230 lines
8.4 KiB
Python
230 lines
8.4 KiB
Python
import pytz
|
|
from pkg.tool import get_tweets_since_last_friday, aggregate_data
|
|
import numpy as np
|
|
from scipy.stats import norm
|
|
from datetime import timedelta, datetime
|
|
from pkg.config import render_data
|
|
|
|
def get_last_7_days_data():
|
|
est = pytz.timezone('US/Eastern')
|
|
now = datetime.now(est).date()
|
|
last_7_days = [now - timedelta(days=i) for i in range(7)]
|
|
data = render_data.global_agg_df[
|
|
render_data.global_agg_df['date'].isin(last_7_days)].copy()
|
|
return data
|
|
|
|
def get_hourly_weighted_array():
|
|
data = get_last_7_days_data()
|
|
if data.empty:
|
|
return [1 / 24] * 24
|
|
|
|
agg_data = aggregate_data(data, 60)
|
|
one_day_data = agg_data.groupby('interval_group')['tweet_count'].sum().reset_index()
|
|
tweet_count_total = one_day_data['tweet_count'].sum()
|
|
|
|
hourly_rates = [0] * 24
|
|
for _, row in one_day_data.iterrows():
|
|
minute = row['interval_group']
|
|
hour = int(minute // 60)
|
|
if hour < 24:
|
|
hourly_rates[hour] = row['tweet_count'] / tweet_count_total if tweet_count_total > 0 else 0
|
|
|
|
total_rate = sum(hourly_rates)
|
|
if total_rate > 0:
|
|
hourly_rates = [rate / total_rate for rate in hourly_rates]
|
|
else:
|
|
hourly_rates = [1 / 24] * 24
|
|
return hourly_rates
|
|
|
|
def calculate_variance_factor():
|
|
data = get_last_7_days_data()
|
|
if data.empty or 'tweet_count' not in data.columns:
|
|
return 1.5
|
|
|
|
data['hour'] = data['minute_of_day'] // 60
|
|
hourly_data = data.groupby(['date', 'hour'])['tweet_count'].sum().reset_index()
|
|
hourly_stats = hourly_data.groupby('hour')['tweet_count'].agg(['mean', 'var']).reset_index()
|
|
variance_factors = hourly_stats['var'] / hourly_stats['mean']
|
|
return np.mean(variance_factors[variance_factors.notna()]) or 1.5
|
|
|
|
def get_dynamic_hourly_weights():
|
|
data = get_last_7_days_data()
|
|
if data.empty:
|
|
return [1 / 24] * 24
|
|
|
|
weights = [0.2, 0.2, 0.3, 0.3, 0.5, 0.5, 0.5]
|
|
hourly_rates = [0] * 24
|
|
|
|
for day_idx, day in enumerate(get_last_7_days_data()['date'].unique()):
|
|
day_data = data[data['date'] == day].copy()
|
|
if day_data.empty:
|
|
continue
|
|
agg_data = aggregate_data(day_data, 60)
|
|
day_tweets = agg_data.groupby('interval_group')['tweet_count'].sum().reset_index()
|
|
day_total = day_tweets['tweet_count'].sum()
|
|
for _, row in day_tweets.iterrows():
|
|
minute = row['interval_group']
|
|
hour = int(minute // 60)
|
|
if hour < 24:
|
|
hourly_rates[hour] += (row['tweet_count'] / day_total if day_total > 0 else 0) * weights[day_idx % 7]
|
|
|
|
total_rate = sum(hourly_rates)
|
|
if total_rate > 0:
|
|
hourly_rates = [rate / total_rate for rate in hourly_rates]
|
|
else:
|
|
hourly_rates = [1 / 24] * 24
|
|
return hourly_rates
|
|
|
|
def get_pace_params():
|
|
est = pytz.timezone('US/Eastern')
|
|
now = datetime.now(est)
|
|
today = now.date()
|
|
days_to_next_friday = (4 - today.weekday()) % 7
|
|
next_friday = (now.replace(hour=12, minute=0, second=0, microsecond=0) +
|
|
timedelta(days=days_to_next_friday))
|
|
if now > next_friday:
|
|
next_friday += timedelta(days=7)
|
|
days_to_next_friday = (next_friday - now).total_seconds() / (24 * 60 * 60)
|
|
tweet_count = get_tweets_since_last_friday()
|
|
return tweet_count, days_to_next_friday
|
|
|
|
def calculate_tweet_pace():
|
|
tweet_count, days_to_next_friday = get_pace_params()
|
|
pace = (tweet_count / (7 - days_to_next_friday)) * days_to_next_friday + tweet_count
|
|
return round(pace, 6) if pace > 0 else float(tweet_count)
|
|
|
|
def calculate_pace_decline_rate():
|
|
tweet_count, days_to_next_friday = get_pace_params()
|
|
T = 7
|
|
decline_per_day = -(tweet_count * T) / ((T - days_to_next_friday) ** 2)
|
|
decline_per_hour = decline_per_day / 24
|
|
return round(decline_per_hour, 2)
|
|
|
|
def calculate_pace_for_increment(increment, hours, tweet_count, days_to_next_friday, current_pace):
|
|
future_days = days_to_next_friday - (hours / 24)
|
|
new_tweet_count = tweet_count + increment
|
|
if future_days <= 0:
|
|
return round(new_tweet_count, 2)
|
|
new_pace = (new_tweet_count / (7 - future_days)) * future_days + new_tweet_count
|
|
return round(max(new_pace, new_tweet_count), 2)
|
|
|
|
def calculate_pace_increase_in_hour(increment_value, hour_value):
|
|
tweet_count, days_to_next_friday = get_pace_params()
|
|
current_pace = (tweet_count / (7 - days_to_next_friday)) * days_to_next_friday + tweet_count
|
|
increments = [0, 1, 5, 10, 20]
|
|
pace_increases = {}
|
|
for inc in increments:
|
|
pace_increases[f'increase_{inc}'] = calculate_pace_for_increment(
|
|
inc, 1, tweet_count, days_to_next_friday, current_pace
|
|
)
|
|
if increment_value is None or hour_value is None:
|
|
pace_increases['custom_increment'] = None
|
|
else:
|
|
increment = int(increment_value)
|
|
hours = int(hour_value)
|
|
pace_increases['custom_increment'] = calculate_pace_for_increment(
|
|
increment, hours, tweet_count, days_to_next_friday, current_pace
|
|
)
|
|
pace_increases['custom_increment_key'] = increment
|
|
return pace_increases
|
|
|
|
def calculate_avg_tweets_per_day(target, now, remain):
|
|
Xi = get_hourly_weighted_array()
|
|
if remain <= 0:
|
|
return "remain<=0"
|
|
if target <= now:
|
|
return "Already reach"
|
|
|
|
fx = max(remain - 12, 0)
|
|
|
|
if remain > 12:
|
|
fy = sum(Xi[0:12]) * 24
|
|
else:
|
|
full_hours = int(remain)
|
|
fractional_hour = remain - full_hours
|
|
if full_hours >= 24:
|
|
full_hours = 23
|
|
fractional_hour = 0
|
|
|
|
if full_hours < 0:
|
|
full_hours = 0
|
|
|
|
if full_hours > 0:
|
|
fy = sum(Xi[0:full_hours]) + Xi[full_hours] * fractional_hour
|
|
else:
|
|
fy = Xi[0] * fractional_hour
|
|
fy *= 24
|
|
|
|
if fx + fy == 0:
|
|
return "fx + fy = 0"
|
|
|
|
result = (target - now) / ((fx + fy) / 24)
|
|
return round(result, 2)
|
|
|
|
def calculate_tweet_probability(tweet_count, days_to_next_friday, prob_start, prob_end, peak_percentile=75):
|
|
remaining_hours = days_to_next_friday * 24
|
|
hourly_weights = get_dynamic_hourly_weights()
|
|
|
|
data = get_last_7_days_data()
|
|
if data.empty:
|
|
recent_tweets = [70] * 7
|
|
else:
|
|
agg_data = aggregate_data(data, 1440)
|
|
daily_tweets = agg_data.groupby('date')['tweet_count'].sum().reset_index()
|
|
recent_tweets = daily_tweets['tweet_count'].tolist()[-7:]
|
|
if len(recent_tweets) < 7:
|
|
recent_tweets = recent_tweets + [70] * (7 - len(recent_tweets))
|
|
|
|
recent_3_days = np.mean(recent_tweets[-3:])
|
|
past_4_days = np.mean(recent_tweets[:-3]) if len(recent_tweets) > 3 else 70
|
|
daily_avg = 0.8 * recent_3_days + 0.2 * past_4_days
|
|
daily_avg_std = np.std(recent_tweets) if len(recent_tweets) >= 7 else np.std([70] * 7)
|
|
|
|
peak_threshold = np.percentile(hourly_weights, peak_percentile)
|
|
segments = []
|
|
current_segment = []
|
|
for i in range(24):
|
|
if hourly_weights[i] >= peak_threshold:
|
|
current_segment.append(i)
|
|
elif current_segment:
|
|
segments.append(current_segment)
|
|
current_segment = []
|
|
if current_segment:
|
|
segments.append(current_segment)
|
|
|
|
lambda_remaining = 0
|
|
variance_factor = calculate_variance_factor()
|
|
total_weight = sum(hourly_weights)
|
|
for segment in segments:
|
|
hours_in_segment = len(segment) * (remaining_hours / 24)
|
|
segment_weight_avg = np.mean([hourly_weights[i] for i in segment])
|
|
lambda_segment = daily_avg * (hours_in_segment / remaining_hours) * (segment_weight_avg / (total_weight / 24))
|
|
lambda_remaining += lambda_segment
|
|
|
|
mu_low = (daily_avg - daily_avg_std) * (remaining_hours / 24)
|
|
mu_high = (daily_avg + daily_avg_std) * (remaining_hours / 24)
|
|
var_low = mu_low * variance_factor
|
|
var_high = mu_high * variance_factor
|
|
sigma_low = np.sqrt(var_low)
|
|
sigma_high = np.sqrt(var_high)
|
|
|
|
a = prob_start - tweet_count
|
|
b = prob_end - tweet_count
|
|
|
|
if tweet_count > prob_end:
|
|
return "0.0000 - 0.0000"
|
|
|
|
if a < 0:
|
|
a = 0
|
|
if b < 0:
|
|
return "0.0000 - 0.0000"
|
|
|
|
prob_low = norm.cdf((b - mu_low) / sigma_low) - norm.cdf((a - mu_low) / sigma_low)
|
|
prob_high = norm.cdf((b - mu_high) / sigma_high) - norm.cdf((a - mu_high) / sigma_high)
|
|
|
|
prob_low = max(0.0, min(1.0, prob_low))
|
|
prob_high = max(0.0, min(1.0, prob_high))
|
|
|
|
prob_min = min(prob_low, prob_high)
|
|
prob_max = max(prob_low, prob_high)
|
|
return f"{prob_min:.4f} - {prob_max:.4f}" |