def main():
format = ('%(asctime)s %(processName)s:%(name)s:'
'%(funcName)s:%(lineno)s:%(levelname)s: %(message)s')
logging.basicConfig(level=logging.INFO, format=format)
directory = os.path.join(config.PROJECT_DIRECTORY, 'scraped_data')
steem = MPSteem(nodes=config.NODES)
current_datetime = '2018-02-01'
crossval_filename = os.path.join(directory,
'xval_{}.gz'.format(current_datetime))
post_frame = tpgd.load_or_scrape_training_data(
steem,
directory,
current_datetime=current_datetime,
days=3,
offset_days=0)
regressor_kwargs = dict(n_estimators=256,
max_leaf_nodes=4096,
max_features=0.2,
n_jobs=-1,
verbose=1,
random_state=42,
oob_score=True)
# topic_kwargs = dict(num_topics=32, no_below=5, no_above=0.1)
doc2vec_kwargs = dict(size=128, epochs=32)
post_frame = tppp.load_or_preprocess(post_frame,
crossval_filename,
ncores=4,
chunksize=1000,
min_en_prob=0.9)
param_grid = {
'feature_generation__topic_model__no_above': [0.33],
#'regressor__max_leaf_nodes': [500, 1000],
# 'regressor__max_features': [0.1, 0.2, 0.3]
}
# tpmo.cross_validate(post_frame, param_grid, topic_kwargs=topic_kwargs,
# regressor_kwargs=regressor_kwargs, n_iter=None,
# n_jobs=4, targets=['reward'])
pipe, test_frame = tpmo.train_test_pipeline(
post_frame,
#topic_kwargs=topic_kwargs,
regressor_kwargs=regressor_kwargs,
doc2vec_kwargs=doc2vec_kwargs,
targets=['reward', 'votes'])
# topic_model = pipe.named_steps['feature_generation'].transformer_list[1][1]
# logging.getLogger().info(topic_model.print_topics(n_best=None))
tpmo.find_truffles(test_frame, pipe)
def test_find_truffles():
posts = create_n_random_posts(300)
post_frame = pd.DataFrame(posts)
regressor_kwargs = dict(n_estimators=20,
max_leaf_nodes=100,
max_features=0.1,
n_jobs=-1,
verbose=1,
random_state=42)
topic_kwargs = dict(num_topics=50, no_below=5, no_above=0.7)
post_frame = tppp.preprocess(post_frame, ncores=4, chunksize=50)
pipeline = tpmo.train_pipeline(post_frame,
topic_kwargs=topic_kwargs,
regressor_kwargs=regressor_kwargs)
posts = create_n_random_posts(50)
post_frame = pd.DataFrame(posts)
post_frame = tppp.preprocess(post_frame, ncores=4, chunksize=50)
truffles = tpmo.find_truffles(post_frame, pipeline, account='aa')
assert truffles.iloc[0].rank_score == truffles.rank_score.max()
def test_find_truffles_with_real_data(steem):
df = tpbg.scrape_hour_data(steem, stop_after=20)
df = tppp.preprocess(df)
sorted = tpmo.find_truffles(df, MockPipeline())
assert sorted.rank_score.iloc[0] == sorted.rank_score.max()
def main():
format = ('%(asctime)s %(processName)s:%(name)s:'
'%(funcName)s:%(lineno)s:%(levelname)s: %(message)s')
logging.basicConfig(level=logging.INFO, format=format)
directory = os.path.join(config.PROJECT_DIRECTORY, 'scraped_data')
crossval_filename = os.path.join(directory, 'xval_first_proto.gz')
post_frame = pd.read_pickle('../scraped_data/first_post_set.gz')
regressor_kwargs = dict(n_estimators=256,
max_leaf_nodes=1024,
max_features=0.3,
n_jobs=-1,
verbose=1,
random_state=42)
doc2vec_kwargs = dict(size=32, epochs=20)
post_frame['votes'] = post_frame.reward.astype(int).astype(float)
post_frame = tppp.load_or_preprocess(post_frame,
crossval_filename,
ncores=4,
chunksize=1000,
min_en_prob=0.9)
# param_grid = {
# #'feature_generation__topic_model__no_above':[0.05, 0.1, 0.2, 0.33],
# #'feature_generation__topic_model__num_topics':[50, 100, 200],
# 'regressor__max_leaf_nodes': [50, 100, 200]
# # 'regressor__max_features': [0.1, 0.2, 0.3, 0.66]
# }
#
# tpmo.cross_validate(post_frame, param_grid, topic_kwargs=topic_kwargs,
# regressor_kwargs=regressor_kwargs, n_iter=None,
# n_jobs=4, targets=['reward'])
pipe, test_frame = tpmo.train_test_pipeline(
post_frame,
doc2vec_kwargs=doc2vec_kwargs,
regressor_kwargs=regressor_kwargs,
targets=['reward', 'votes'])
tpmo.find_truffles(test_frame, pipe)
def main():
logging.basicConfig(level=logging.INFO)
author, permalink, current_datetime = parse_args()
if current_datetime is None:
current_datetime = pd.datetime.utcnow()
else:
current_datetime = pd.to_datetime(current_datetime)
model_directoy = os.path.join(config.PROJECT_DIRECTORY, 'trained_models')
pipeline = tpmo.load_or_train_pipeline(None, model_directoy,
current_datetime)
steem = MPSteem(nodes=config.NODES, no_broadcast=True)
posts = tpgd.get_post_data([(author, permalink)], steem, {})
posts = pd.DataFrame(posts)
post_frame = tppp.preprocess(posts)
tpmo.find_truffles(post_frame, pipeline)
def execute_call(comment_authors_and_permalinks, poster, pipeline,
topN_permalink, max_comments, overview_permalink):
"""Executes the pig on duty call"""
ncomments = len(comment_authors_and_permalinks)
logger.info('Found {} comments mentioning {}'.format(
ncomments, poster.account))
if ncomments > max_comments:
logger.info('To many comments, reducing to {}'.format(max_comments))
comment_authors_and_permalinks = comment_authors_and_permalinks[:
max_comments]
posts = tpco.get_parent_posts(comment_authors_and_permalinks, poster.steem)
initial_frame = pd.DataFrame(posts)
post_frame = initial_frame.copy()
post_frame = tppp.preprocess(post_frame, ncores=4)
if len(post_frame):
truffle_frame = tpmo.find_truffles(post_frame,
pipeline,
k=0,
account='',
add_rank_score=False)
truffle_frame['passed'] = True
else:
truffle_frame = pd.DataFrame()
filtered_posts = initial_frame[~initial_frame.index.isin(truffle_frame.
index)]
filtered_posts['passed'] = False
combined = pd.concat([truffle_frame, filtered_posts], axis=0)
topN_link = 'https://steemit.com/@{author}/{permalink}'.format(
author=poster.account, permalink=topN_permalink)
tpoc.post_on_call(combined,
poster=poster,
topN_link=topN_link,
overview_permalink=overview_permalink)
def main():
"""Main loop started from command line"""
no_broadcast, current_datetime = parse_args()
if current_datetime is None:
current_datetime = pd.datetime.utcnow()
else:
current_datetime = pd.to_datetime(current_datetime)
data_directory = os.path.join(config.PROJECT_DIRECTORY, 'scraped_data')
model_directoy = os.path.join(config.PROJECT_DIRECTORY, 'trained_models')
log_directory = os.path.join(config.PROJECT_DIRECTORY, 'logs')
configure_logging(log_directory, current_datetime)
logger.info('STARTING main script at {}'.format(current_datetime))
if no_broadcast:
logger.info('Run without broadcasting.')
else:
logger.info('ATTENTION I WILL BROADCAST TO STEEMIT!!!')
time.sleep(2)
steem = MPSteem(nodes=config.NODES, no_broadcast=no_broadcast)
# hack to allow for payments, because of https://github.com/steemit/steem-python/issues/191
noapisteem = MPSteem(nodes=config.NODES[1:], no_broadcast=no_broadcast)
# To post stuff
account = config.ACCOUNT
poster = Poster(account=account, steem=noapisteem)
tppd.create_wallet(steem,
config.PASSWORD,
posting_key=config.POSTING_KEY,
active_key=config.ACTIVE_KEY)
logger.info('Paying out investors')
tpde.pay_delegates(
account=account,
steem=noapisteem, # use a steem instance without api.steem!
current_datetime=current_datetime)
if not tpmo.model_exists(current_datetime, model_directoy):
post_frame = load_and_preprocess_2_frames(
log_directory=log_directory,
current_datetime=current_datetime,
steem=steem,
noapisteem=noapisteem,
data_directory=data_directory)
logger.info('Garbage collecting')
gc.collect()
else:
post_frame = None
regressor_kwargs = dict(n_estimators=256,
max_leaf_nodes=5000,
max_features=0.2,
n_jobs=-1,
verbose=1,
random_state=42)
topic_kwargs = dict(num_topics=128,
no_below=7,
no_above=0.1,
ngrams=(1, 2),
keep_n=333000)
if post_frame is not None and len(post_frame) > MAX_DOCUMENTS:
logger.info('Frame has {} Documents, too many, '
'reducing to {}'.format(len(post_frame), MAX_DOCUMENTS))
post_frame.sort_values('created', inplace=True, ascending=False)
train_frame = post_frame.iloc[:MAX_DOCUMENTS, :]
else:
train_frame = post_frame
pipeline = tpmo.load_or_train_pipeline(
train_frame,
model_directoy,
current_datetime,
regressor_kwargs=regressor_kwargs,
topic_kwargs=topic_kwargs,
targets=['adjusted_reward', 'adjusted_votes'])
tpmo.log_pipeline_info(pipeline=pipeline)
overview_permalink = tppw.return_overview_permalink_if_exists(
account=account, current_datetime=current_datetime, steem=steem)
if not overview_permalink:
if post_frame is None:
logger.info('Need to reaload data for weekly overview')
post_frame = load_and_preprocess_2_frames(
log_directory=log_directory,
current_datetime=current_datetime,
steem=steem,
noapisteem=noapisteem,
data_directory=data_directory)
logger.info('I want to post my weekly overview')
overview_permalink = tppw.post_weakly_update(
pipeline=pipeline,
post_frame=post_frame,
poster=poster,
current_datetime=current_datetime)
logger.info('Garbage collecting')
del post_frame
gc.collect()
prediction_frame = tpgd.scrape_hour_data(steem=steem,
current_datetime=current_datetime,
ncores=32,
offset_hours=2)
prediction_frame = tppp.preprocess(prediction_frame, ncores=8)
sorted_frame = tpmo.find_truffles(prediction_frame,
pipeline,
account=account)
permalink = tppd.post_topN_list(sorted_frame,
poster=poster,
current_datetime=current_datetime,
overview_permalink=overview_permalink)
tppd.comment_on_own_top_list(sorted_frame,
poster=poster,
topN_permalink=permalink)
tppd.vote_and_comment_on_topK(sorted_frame,
poster=poster,
topN_permalink=permalink,
overview_permalink=overview_permalink)
logger.info('Computing the top trending without bidbots')
logger.info('Searching for bid bots and bought votes')
min_datetime = sorted_frame.created.min()
max_datetime = sorted_frame.created.max() + pd.Timedelta(days=1)
upvote_payments, bots = tpad.get_upvote_payments_to_bots(
steem=noapisteem, min_datetime=min_datetime, max_datetime=max_datetime)
logger.info('Adjusting votes and reward')
sorted_frame = tppp.compute_bidbot_correction(
post_frame=sorted_frame, upvote_payments=upvote_payments)
tt0b.create_trending_post(sorted_frame,
upvote_payments=upvote_payments,
poster=poster,
topN_permalink=permalink,
overview_permalink=overview_permalink,
current_datetime=current_datetime,
bots=bots)
logger.info('Done with normal duty, answering manual calls!')
tfod.call_a_pig(poster=poster,
pipeline=pipeline,
topN_permalink=permalink,
current_datetime=current_datetime,
offset_hours=2,
hours=24,
overview_permalink=overview_permalink)
logger.info('Cleaning up after myself')
tfut.clean_up_directory(model_directoy, keep_last=3)
tfut.clean_up_directory(data_directory, keep_last=25)
tfut.clean_up_directory(log_directory, keep_last=14)
logger.info('Preloading -8 days for later training')
tpgd.load_or_scrape_training_data(steem,
data_directory,
current_datetime=current_datetime,
days=1,
offset_days=8,
ncores=32)
logger.info('DONE at {}'.format(current_datetime))