|
| 1 | +import json |
| 2 | +import logging |
| 3 | +from pathlib import Path |
| 4 | +from typing import Any, Dict, List, Optional |
| 5 | + |
| 6 | +import pandas as pd |
| 7 | +from openpyxl import Workbook, load_workbook |
| 8 | + |
| 9 | +from dbgpt.util.benchmarks.ExcelUtils import ExcelUtils |
| 10 | + |
| 11 | +from ..file_parse_service import FileParseService |
| 12 | +from ..models import ( |
| 13 | + AnswerExecuteModel, |
| 14 | + BaseInputModel, |
| 15 | + BenchmarkDataSets, |
| 16 | + DataCompareStrategyConfig, |
| 17 | +) |
| 18 | + |
| 19 | +logger = logging.getLogger(__name__) |
| 20 | + |
| 21 | + |
| 22 | +class ExcelFileParseService(FileParseService): |
| 23 | + def parse_input_sets(self, path: str) -> BenchmarkDataSets: |
| 24 | + """ |
| 25 | + Parse input sets from excel file |
| 26 | + Args: |
| 27 | + path: File location path |
| 28 | + Returns: |
| 29 | + BenchmarkDataSets: Parsed data sets |
| 30 | + """ |
| 31 | + input_stream = self.get_input_stream(path) |
| 32 | + |
| 33 | + if input_stream is None: |
| 34 | + raise RuntimeError(f"file not found! path: {path}") |
| 35 | + |
| 36 | + # Parse excel file to get data sets |
| 37 | + input_sets = BenchmarkDataSets() |
| 38 | + workbook = None |
| 39 | + |
| 40 | + try: |
| 41 | + workbook = load_workbook(input_stream, data_only=True) |
| 42 | + input_list = [] |
| 43 | + |
| 44 | + # Get the first worksheet |
| 45 | + sheet = workbook.worksheets[0] |
| 46 | + |
| 47 | + for row_num in range( |
| 48 | + 2, sheet.max_row + 1 |
| 49 | + ): # Skip header row (start from 1-based index) |
| 50 | + row = sheet[row_num] |
| 51 | + if ExcelUtils.is_row_empty(row): |
| 52 | + continue |
| 53 | + |
| 54 | + # Get content from columns 1-6 (0-based index 0-5) |
| 55 | + serial_no_cell = row[0] |
| 56 | + analysis_model_id_cell = row[1] |
| 57 | + question_cell = row[2] |
| 58 | + self_define_tags_cell = row[3] |
| 59 | + knowledge_cell = row[4] |
| 60 | + llm_output_cell = row[5] |
| 61 | + prompt_cell = row[8] |
| 62 | + |
| 63 | + # Build input model |
| 64 | + input_model = BaseInputModel( |
| 65 | + serial_no=int( |
| 66 | + ExcelUtils.get_cell_value_as_string(serial_no_cell) or "0" |
| 67 | + ), |
| 68 | + analysis_model_id=ExcelUtils.get_cell_value_as_string( |
| 69 | + analysis_model_id_cell |
| 70 | + ), |
| 71 | + question=ExcelUtils.get_cell_value_as_string(question_cell), |
| 72 | + self_define_tags=ExcelUtils.get_cell_value_as_string( |
| 73 | + self_define_tags_cell |
| 74 | + ), |
| 75 | + llm_output=ExcelUtils.get_cell_value_as_string(llm_output_cell), |
| 76 | + knowledge=ExcelUtils.get_cell_value_as_string(knowledge_cell), |
| 77 | + prompt=ExcelUtils.get_cell_value_as_string(prompt_cell), |
| 78 | + ) |
| 79 | + |
| 80 | + input_list.append(input_model) |
| 81 | + |
| 82 | + input_sets.data_list = input_list |
| 83 | + except Exception as e: |
| 84 | + logger.error(f"parse excel error, path: {path}, errorMsg: {e}") |
| 85 | + finally: |
| 86 | + try: |
| 87 | + if workbook is not None: |
| 88 | + workbook.close() |
| 89 | + except Exception as e: |
| 90 | + logger.error(f"close workbook error, path: {path}, errorMsg: {e}") |
| 91 | + |
| 92 | + return input_sets |
| 93 | + |
| 94 | + def parse_standard_benchmark_sets( |
| 95 | + self, standard_excel_path: str |
| 96 | + ) -> List[AnswerExecuteModel]: |
| 97 | + df = pd.read_excel(standard_excel_path, sheet_name=0) |
| 98 | + outputs: List[AnswerExecuteModel] = [] |
| 99 | + for _, row in df.iterrows(): |
| 100 | + try: |
| 101 | + serial_no = int(row["编号"]) |
| 102 | + except Exception: |
| 103 | + continue |
| 104 | + question = row.get("用户问题") |
| 105 | + analysis_model_id = row.get("数据集ID") |
| 106 | + llm_output = ( |
| 107 | + None if pd.isna(row.get("标准答案SQL")) else str(row.get("标准答案SQL")) |
| 108 | + ) |
| 109 | + order_by = True |
| 110 | + if not pd.isna(row.get("是否排序")): |
| 111 | + try: |
| 112 | + order_by = bool(int(row.get("是否排序"))) |
| 113 | + except Exception: |
| 114 | + order_by = True |
| 115 | + |
| 116 | + std_result: Optional[List[Dict[str, List[str]]]] = None |
| 117 | + if not pd.isna(row.get("标准结果")): |
| 118 | + std_result_raw = str(row.get("标准结果")) |
| 119 | + std_result = self._parse_multi_standard_result(std_result_raw) |
| 120 | + |
| 121 | + strategy_config = DataCompareStrategyConfig( |
| 122 | + strategy="CONTAIN_MATCH", |
| 123 | + order_by=order_by, |
| 124 | + standard_result=std_result if std_result is not None else None, |
| 125 | + ) |
| 126 | + outputs.append( |
| 127 | + AnswerExecuteModel( |
| 128 | + serialNo=serial_no, |
| 129 | + analysisModelId=analysis_model_id, |
| 130 | + question=question, |
| 131 | + llmOutput=llm_output, |
| 132 | + executeResult=std_result, |
| 133 | + strategyConfig=strategy_config, |
| 134 | + ) |
| 135 | + ) |
| 136 | + return outputs |
| 137 | + |
| 138 | + def _parse_multi_standard_result( |
| 139 | + self, std_result_raw: str |
| 140 | + ) -> Optional[List[Dict[str, List[str]]]]: |
| 141 | + """ |
| 142 | + Parse multiple standard results from raw string data. |
| 143 | +
|
| 144 | + Handles multiple results separated by newlines and parses each line as a dict. |
| 145 | +
|
| 146 | + Args: |
| 147 | + std_result_raw (str): Raw standard result string with multiple lines |
| 148 | +
|
| 149 | + Returns: |
| 150 | + Optional[List[Dict[str, List[str]]]]: List of parsed dictionaries, |
| 151 | + or None if parsing fails or no valid data |
| 152 | + """ |
| 153 | + try: |
| 154 | + std_result_raw = std_result_raw.strip() |
| 155 | + if not std_result_raw: |
| 156 | + return None |
| 157 | + |
| 158 | + # 处理多个结果，通过换行符分隔 |
| 159 | + result_lines = std_result_raw.split("\n") |
| 160 | + result_list = [] |
| 161 | + |
| 162 | + for line in result_lines: |
| 163 | + line = line.strip() |
| 164 | + if line: |
| 165 | + try: |
| 166 | + result_list.append(json.loads(line)) |
| 167 | + except Exception as e: |
| 168 | + logger.warning( |
| 169 | + f"Failed to parse line as JSON: {line}, error: {e}" |
| 170 | + ) |
| 171 | + continue |
| 172 | + |
| 173 | + return result_list if result_list else None |
| 174 | + except Exception as e: |
| 175 | + logger.error(f"parse multiple standard results error: {e}") |
| 176 | + return None |
0 commit comments