Files
AIclinicalresearch/extraction_service/operations/dropna.py
HaHafeng b47079b387 feat(iit): Phase 1.5 AI对话集成REDCap真实数据完成
- feat: ChatService集成DeepSeek-V3实现AI对话(390行)
- feat: SessionMemory实现上下文记忆(最近3轮对话,170行)
- feat: 意图识别支持REDCap数据查询(关键词匹配)
- feat: REDCap数据注入LLM(queryRedcapRecord, countRedcapRecords, getProjectInfo)
- feat: 解决LLM幻觉问题(基于真实数据回答,明确system prompt)
- feat: 即时反馈(正在查询...提示)
- test: REDCap查询测试通过(test0102项目,10条记录,ID 7患者详情)
- docs: 创建Phase1.5开发完成记录(313行)
- docs: 更新Phase1.5开发计划(标记完成)
- docs: 更新MVP开发任务清单(Phase 1.5完成)
- docs: 更新模块当前状态(60%完成度)
- docs: 更新系统总体设计文档(v2.6)
- chore: 删除测试脚本(test-redcap-query-for-ai.ts, check-env-config.ts)
- chore: 移除REDCap测试环境变量(REDCAP_TEST_*)

技术亮点:
- AI基于REDCap真实数据对话,不编造信息
- 从数据库读取项目配置,不使用环境变量
- 企业微信端测试通过,用户体验良好

测试通过:
-  查询项目记录总数(10条)
-  查询特定患者详情(ID 7)
-  项目信息查询
-  上下文记忆(3轮对话)
-  即时反馈提示

影响范围:IIT Manager Agent模块
2026-01-03 22:48:10 +08:00

183 lines
4.7 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
删除缺失值 - 预写函数
支持按行删除、按列删除、阈值控制
"""
import pandas as pd
from typing import Literal, Optional, List
def drop_missing_values(
df: pd.DataFrame,
method: Literal['row', 'column', 'both'] = 'row',
threshold: Optional[float] = None,
subset: Optional[List[str]] = None
) -> pd.DataFrame:
"""
删除缺失值
Args:
df: 输入数据框
method: 删除方式
- 'row': 删除包含缺失值的行
- 'column': 删除缺失值过多的列
- 'both': 先删除列,再删除行
threshold: 缺失率阈值0-1之间仅对'column''both'有效
- 如果列的缺失率超过此阈值,则删除该列
- 默认为0.550%
subset: 仅检查指定列的缺失值(仅对'row'有效)
Returns:
删除缺失值后的数据框
示例:
# 删除包含任何缺失值的行
drop_missing_values(df, method='row')
# 删除缺失率>30%的列
drop_missing_values(df, method='column', threshold=0.3)
# 先删除缺失列,再删除缺失行
drop_missing_values(df, method='both', threshold=0.5)
# 仅检查指定列
drop_missing_values(df, method='row', subset=['年龄', 'BMI'])
"""
result = df.copy()
original_shape = result.shape
print(f'原始数据: {original_shape[0]}× {original_shape[1]}')
print(f'缺失值总数: {result.isna().sum().sum()}')
print('')
# 默认阈值
if threshold is None:
threshold = 0.5
# 按列删除
if method in ('column', 'both'):
# 计算每列的缺失率
missing_rate = result.isna().sum() / len(result)
cols_to_drop = missing_rate[missing_rate > threshold].index.tolist()
if cols_to_drop:
print(f'检测到缺失率>{threshold*100:.0f}%的列: {len(cols_to_drop)}')
for col in cols_to_drop:
rate = missing_rate[col]
count = result[col].isna().sum()
print(f' - {col}: 缺失率={rate*100:.1f}% ({count}/{len(result)})')
result = result.drop(columns=cols_to_drop)
print(f'删除后: {result.shape[0]}× {result.shape[1]}')
print('')
else:
print(f'没有找到缺失率>{threshold*100:.0f}%的列')
print('')
# 按行删除
if method in ('row', 'both'):
before_rows = len(result)
if subset:
# 仅检查指定列
print(f'仅检查指定列的缺失值: {subset}')
result = result.dropna(subset=subset)
else:
# 检查所有列
result = result.dropna()
dropped_rows = before_rows - len(result)
if dropped_rows > 0:
print(f'删除了 {dropped_rows} 行(包含缺失值的行)')
print(f'保留了 {len(result)} 行({len(result)/before_rows*100:.1f}%')
else:
print('没有找到包含缺失值的行')
print('')
# 最终统计
final_shape = result.shape
print(f'最终结果: {final_shape[0]}× {final_shape[1]}')
print(f'删除了 {original_shape[0] - final_shape[0]}')
print(f'删除了 {original_shape[1] - final_shape[1]}')
print(f'剩余缺失值: {result.isna().sum().sum()}')
# 如果结果为空,给出警告
if len(result) == 0:
print('\n⚠️ 警告: 删除后数据为空!')
return result
def get_missing_summary(df: pd.DataFrame) -> dict:
"""
获取缺失值统计摘要
Args:
df: 输入数据框
Returns:
缺失值统计信息
"""
total_cells = df.shape[0] * df.shape[1]
total_missing = df.isna().sum().sum()
# 按列统计
col_missing = df.isna().sum()
col_missing_rate = col_missing / len(df)
cols_with_missing = col_missing[col_missing > 0].to_dict()
cols_missing_rate = col_missing_rate[col_missing > 0].to_dict()
# 按行统计
row_missing = df.isna().sum(axis=1)
rows_with_missing = (row_missing > 0).sum()
return {
'total_cells': total_cells,
'total_missing': int(total_missing),
'missing_rate': total_missing / total_cells if total_cells > 0 else 0,
'rows_with_missing': int(rows_with_missing),
'cols_with_missing': len(cols_with_missing),
'col_missing_detail': {
col: {
'count': int(count),
'rate': float(cols_missing_rate[col])
}
for col, count in cols_with_missing.items()
}
}