characterization__location_depth_{0}'.format(location_depth))
tweet_features = list(tweets.values(*feature_keys))
if not tweet_features or len(tweet_features) < size:
print('ERROR, not enough tweets', len(tweet_features))
return None
if location_depth > 0:
pick_strategy, approve_fn = TimelineFilter.select_tweet_and_sideline(TimelineFilter.select_popular_bucketed, location_pks, turns=sideline_turns)
start_strategy = pick_strategy
else:
pick_strategy = TimelineFilter.select_popular_bucketed
start_strategy = TimelineFilter.starting_tweet
approve_fn = None
if skip_fields is None:
skip_fields = {'geography'}
else:
skip_fields.add('geography')
generator = TimelineFilter(characterizer,
skip_fields=skip_fields,
min_date=time_range[0],
max_entropy_percentile=100.0,
time_bucket_size=time_bucket_size,
start_strategy=start_strategy,
pick_strategy=pick_strategy,
approve_tweet_fn=approve_fn,
n_candidates=n_candidates,
target_entropy=target_entropy)
for t in tweet_features:
if location_depth > 0:
t['geography'] = int(t['user__characterization__location_depth_{0}'.format(location_depth)])
t['popularity'] = 2.0 * t['characterization__count_rts'] + t['retweet_count'] + 0.5 * t['favourite_count']
generator.prepare_tweet(t)
tweet_features = filter(lambda x: x['__shout_score__'] < shout_score, tweet_features)
tweet_features = sorted(tweet_features, key=lambda x: x['buckets']['popularity'], reverse=True)
print('N features', len(tweet_features))
for i in range(0, min([size, len(tweet_features)])):
generator(tweet_features)
# estimate data
tweet_pks = list(pluck('pk', generator))
tl_tweets = Tweet.objects.in_bulk(tweet_pks)
to_serialize = [tl_tweets[pk] for pk in tweet_pks]
popularity = list(pluck('popularity', generator))
html = []
media = []
sources = []
rted_user_pks = set()
for t in to_serialize:
if t.media.exists():
media.append(json.loads(serializers.serialize("json", t.media.all())))
else:
media.append(None)
if t.rt_instance_tweet.exists():
rt = t.rt_instance_tweet.all()[0].source_tweet
rt_serialized = json.loads(serializers.serialize("json", [rt]))[0]
rted_user_pks.add(rt_serialized['fields']['user'])
# NOTE: this user is not updated. need to update it at some point.
#rt_serialized['fields']['user'] = json.loads(serializers.serialize("json", [User.objects.get(pk=pk)]))[0]
html.append(rt.beautify_html())
sources.append(rt_serialized)
else:
html.append(t.beautify_html())
sources.append(None)
if not twitter_api or not update_users:
print('not updating users')
users = json.loads(serializers.serialize("json", [t.user for t in to_serialize]))
else:
print('updating user profiles')
user_pks = [t.user_id for t in to_serialize]
user_pks.extend(rted_user_pks)
user_ids = User.objects.filter(pk__in=user_pks).values_list('internal_id', flat=True)
tl_user_data = twitter_api.lookup_users(user_ids=user_ids)
for tl_user in tl_user_data:
user, created = User.import_json(valfilter(lambda x: x, tl_user._json))
user_dict = User.objects.in_bulk([t.user_id for t in to_serialize])
users = json.loads(serializers.serialize("json", [user_dict[t.user_id] for t in to_serialize]))
for t, source in zip(to_serialize, sources):
if source is not None:
source['fields']['user'] = json.loads(serializers.serialize("json", [User.objects.get(pk=source['fields']['user'])]))[0]
weight = []
if location_depth > 0:
max_population = 1.0 * max([loc.population for loc in locations])
else:
max_population = None
for t in generator:
if location_depth > 0:
w = gmean([t['popularity'] + 1.0,
max_population / location_populations[t['geography']],
t['user__followers_count'] + 1.0,
t['user__friends_count'] + 1.0])
else:
w = gmean([t['popularity'] + 1.0,
t['user__followers_count'] + 1.0,
t['user__friends_count'] + 1.0])
weight.append(w)
hour_norm = 1.0 / (60.0 * 60.0)
kwargs = {
'user': users,
'html': html,
'popularity': popularity,
'weight': weight,
'recency': [(time_range[1] - t.datetime).total_seconds() * hour_norm for t in to_serialize],
'media': media,
'source_user': sources
}
if location_depth > 0:
kwargs['geolocation'] = list(pluck('geography', generator))
if time_zone:
kwargs['datetime'] = [t.datetime.astimezone(time_zone).isoformat() for t in to_serialize]
# metadata
metadata = {
'locations': json.loads(serializers.serialize("json", locations)),
}
if not save_to_db:
return {'tweets': to_serialize, 'metadata': kwargs}
# save filtered set
tl = Timeline.from_tweets(to_serialize, metadata, **kwargs)
if twitter_api and to_serialize:
tl_url = u'http://auroratwittera.cl{0}'.format(reverse('timelines:timeline-home'))
if len(users) > 3:
top_users = random.sample([u'@{0}'.format(t['fields']['screen_name']) for t in users], 3)
else:
top_users = ['@{0}'.format(t['fields']['screen_name']) for t in users]
status_1 = 'Nvo. resumen informativo en {0} con tweets de {1}'.format(tl_url, u' '.join(top_users))
print(repr(status_1))
if post_update and post_to_twitter:
try:
twitter_api.update_status(status_1)
time.sleep(30)
except tweepy.error.TweepError:
pass
if rted_user_pks:
try:
top_rted = random.sample([u'@{0}'.format(t['fields']['user']['fields']['screen_name']) for t in sources if t], min([3, len(rted_user_pks)]))
status_2 = 'Nvo. resumen informativo en {0} con RTs de {1}'.format(tl_url, u' '.join(top_rted))
print(repr(status_2))
if post_update and post_to_twitter:
try:
twitter_api.update_status(status_2)
time.sleep(30)
except tweepy.error.TweepError:
pass
except ValueError:
pass
if retweet_tweeps:
for w, t in sorted(zip(kwargs['weight'], to_serialize), reverse=True)[0:size - 1]:
if not post_to_twitter:
print('retweet', t.internal_id)
else:
try:
print('retweet', t.internal_id)
twitter_api.retweet(t.internal_id)
time.sleep(60)
except tweepy.error.TweepError as e:
print([e])
return tl
