def main():
path = './'
n_runs = 1
simulation_periods = [['2019-02-01', '2019-02-15'],
['2019-02-08', '2019-02-22'],
['2019-02-15', '2019-03-01'],
['2019-02-22', '2019-03-01']]
#files are in weekly subsets, e.g. venezuela_v2_extracted_twitter_2019-02-01_2019-02-08.json
all_files = glob.glob(path + 'venezuela_v2_extracted*.json')
#extract dates and platforms from file names
date_re = '(20\d\d-\d\d-\d\d)_(20\d\d-\d\d-\d\d)'
dates = [re.search(date_re, fn) for fn in all_files]
start_dates = [d.group(1) for d in dates]
end_dates = [d.group(2) for d in dates]
platforms = [re.search('twitter|youtube', fn).group(0) for fn in all_files]
#create data frame with files, dates, and platforms
fn_df = pd.DataFrame({
'fn': all_files,
'start': start_dates,
'end': end_dates,
'platform': platforms
})
fn_df['start'] = pd.to_datetime(fn_df['start'])
fn_df['end'] = pd.to_datetime(fn_df['end'])
fn_df = fn_df.sort_values('start')
#loop over simulation periods
for sim_period in simulation_periods:
#start and end time of the simulation
start = pd.to_datetime(sim_period[0])
end = pd.to_datetime(sim_period[1])
#select files to sample from based on dates (same length as simulation period, just before simulation period)
sim = fn_df[(fn_df['start'] >= start) & (fn_df['start'] < end)]
hist = fn_df[(fn_df['start'] < start) & (fn_df['start'] >= start -
(end - start))]
previous = fn_df[(fn_df['start'] < start - (end - start))]
print(start, end)
print('Historical Data to Sample From')
print(hist)
print('Prior Data to Track Users From')
print(previous)
previous_history_data = list(previous['fn'].values)
history_data = list(hist['fn'].values)
#load data
hist = []
for data in history_data:
hist.append(
ss.load_data(data, ignore_first_line=False, verbose=False))
hist = pd.concat(hist)
hist = hist.sort_values('nodeTime')
previous_hist = []
for data in previous_history_data:
previous_hist.append(
ss.load_data(data, ignore_first_line=False, verbose=False))
previous_hist = pd.concat(previous_hist)
previous_hist = previous_hist[['nodeUserID',
'informationID']].drop_duplicates()
#multiple runs of the baseline sampling
for i in range(n_runs):
dfs = []
# for each platform and information ID
for (plat,
info), grp in hist.groupby(['platform', 'informationID']):
print(plat, info)
starting = time.time()
sampled_df = sample_from_historical_data(
grp,
info,
plat,
hist['nodeTime'].min(),
hist['nodeTime'].max(),
start,
end,
previous_hist=previous_hist,
new_users=True)
ending = time.time()
elapsed = (ending - starting) / 60.0
print(f'Time elapsed: {elapsed} minutes')
dfs.append(sampled_df)
baseline = pd.concat(dfs).reset_index(drop=True)
#save generated baseline
start_str = start.strftime('%Y-%m-%d')
end_str = end.strftime('%Y-%m-%d')
baseline.to_json(f'baseline_{start_str}_{end_str}_{i}.json',
orient='records',
lines=True)
import socialsim as ss
# Load the configuration file
config = 'data/cp4_configuration.json'
config = ss.load_config(config)
# Get metadata
metadata = ss.MetaData()
# Instantiate the task runner with the specified ground truth
ground_truth_filepath = 'data/test_dataset.json'
ground_truth = ss.load_data(ground_truth_filepath,
ignore_first_line=True,
verbose=False)
eval_runner = ss.EvaluationRunner(ground_truth, config, metadata=metadata)
# Evaluate a series of submissions that contain submission metadata as the first line of the submission file\
submission_filepaths = ['data/test_dataset.json']
for simulation_filepath in submission_filepaths:
# Run measurements and metrics on the simulation data
results, logs = eval_runner(simulation_filepath,
verbose=True,
submission_meta=True)
import socialsim as ss
# Load the example dataset
dataset = 'data/test_dataset.txt'
dataset = ss.load_data(dataset)
# Subset the dataset to a particular platform
dataset = dataset[dataset['platform'] == 'twitter'].head(n=2000)
# Load the configuration file
config = 'cp1_configuration.json'
config = ss.load_config(config)
# Subset the configuration for the given task
config = config['twitter']['social_structure']
# Define the measurement object
social_structure_measurements = ss.SocialStructureMeasurements(
dataset, config, None, 'twitter')
# Run all measurements in the config file
results = social_structure_measurements.run(verbose=True)
def check_records(submission_filepath, nodelist, simulation_period):
errors, warnings = [], []
try:
# test that submission file can be loaded
subm = ss.load_data(submission_filepath,
ignore_first_line=True,
verbose=False)
loaded = True
except Exception as e:
errors.append('Submission could not be loaded: ' + str(e))
loaded = False
if loaded:
# platform tests
valid_items = VALID_OPTIONS[challenge]['platforms']
subm_items = set(subm['platform'].unique())
platform_errors, platform_warnings = check_all_present(
valid_items, subm_items, 'platforms')
errors.extend(platform_errors)
warnings.extend(platform_warnings)
if nodelist is not None:
# informationID tests
valid_items = VALID_OPTIONS[challenge]['informationID']
subm_items = set(subm['informationID'].unique())
informationID_errors, informationID_warnings = check_all_present(
valid_items, subm_items, 'informationIDs')
errors.extend(informationID_errors)
warnings.extend(informationID_warnings)
# test that there are no NaN items in required event details
for c in [
'informationID', 'nodeTime', 'nodeID', 'parentID', 'rootID',
'platform', 'actionType', 'nodeUserID'
]:
if len(subm[c]) != len(subm[c].dropna()):
errors.append(f'{c} can not be NaN values.')
print(c)
print(subm[c].astype(str).unique())
# check for empty user-user network
parentID_nodeID_overlap = set(subm['parentID']).intersection(
set(subm['nodeID']))
if len(parentID_nodeID_overlap) == 0:
warnings.append(
'There is no overlap between nodeID values and parentID values -- the user-to-user network created from this submission will be empty.'
)
# check that nodeTimes fall within simulation window
try:
simulation_window = VALID_OPTIONS[challenge]['simulation_windows'][
simulation_period]
minday = f'2019-{simulation_window[0]}'
maxday = f'2019-{simulation_window[1]} 23:59'
maxday_str = maxday.split(' ')[0]
subm['nodeTime'] = pd.to_datetime(subm['nodeTime']).astype(str)
subm_minday, subm_maxday = subm['nodeTime'].min(
), subm['nodeTime'].max()
if subm_maxday <= minday:
errors.append(
f'There is no data within the simulation period, all nodeTime values occur before the simulation period ({minday} - {maxday_str}).\n\tSubmission nodeTime values -- Min: {subm_minday} Max: {subm_maxday}'
)
elif subm_minday > maxday:
errors.append(
f'There is no data within the simulation period, all nodeTime values occur after the simulation period ({minday} - {maxday_str}).\n\tSubmission nodeTime values -- Min: {subm_minday} Max: {subm_maxday}'
)
elif subm_minday < minday:
warnings.append(
f'Some events occur before the simulation period, earliest nodeTime value is {subm_minday}'
)
elif subm_maxday > maxday:
warnings.append(
f'Some events occur after the simulation period, latest nodeTime value is {subm_maxday}'
)
except Exception as e:
warnings.append(
'Could not validate nodeTimes occur within the simulation period: '
+ str(e))
# actionType tests
valid_items = VALID_OPTIONS[challenge]['actiontypes']
subm_items = set(subm['actionType'].unique())
platform_errors, platform_warnings = check_all_present(
valid_items, subm_items, 'actionType')
errors.extend(platform_errors)
warnings.extend(platform_warnings)
result = ''
if len(errors) > 0:
result = result + 'ERRORS:\n\t'
result = result + '\n\n\t'.join(errors) + '\n\n'
if len(warnings) > 0:
result = result + 'WARNINGS:\n\t'
result = result + '\n\n\t'.join(warnings)
if result == '': result = 'success'
return result
import socialsim as ss # Load the simulation data simulation = 'data/test_dataset.txt' simulation = ss.load_data(simulation) # Load the ground truth data ground_truth = 'data/test_dataset.txt' ground_truth = ss.load_data(ground_truth) # Load the configuration file config = 'data/cp2_configuration.json' config = ss.load_config(config) # Get metadata metadata = ss.MetaData(community_directory='data/communities/') # Instantiate the task runner task_runner = ss.TaskRunner(ground_truth, config, metadata=metadata, test=True) # Run measurements and metrics on the simulation data results = task_runner(simulation, verbose=True)
from pprint import pprint import socialsim as ss # Load the simulation data simulation = 'data/test_dataset.json' simulation = ss.load_data(simulation, ignore_first_line=True, verbose=False) # Load the ground truth data ground_truth = 'data/test_dataset.json' ground_truth = ss.load_data(ground_truth, ignore_first_line=True, verbose=False) # Load the configuration file config = 'data/cp3_s1_configuration.json' config = ss.load_config(config) # Get metadata metadata = ss.MetaData() # Instantiate the task runner task_runner = ss.TaskRunner(ground_truth, config, metadata=metadata) # Run measurements and metrics on the simulation data results, logs = task_runner(simulation, verbose=True)
