from tideh import training_cross_validation
number_of_files = 100 # number of files to train on
file_name_prefix = 'data/training/RT' # file names prefix of files used for training
iterations = 5 # number of cross validation iterations
pred_time = 168 # prediction time (hours)
# get file paths of files to use for training
file_names = [
file_name_prefix + str(i) + '.txt' for i in range(1, number_of_files + 1)
]
# load events for optimized training
events_data = []
for file in file_names:
events_data.append(load_events_vec(
file, 1 / 3600, 24)) # convert event_times and start_time to hours
# initial simplex matrix passed to simplex algorithm
# ordering is r0, phi, taum; taum is trained by 1/taum
simplex = [
[0.2, 0, 0.25],
[0, -0.1, 0.25],
[0.2, 0.1, 0.15],
[0.4, 0.0, 0.15],
]
start_values = [0.2, 0, 0.25]
(mean_err, median_err,
param), _ = training_cross_validation(events_data,
iterations,
start_values,
args = parser.parse_args()
month = args.m if args.m else 'nov'
month = month.upper()
input_path = os.path.join("data", "reddit_data", month + "_INPUT")
output_path = os.path.join("data", "reddit_data", month + "_OUTPUT")
subreddit_number = args.srd
filename = args.fl + ".txt"
filename = os.path.join(input_path, subreddit_number, filename)
obs_time = int(args.ot) if args.ot else 1
pred_time = int(args.pt) if args.pt else 720 # 24 * 30
# the number of retweets is not necessary for the further steps
# make sure that all times are loaded in the correct time unit (hours)
# here it is important that there is one nd-array for event times and one for the follower counts
(_, start_time), (event_times, follower) = load_events_vec(filename)
# additional parameters passed to infectious rate function
add_params = {'t0': start_time, 'bounds': [(-1, 0.5), (1, 20.)]}
params, err, _ = estimate_parameters_optimized(event_times=event_times,
follower=follower,
obs_time=obs_time,
**add_params)
results = {}
results['p0'] = params[0]
results['r0'] = params[1]
results['phi0'] = params[2]
results['tm'] = params[3]
results['avg_fit_error'] = err * 100
1) Model parameters of TiDeH (p_0, r_0, phi_0, t_m).
2) Observation time (= obs_time).
Outputs are
1) Estimate of model parameters of TiDeH (p_0, r_0, phi_0, t_m).
This code is developed by Sylvain Gauthier and Sebastian Rühl under the supervision of Ryota Kobayashi.
"""
from tideh.simulate import simulate_time_rescaling
from tideh.functions import infectious_rate_tweets
from tideh import load_events_vec
from tideh import estimate_parameters_optimized
# load pool of follower counts used for simulation from file
file_path = 'data/example/sample_file.txt'
_, (_, follower_pool) = load_events_vec(file_path)
runtime = 72 # simulate for 3 days
# parameters of infectious rate
p0 = 0.001
r0 = 0.424
phi0 = 0.125
taum = 2.
# simulate
event_times, follower = simulate_time_rescaling(
runtime=runtime,
p=lambda t: infectious_rate_tweets(t, p0, r0, phi0, taum),
follower_pool=follower_pool[1:],
int_fol_cnt=follower_pool[0])