def _fetch_member_friends(self, user_subset):
if self.member_friends is not None:
return
log_method_begin()
members = [
member for member in self.members if member['id'] in user_subset
]
print('{} users to fetch'.format(len(members)))
pool_results = []
with vk_api.VkRequestsPool(self.vk_session) as pool:
for member in members:
pool_results.append(
(member['id'],
pool.method('friends.get', {
'user_id': member['id'],
'fields': 'photo'
})))
self.member_friends = defaultdict(list)
for member_id, friend_request in pool_results:
if friend_request.ok:
for friend in friend_request.result['items']:
if friend['id'] not in user_subset:
friend['is_member'] = False
self.member_friends[member_id].append(friend)
self._compress_users()
self._save_pickle('raw_users_data.member_friends', self.member_friends)
log_method_end()
def fit(self, post_subset=None):
df = self.action_data.get_all()
if post_subset is not None:
df = df[df['post_id'].isin(post_subset)]
log_method_begin()
x_df = df.drop(
['user_id', 'post_id', 'is_member', 'is_liked', 'is_reposted'],
axis=1)
self.like_model.fit(x_df, df['is_liked'])
self.repost_model.fit(x_df, df['is_reposted'])
self.is_fitted = True
log_method_end()
def get_true(self, subset=None):
print('GroupPredict.get_true for group {}'.format(self.group_id))
log_method_begin()
direct_likes_count = Counter()
direct_reposts_count = Counter()
non_direct_likes_count = Counter()
non_direct_reposts_count = Counter()
for post in tqdm(self.raw_wall_data.posts):
post_id = post['id']
if subset is not None and post_id not in subset:
continue
for user_id in post['likes']['user_ids']:
user = self.raw_users_data.find_user(user_id)
if user is None:
continue
if user['is_member']:
direct_likes_count[post_id] += 1
else:
non_direct_likes_count[post_id] += 1
for user_id in post['reposts']['user_ids']:
user = self.raw_users_data.find_user(user_id)
if user is None:
continue
if user['is_member']:
direct_reposts_count[post_id] += 1
else:
non_direct_reposts_count[post_id] += 1
post_ids = list(direct_likes_count.keys()
| direct_reposts_count.keys()
| non_direct_likes_count.keys()
| non_direct_reposts_count.keys())
rows = []
for post_id in post_ids:
rows.append([
direct_likes_count[post_id], direct_reposts_count[post_id],
non_direct_likes_count[post_id],
non_direct_reposts_count[post_id]
])
result = pd.DataFrame(rows,
index=post_ids,
columns=[
'direct_likes_count', 'direct_reposts_count',
'non_direct_likes_count',
'non_direct_reposts_count'
])
log_method_end()
return result
def _fetch_members(self):
if self.members is not None:
return
log_method_begin()
self.members = self.vk_tools.get_all('groups.getMembers', 1000, {
'group_id': self.group_id,
'fields': self.member_fields
})['items']
print('{} members'.format(len(self.members)))
for member in self.members:
member['is_member'] = True
log_method_end()
def fit(self):
log_method_begin()
print("{} members, {} posts".format(len(self.raw_users_data.members),
len(self.raw_wall_data.posts)))
rows = []
friend_post_pairs = set()
for user in tqdm(self.raw_users_data.members,
'ActionData.get_all: for members'):
if 'groups' not in user:
continue
for post in self.raw_wall_data.posts:
is_liked = user['id'] in post['likes']['user_ids']
is_reposted = user['id'] in post['reposts']['user_ids']
if is_reposted:
for friend in self.raw_users_data.member_friends[
user['id']]:
if 'groups' not in friend:
continue
friend_post_pair = (friend['id'], post['id'])
if friend_post_pair not in friend_post_pairs:
friend_is_liked = friend['id'] in post['likes'][
'user_ids']
friend_is_reposted = friend['id'] in post[
'reposts']['user_ids']
rows.append(
self.get_row(friend, post, False,
friend_is_liked,
friend_is_reposted))
friend_post_pairs.add(friend_post_pair)
rows.append(
self.get_row(user, post, True, is_liked, is_reposted))
result = pd.DataFrame(rows, columns=self.get_labels())
self.table = result
print("{} rows".format(len(result)))
print("{} liked, {} reposted".format(sum(result['is_liked']),
sum(result['is_reposted'])))
print("{} liked, {} reposted by members".format(
sum(result[result['is_member']]['is_liked']),
sum(result[result['is_member']]['is_reposted'])))
log_method_end()
return result
def predict(self, post_subset=None):
df = self.action_data.get_all()
if post_subset is not None:
df = df[df['post_id'].isin(post_subset)]
log_method_begin()
x_df = df.drop(
['user_id', 'post_id', 'is_member', 'is_liked', 'is_reposted'],
axis=1)
pred = [
df['user_id'], df['post_id'], df['is_member'],
self.like_model.predict(x_df),
self.repost_model.predict(x_df)
]
result = pd.DataFrame(np.array(pred).T,
columns=[
'user_id', 'post_id', 'is_member',
'is_liked', 'is_reposted'
])
log_method_end()
return result
def predict(self, post_subset=None):
log_method_begin()
direct_likes_count = Counter()
direct_reposts_count = Counter()
non_direct_likes_count = Counter()
non_direct_reposts_count = Counter()
pred_df = self.predict_action_model.predict(post_subset)
for i, row in pred_df.iterrows():
if row['is_liked']:
if row['is_member']:
direct_likes_count[row['post_id']] += 1
else:
non_direct_likes_count[row['post_id']] += 1
if row['is_reposted']:
if row['is_member']:
direct_reposts_count[row['post_id']] += 1
else:
non_direct_reposts_count[row['post_id']] += 1
post_ids = list(direct_likes_count.keys()
| direct_reposts_count.keys()
| non_direct_likes_count.keys()
| non_direct_reposts_count.keys())
rows = []
for post_id in post_ids:
rows.append([
direct_likes_count[post_id], direct_reposts_count[post_id],
non_direct_likes_count[post_id],
non_direct_reposts_count[post_id]
])
result = pd.DataFrame(rows,
index=post_ids,
columns=[
'direct_likes_count', 'direct_reposts_count',
'non_direct_likes_count',
'non_direct_reposts_count'
])
log_method_end()
return result
def _fetch_activity(self):
log_method_begin()
print('{} posts to fetch'.format(len(self.posts)))
pool_results = []
with vk_api.VkRequestsPool(self.vk_session) as pool:
for post in self.posts:
likes = pool.method(
'likes.getList',
{'item_id': post['id'], 'owner_id': -self.group_id, 'type': 'post', 'count': 1000, 'filter': 'likes'}
)
reposts = pool.method(
'likes.getList',
{'item_id': post['id'], 'owner_id': -self.group_id, 'type': 'post', 'count': 1000, 'filter': 'copies'}
)
pool_results.append((post, likes, reposts))
for post, likes, reposts in pool_results:
if 'likes' not in post:
post['likes'] = dict()
if likes.ok:
likes = likes.result['items']
post['likes']['user_ids'] = set(likes)
else:
post['likes']['user_ids'] = set()
if 'reposts' not in post:
post['reposts'] = dict()
if reposts.ok:
reposts = reposts.result['items']
post['reposts']['user_ids'] = set(reposts)
else:
post['reposts']['user_ids'] = set()
log_method_end()
def fit(self):
log_method_begin()
self.lda_maker = LdaMaker(self._get_corpora_for_lda(), self.num_topics)
log_method_end()
def _fetch_wall(self):
log_method_begin()
self.posts = self.vk_tools.get_all('wall.get', 100, {'owner_id': -self.group_id, 'extended': 1})['items']
print('{} posts'.format(len(self.posts)))
log_method_end()
def _fetch_groups(self, user_subset):
log_method_begin()
all_users = [
user for user in self.get_all_users() if user['id'] in user_subset
]
print('{} users to fetch'.format(len(all_users)))
all_users_processing_step = 1000
fetch_start = time.time()
for i in range(0, len(all_users), all_users_processing_step):
print('Fetching from {} to {}...'.format(
i, i + all_users_processing_step))
users = all_users[i:i + all_users_processing_step]
if time.time() - fetch_start > 2 * 60 * 60:
print('Cooldown for 30 minutes')
time.sleep(30 * 60)
fetch_start = time.time()
do_fetch = True
last_error_time = -1
while do_fetch:
try:
pool_results = []
with vk_api.VkRequestsPool(self.vk_session) as pool:
for user in users:
if 'groups' not in user:
pool_results.append(
(user,
pool.method(
'groups.get', {
'user_id': user['id'],
'count': 1000,
'extended': 1,
'fields': self.group_fields
})))
do_fetch = False
self.unmark_fetch_groups_error()
except Exception as e:
print('Can\'t fetch groups because of', e)
traceback.print_exc()
if time.time() - last_error_time < 120:
print(
'Can\'t do anything, exit. Restart will reuse fetched users'
)
do_fetch = False
self.mark_fetch_groups_error()
else:
print('Trying again in 1 minute')
time.sleep(60)
last_error_time = time.time()
finally:
for user, groups_request in pool_results:
if groups_request.ok and groups_request.ready:
user['groups'] = []
for group in groups_request.result['items']:
if 'description' in group:
user['groups'].append(
{'description': group['description']})
self._save_pickle('raw_users_data.members', self.members)
self._save_pickle('raw_users_data.member_friends',
self.member_friends)
log_method_end()