shenjianZ
77e38fd15b
- 将 LSTMMLPRegressor 重构为 TemporalFusionRegressor，采用 Transformer Encoder 替代 LSTM - 新增 LearnedAttentionPooling 和 GatedResidualBlock 模块增强模型表达能力 - 优化训练策略，使用 OneCycleLR 调度器和样本加权机制 - 改进缺勤事件采样算法，基于压力、健康、家庭等维度更精确地计算缺勤时长 - 更新 .gitignore 排除原始数据文件，删除不再使用的原始 CSV 文件
406 lines
17 KiB
Python
406 lines
17 KiB
Python
import os
|
|
import sys
|
|
|
|
import numpy as np
|
|
import pandas as pd
|
|
|
|
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
|
|
|
import config
|
|
|
|
|
|
INDUSTRIES = {
|
|
'制造业': {'shift_bias': 0.9, 'overtime_bias': 0.8, 'night_bias': 0.8},
|
|
'互联网': {'shift_bias': 0.2, 'overtime_bias': 1.0, 'night_bias': 0.2},
|
|
'零售连锁': {'shift_bias': 0.7, 'overtime_bias': 0.5, 'night_bias': 0.3},
|
|
'物流运输': {'shift_bias': 0.9, 'overtime_bias': 0.7, 'night_bias': 0.9},
|
|
'金融服务': {'shift_bias': 0.1, 'overtime_bias': 0.7, 'night_bias': 0.1},
|
|
'医药健康': {'shift_bias': 0.6, 'overtime_bias': 0.6, 'night_bias': 0.5},
|
|
'建筑工程': {'shift_bias': 0.5, 'overtime_bias': 0.8, 'night_bias': 0.3},
|
|
}
|
|
|
|
|
|
def season_from_month(month):
|
|
if month in [12, 1, 2]:
|
|
return 1
|
|
if month in [3, 4, 5]:
|
|
return 2
|
|
if month in [6, 7, 8]:
|
|
return 3
|
|
return 4
|
|
|
|
|
|
def weighted_choice(rng, items, probs):
|
|
probs = np.array(probs, dtype=float)
|
|
probs = probs / probs.sum()
|
|
return rng.choice(items, p=probs)
|
|
|
|
|
|
def build_company_pool(rng, company_count=180):
|
|
industries = list(INDUSTRIES.keys())
|
|
scales = ['100人以下', '100-499人', '500-999人', '1000-4999人', '5000人及以上']
|
|
city_tiers = ['一线', '新一线', '二线', '三线及以下']
|
|
companies = []
|
|
for idx in range(company_count):
|
|
industry = weighted_choice(rng, industries, [0.22, 0.14, 0.14, 0.14, 0.1, 0.12, 0.14])
|
|
companies.append({
|
|
'企业编号': f'C{idx + 1:03d}',
|
|
'所属行业': industry,
|
|
'企业规模': weighted_choice(rng, scales, [0.15, 0.28, 0.2, 0.24, 0.13]),
|
|
'所在城市等级': weighted_choice(rng, city_tiers, [0.18, 0.34, 0.3, 0.18]),
|
|
})
|
|
return companies
|
|
|
|
|
|
def build_employee_pool(rng, companies, employee_count=2600):
|
|
genders = ['男', '女']
|
|
employment_types = ['正式员工', '劳务派遣', '外包驻场', '实习生']
|
|
departments = ['生产', '研发', '销售', '客服', '职能', '仓储物流', '门店运营']
|
|
job_families = ['管理', '专业技术', '销售业务', '生产操作', '行政支持', '客服坐席']
|
|
job_levels = ['初级', '中级', '高级', '主管', '经理及以上']
|
|
educations = ['中专及以下', '大专', '本科', '硕士', '博士']
|
|
marital = ['未婚', '已婚', '离异/其他']
|
|
housing = ['自有住房', '租房', '宿舍']
|
|
shifts = ['标准白班', '两班倒', '三班倒', '弹性班']
|
|
performance = ['A', 'B', 'C', 'D']
|
|
stress = ['低', '中', '高']
|
|
|
|
employees = []
|
|
for idx in range(employee_count):
|
|
company = companies[rng.integers(0, len(companies))]
|
|
industry = company['所属行业']
|
|
age = int(np.clip(rng.normal(33, 7), 20, 55))
|
|
tenure = round(float(np.clip(age - 21 + rng.normal(0, 2), 0.2, 32)), 1)
|
|
family_bias = 0.6 if age >= 30 else 0.25
|
|
married = weighted_choice(rng, marital, [0.45, 0.48, 0.07] if age < 30 else [0.18, 0.72, 0.1])
|
|
children = int(np.clip(rng.poisson(0.4 if married == '未婚' else family_bias), 0, 3))
|
|
industry_profile = INDUSTRIES[industry]
|
|
shift = weighted_choice(
|
|
rng,
|
|
shifts,
|
|
[
|
|
max(0.1, 1 - industry_profile['shift_bias']),
|
|
0.35 * industry_profile['shift_bias'],
|
|
0.25 * industry_profile['shift_bias'],
|
|
0.2,
|
|
],
|
|
)
|
|
night_flag = int(shift == '三班倒' or (shift == '两班倒' and rng.random() < industry_profile['night_bias']))
|
|
overtime = float(np.clip(rng.normal(22 + 18 * industry_profile['overtime_bias'], 10), 0, 90))
|
|
commute_minutes = float(np.clip(rng.normal(42, 18), 8, 130))
|
|
commute_km = float(np.clip(commute_minutes * rng.uniform(0.35, 0.75), 2, 65))
|
|
performance_level = weighted_choice(rng, performance, [0.18, 0.46, 0.26, 0.1])
|
|
chronic_flag = int(rng.random() < max(0.05, (age - 26) * 0.01))
|
|
check_abnormal = int(chronic_flag == 1 or rng.random() < 0.14)
|
|
sleep_hours = round(float(np.clip(rng.normal(6.9 - 0.35 * night_flag, 0.8), 4.5, 9.0)), 1)
|
|
exercise = int(np.clip(rng.poisson(2.2), 0, 7))
|
|
smoking = int(rng.random() < (0.22 if rng.random() < 0.55 else 0.08))
|
|
drinking = int(rng.random() < 0.27)
|
|
stress_level = weighted_choice(
|
|
rng,
|
|
stress,
|
|
[0.22, 0.52, 0.26 + min(0.15, overtime / 120)],
|
|
)
|
|
bmi = round(float(np.clip(rng.normal(24.2, 3.2), 17.5, 36.5)), 1)
|
|
history_count = int(np.clip(rng.poisson(1.2 + chronic_flag * 0.6 + children * 0.15), 0, 8))
|
|
history_hours = float(np.clip(rng.normal(18 + chronic_flag * 10 + history_count * 3, 10), 0, 120))
|
|
discipline = int(np.clip(rng.poisson(0.2), 0, 4))
|
|
team_size = int(np.clip(rng.normal(11, 5), 3, 40))
|
|
manager_span = int(np.clip(team_size + rng.normal(3, 2), 4, 60))
|
|
local_hukou = int(rng.random() < 0.58)
|
|
cross_city = int(commute_minutes > 65 or (local_hukou == 0 and rng.random() < 0.35))
|
|
sedentary = int(weighted_choice(rng, [0, 1], [0.45, 0.55]) if company['所属行业'] in ['互联网', '金融服务'] else rng.random() < 0.3)
|
|
|
|
employees.append({
|
|
'企业编号': company['企业编号'],
|
|
'所属行业': industry,
|
|
'企业规模': company['企业规模'],
|
|
'所在城市等级': company['所在城市等级'],
|
|
'用工类型': weighted_choice(rng, employment_types, [0.74, 0.12, 0.1, 0.04]),
|
|
'部门条线': weighted_choice(rng, departments, [0.18, 0.16, 0.14, 0.11, 0.12, 0.14, 0.15]),
|
|
'岗位序列': weighted_choice(rng, job_families, [0.08, 0.24, 0.16, 0.2, 0.12, 0.2]),
|
|
'岗位级别': weighted_choice(rng, job_levels, [0.34, 0.32, 0.18, 0.11, 0.05]),
|
|
'员工编号': f'E{idx + 1:05d}',
|
|
'性别': weighted_choice(rng, genders, [0.56, 0.44]),
|
|
'年龄': age,
|
|
'司龄年数': tenure,
|
|
'最高学历': weighted_choice(rng, educations, [0.14, 0.28, 0.4, 0.15, 0.03]),
|
|
'婚姻状态': married,
|
|
'是否本地户籍': local_hukou,
|
|
'子女数量': children,
|
|
'是否独生子女家庭负担': int(children >= 2 or (married == '已婚' and rng.random() < 0.18)),
|
|
'居住类型': weighted_choice(rng, housing, [0.38, 0.48, 0.14]),
|
|
'班次类型': shift,
|
|
'是否夜班岗位': night_flag,
|
|
'月均加班时长': round(overtime, 1),
|
|
'近30天出勤天数': int(np.clip(rng.normal(21.5, 2.2), 14, 27)),
|
|
'近90天缺勤次数': history_count,
|
|
'近180天请假总时长': round(history_hours, 1),
|
|
'通勤时长分钟': round(commute_minutes, 1),
|
|
'通勤距离公里': round(commute_km, 1),
|
|
'是否跨城通勤': cross_city,
|
|
'绩效等级': performance_level,
|
|
'近12月违纪次数': discipline,
|
|
'团队人数': team_size,
|
|
'直属上级管理跨度': manager_span,
|
|
'BMI': bmi,
|
|
'是否慢性病史': chronic_flag,
|
|
'年度体检异常标记': check_abnormal,
|
|
'近30天睡眠时长均值': sleep_hours,
|
|
'每周运动频次': exercise,
|
|
'是否吸烟': smoking,
|
|
'是否饮酒': drinking,
|
|
'心理压力等级': stress_level,
|
|
'是否长期久坐岗位': sedentary,
|
|
})
|
|
return employees
|
|
|
|
|
|
def sample_event(rng, employee):
|
|
month = int(rng.integers(1, 13))
|
|
weekday = int(rng.integers(1, 8))
|
|
near_holiday = int(rng.random() < (0.3 if month in [1, 2, 4, 5, 9, 10] else 0.16))
|
|
leave_type_items = ['病假', '事假', '年假', '调休', '婚假', '丧假', '产检育儿假', '工伤假', '其他']
|
|
leave_probs = [0.26, 0.22, 0.11, 0.14, 0.03, 0.02, 0.07, 0.03, 0.12]
|
|
if employee['是否慢性病史'] == 1 or employee['年度体检异常标记'] == 1:
|
|
leave_probs = [0.34, 0.18, 0.08, 0.1, 0.02, 0.02, 0.08, 0.04, 0.14]
|
|
elif employee['子女数量'] >= 2:
|
|
leave_probs = [0.22, 0.24, 0.1, 0.12, 0.03, 0.02, 0.12, 0.02, 0.13]
|
|
leave_type = weighted_choice(rng, leave_type_items, leave_probs)
|
|
|
|
if leave_type in ['病假', '工伤假']:
|
|
reason_category = weighted_choice(rng, ['身体不适', '就医复查', '职业疲劳'], [0.52, 0.3, 0.18])
|
|
elif leave_type == '产检育儿假':
|
|
reason_category = weighted_choice(rng, ['子女照护', '家庭事务', '身体不适'], [0.6, 0.25, 0.15])
|
|
elif leave_type in ['婚假', '丧假']:
|
|
reason_category = weighted_choice(rng, ['家庭事务', '突发事件'], [0.72, 0.28])
|
|
elif leave_type in ['年假', '调休']:
|
|
reason_category = weighted_choice(rng, ['职业疲劳', '家庭事务', '交通受阻'], [0.52, 0.28, 0.2])
|
|
else:
|
|
reason_category = weighted_choice(
|
|
rng,
|
|
['身体不适', '家庭事务', '子女照护', '交通受阻', '突发事件', '职业疲劳'],
|
|
[0.2, 0.22, 0.14, 0.12, 0.12, 0.2],
|
|
)
|
|
|
|
medical_certificate = int(
|
|
leave_type in ['病假', '工伤假']
|
|
or reason_category in ['身体不适', '就医复查']
|
|
or (employee['是否慢性病史'] == 1 and leave_type == '其他')
|
|
)
|
|
urgent_leave = int(
|
|
leave_type in ['病假', '工伤假']
|
|
or reason_category in ['突发事件', '身体不适']
|
|
or (near_holiday == 0 and leave_type == '事假' and rng.random() < 0.35)
|
|
)
|
|
continuous_absence = int(
|
|
leave_type in ['病假', '工伤假', '产检育儿假']
|
|
and (employee['近90天缺勤次数'] >= 2 or employee['近180天请假总时长'] >= 28)
|
|
)
|
|
previous_overtime = int(
|
|
employee['月均加班时长'] >= 30
|
|
or (employee['月均加班时长'] >= 24 and weekday in [1, 2, 5])
|
|
or (employee['是否夜班岗位'] == 1 and rng.random() < 0.65)
|
|
)
|
|
season = season_from_month(month)
|
|
channel = weighted_choice(rng, ['系统申请', '主管代提', '临时电话报备'], [0.68, 0.18, 0.14])
|
|
|
|
pressure_score = (
|
|
employee['月均加班时长'] * 0.032
|
|
+ employee['通勤时长分钟'] * 0.018
|
|
+ employee['是否夜班岗位'] * 0.75
|
|
+ employee['是否跨城通勤'] * 0.32
|
|
+ previous_overtime * 0.35
|
|
)
|
|
health_score = (
|
|
employee['是否慢性病史'] * 1.2
|
|
+ employee['年度体检异常标记'] * 0.55
|
|
+ (employee['BMI'] >= 28) * 0.3
|
|
+ (employee['近30天睡眠时长均值'] < 6.4) * 0.45
|
|
)
|
|
family_score = employee['子女数量'] * 0.18 + employee['是否独生子女家庭负担'] * 0.28
|
|
resilience_score = (
|
|
(0.55 if employee['绩效等级'] == 'A' else 0.25 if employee['绩效等级'] == 'B' else 0.0)
|
|
+ min(employee['司龄年数'] / 26, 0.65)
|
|
+ min(employee['每周运动频次'] * 0.06, 0.25)
|
|
)
|
|
|
|
base = 0.35
|
|
base += pressure_score
|
|
base += health_score
|
|
base += family_score
|
|
base += 0.4 if employee['心理压力等级'] == '高' else (0.18 if employee['心理压力等级'] == '中' else -0.05)
|
|
base += 0.18 if near_holiday else 0.0
|
|
base += 0.35 if continuous_absence else 0.0
|
|
base += 0.28 if employee['近90天缺勤次数'] >= 3 else 0.0
|
|
base += 0.18 if employee['近180天请假总时长'] >= 36 else 0.0
|
|
base -= resilience_score
|
|
|
|
leave_bonus = {
|
|
'病假': 2.1,
|
|
'事假': 0.8,
|
|
'年假': 0.15,
|
|
'调休': 0.1,
|
|
'婚假': 3.1,
|
|
'丧假': 2.8,
|
|
'产检育儿假': 2.35,
|
|
'工伤假': 3.9,
|
|
'其他': 0.55,
|
|
}
|
|
reason_bonus = {
|
|
'身体不适': 1.0,
|
|
'家庭事务': 0.55,
|
|
'子女照护': 0.75,
|
|
'交通受阻': 0.2,
|
|
'突发事件': 0.6,
|
|
'职业疲劳': 0.7,
|
|
'就医复查': 1.15,
|
|
}
|
|
industry_bonus = {
|
|
'制造业': 0.42,
|
|
'互联网': 0.22,
|
|
'零售连锁': 0.28,
|
|
'物流运输': 0.5,
|
|
'金融服务': 0.12,
|
|
'医药健康': 0.24,
|
|
'建筑工程': 0.4,
|
|
}
|
|
season_bonus = {1: 0.35, 2: 0.0, 3: 0.15, 4: 0.05}
|
|
weekday_bonus = {1: 0.05, 2: 0.0, 3: 0.0, 4: 0.05, 5: 0.15, 6: 0.25, 7: 0.3}
|
|
|
|
duration = base
|
|
duration += leave_bonus[leave_type]
|
|
duration += reason_bonus[reason_category]
|
|
duration += industry_bonus[employee['所属行业']]
|
|
duration += season_bonus[season]
|
|
duration += weekday_bonus[weekday]
|
|
duration += 0.55 if medical_certificate else 0.0
|
|
duration += 0.28 if urgent_leave else -0.06
|
|
|
|
if leave_type == '病假' and employee['是否慢性病史'] == 1:
|
|
duration += 0.85
|
|
if leave_type == '工伤假':
|
|
duration += 1.0 + employee['是否夜班岗位'] * 0.3
|
|
if leave_type in ['婚假', '丧假']:
|
|
duration += 0.7 + 0.18 * near_holiday
|
|
if leave_type == '产检育儿假':
|
|
duration += 0.55 + employee['子女数量'] * 0.12
|
|
if leave_type in ['年假', '调休']:
|
|
duration *= 0.82 if near_holiday == 0 else 0.9
|
|
|
|
duration = round(float(np.clip(duration + rng.normal(0, 0.35), 0.5, 18.0)), 1)
|
|
|
|
event = employee.copy()
|
|
event.update({
|
|
'缺勤月份': month,
|
|
'星期几': weekday,
|
|
'是否节假日前后': near_holiday,
|
|
'季节': season,
|
|
'请假申请渠道': channel,
|
|
'请假类型': leave_type,
|
|
'请假原因大类': reason_category,
|
|
'是否提供医院证明': medical_certificate,
|
|
'是否临时请假': urgent_leave,
|
|
'是否连续缺勤': continuous_absence,
|
|
'前一工作日是否加班': previous_overtime,
|
|
'缺勤时长（小时）': duration,
|
|
})
|
|
return event
|
|
|
|
|
|
def attach_event_timeline(df):
|
|
df = df.copy()
|
|
rng = np.random.default_rng(config.RANDOM_STATE)
|
|
base_date = np.datetime64('2025-01-01')
|
|
timelines = []
|
|
|
|
for employee_id, group in df.groupby('员工编号', sort=False):
|
|
group = group.copy().reset_index(drop=True)
|
|
event_count = len(group)
|
|
offsets = np.sort(rng.integers(0, 365, size=event_count))
|
|
group['事件日期'] = [
|
|
str(pd.Timestamp(base_date + np.timedelta64(int(offset), 'D')).date())
|
|
for offset in offsets
|
|
]
|
|
group['事件日期索引'] = offsets.astype(int)
|
|
group['事件序号'] = np.arange(1, event_count + 1)
|
|
group['员工历史事件数'] = event_count
|
|
timelines.append(group)
|
|
|
|
return pd.concat(timelines, ignore_index=True)
|
|
|
|
|
|
def validate_dataset(df):
|
|
required_columns = [
|
|
'员工编号',
|
|
'所属行业',
|
|
'岗位序列',
|
|
'月均加班时长',
|
|
'通勤时长分钟',
|
|
'是否慢性病史',
|
|
'请假类型',
|
|
'事件序号',
|
|
'事件日期索引',
|
|
'员工历史事件数',
|
|
'缺勤时长（小时）',
|
|
]
|
|
for column in required_columns:
|
|
if column not in df.columns:
|
|
raise ValueError(f'Missing required column: {column}')
|
|
|
|
if len(df) < 10000:
|
|
raise ValueError('Synthetic dataset is smaller than expected')
|
|
if df['员工编号'].nunique() < 2000:
|
|
raise ValueError('Employee coverage is too small')
|
|
|
|
high_risk_ratio = (df['缺勤时长（小时）'] > 8).mean()
|
|
if not 0.15 <= high_risk_ratio <= 0.4:
|
|
raise ValueError(f'High risk ratio out of range: {high_risk_ratio:.3f}')
|
|
|
|
medical_mean = df[df['是否提供医院证明'] == 1]['缺勤时长（小时）'].mean()
|
|
no_medical_mean = df[df['是否提供医院证明'] == 0]['缺勤时长（小时）'].mean()
|
|
if medical_mean <= no_medical_mean:
|
|
raise ValueError('Medical certificate signal is not effective')
|
|
|
|
night_mean = df[df['是否夜班岗位'] == 1]['缺勤时长（小时）'].mean()
|
|
day_mean = df[df['是否夜班岗位'] == 0]['缺勤时长（小时）'].mean()
|
|
if night_mean <= day_mean:
|
|
raise ValueError('Night shift signal is not effective')
|
|
|
|
|
|
def generate_dataset(output_path=None, sample_count=12000, random_state=None):
|
|
rng = np.random.default_rng(config.RANDOM_STATE if random_state is None else random_state)
|
|
companies = build_company_pool(rng)
|
|
employees = build_employee_pool(rng, companies)
|
|
|
|
events = []
|
|
employee_idx = rng.integers(0, len(employees), size=sample_count)
|
|
for idx in employee_idx:
|
|
events.append(sample_event(rng, employees[int(idx)]))
|
|
|
|
df = attach_event_timeline(pd.DataFrame(events))
|
|
validate_dataset(df)
|
|
|
|
if output_path:
|
|
os.makedirs(os.path.dirname(output_path), exist_ok=True)
|
|
df.to_csv(output_path, index=False, encoding='utf-8-sig')
|
|
return df
|
|
|
|
|
|
def ensure_dataset():
|
|
if not os.path.exists(config.RAW_DATA_PATH):
|
|
generate_dataset(config.RAW_DATA_PATH)
|
|
return
|
|
|
|
try:
|
|
df = pd.read_csv(config.RAW_DATA_PATH)
|
|
validate_dataset(df)
|
|
except Exception:
|
|
generate_dataset(config.RAW_DATA_PATH)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
dataset = generate_dataset(config.RAW_DATA_PATH)
|
|
print(f'Generated dataset: {config.RAW_DATA_PATH}')
|
|
print(dataset.head())
|