diff --git a/examples/pipelines/puzzle_configs/InChI2logPbootcamp_test.json b/examples/pipelines/puzzle_configs/InChI2logP_test.json similarity index 100% rename from examples/pipelines/puzzle_configs/InChI2logPbootcamp_test.json rename to examples/pipelines/puzzle_configs/InChI2logP_test.json diff --git a/examples/pipelines/puzzle_configs/InChI2logPbootcamp_train.json b/examples/pipelines/puzzle_configs/InChI2logP_train.json similarity index 100% rename from examples/pipelines/puzzle_configs/InChI2logPbootcamp_train.json rename to examples/pipelines/puzzle_configs/InChI2logP_train.json diff --git a/examples/pipelines/puzzle_configs/SMILES2logPBootCamp_test.json b/examples/pipelines/puzzle_configs/SMILES2logP_test.json similarity index 100% rename from examples/pipelines/puzzle_configs/SMILES2logPBootCamp_test.json rename to examples/pipelines/puzzle_configs/SMILES2logP_test.json diff --git a/examples/pipelines/puzzle_configs/SMILES2logPBootCamp_train.json b/examples/pipelines/puzzle_configs/SMILES2logP_train.json similarity index 100% rename from examples/pipelines/puzzle_configs/SMILES2logPBootCamp_train.json rename to examples/pipelines/puzzle_configs/SMILES2logP_train.json diff --git a/examples/pipelines/puzzle_configs/earth_dew2humidity_test.json b/examples/pipelines/puzzle_configs/earthdew2humidity_test.json similarity index 100% rename from examples/pipelines/puzzle_configs/earth_dew2humidity_test.json rename to examples/pipelines/puzzle_configs/earthdew2humidity_test.json diff --git a/examples/pipelines/puzzle_configs/earth_dew2humidity_train.json b/examples/pipelines/puzzle_configs/earthdew2humidity_train.json similarity index 100% rename from examples/pipelines/puzzle_configs/earth_dew2humidity_train.json rename to examples/pipelines/puzzle_configs/earthdew2humidity_train.json diff --git a/internbootcamp/bootcamp/ChemStructure2Property/InChI2MRBootCamp.py b/internbootcamp/bootcamp/ChemStructure2Property/InChI2MRBootCamp.py index 5710aca..3e2bf1e 100755 --- a/internbootcamp/bootcamp/ChemStructure2Property/InChI2MRBootCamp.py +++ b/internbootcamp/bootcamp/ChemStructure2Property/InChI2MRBootCamp.py @@ -2,7 +2,7 @@ from rdkit import Chem from rdkit.Chem import Crippen from .InChI2logPBootCamp import InChI2logPbootcamp -class InChI2MRBootCamp(InChI2logPbootcamp): +class InChI2MRbootCamp(InChI2logPbootcamp): def prompt_func(self, InChI) -> str: diff --git a/internbootcamp/bootcamp/ChemStructure2Property/InChI2logPBootCamp.py b/internbootcamp/bootcamp/ChemStructure2Property/InChI2logPBootCamp.py index 3e53750..702a00d 100755 --- a/internbootcamp/bootcamp/ChemStructure2Property/InChI2logPBootCamp.py +++ b/internbootcamp/bootcamp/ChemStructure2Property/InChI2logPBootCamp.py @@ -8,13 +8,16 @@ from rdkit.Chem import Crippen class InChI2logPbootcamp(Basebootcamp): def __init__(self, max_atoms=15, min_atoms=3, elements=None, seed=None): # super.__init__() - self.InChIGenerator = InChIGenerator(max_atoms=max_atoms, min_atoms=min_atoms, elements=elements, seed=seed) - self.tolerance_factor = tolerance_factor # 1 for 1% error consider true, 0.1 for 0.1% error true, 10 for 10% error + self.max_atoms = max_atoms + self.min_atoms = min_atoms + # self.InChIGenerator = InChIGenerator(max_atoms=max_atoms, min_atoms=min_atoms, elements=elements, seed=seed) + # self.tolerance_factor = tolerance_factor # 1 for 1% error consider true, 0.1 for 0.1% error true, 10 for 10% error def case_generator(self) -> str: """ 生成一组数字和目标值。 """ + self.InChIGenerator = InChIGenerator(max_atoms=self.max_atoms, min_atoms=self.min_atoms, elements=None, seed=None) return self.InChIGenerator.generate_n_valid_inchi(1)[0] def prompt_func(self, InChI) -> str: diff --git a/internbootcamp/bootcamp/ChemStructure2Property/SMILES2MRBootCamp.py b/internbootcamp/bootcamp/ChemStructure2Property/SMILES2MRBootCamp.py index a76d3f5..a2c3e52 100755 --- a/internbootcamp/bootcamp/ChemStructure2Property/SMILES2MRBootCamp.py +++ b/internbootcamp/bootcamp/ChemStructure2Property/SMILES2MRBootCamp.py @@ -4,9 +4,9 @@ from .utils import last_boxed_only_string, remove_boxed from rdkit import Chem from rdkit.Chem import Crippen -from .SMILES2logPBootCamp import SMILES2logPBootCamp +from .SMILES2logPBootCamp import SMILES2logPbootcamp -class SMILES2MRBootCamp(SMILES2logPBootCamp): +class SMILES2MRbootCamp(SMILES2logPbootcamp): def prompt_func(self, SMILES) -> str: diff --git a/internbootcamp/bootcamp/ChemStructure2Property/SMILES2logPBootCamp.py b/internbootcamp/bootcamp/ChemStructure2Property/SMILES2logPBootCamp.py index 1bb1303..79144fd 100755 --- a/internbootcamp/bootcamp/ChemStructure2Property/SMILES2logPBootCamp.py +++ b/internbootcamp/bootcamp/ChemStructure2Property/SMILES2logPBootCamp.py @@ -6,16 +6,19 @@ from rdkit.Chem import Crippen from .InChI2logPBootCamp import InChI2logPbootcamp -class SMILES2logPBootCamp(InChI2logPbootcamp): +class SMILES2logPbootcamp(InChI2logPbootcamp): def __init__(self,min_len=5, max_len=25, seed=None): # super.__init__() - self.SMILESGenerator = SMILESGenerator(min_len=min_len, max_len=max_len, seed=seed) + self.min_len = min_len + self.max_len = max_len + # self.SMILESGenerator = SMILESGenerator(min_len=min_len, max_len=max_len, seed=seed) def case_generator(self) -> str: """ 生成一组数字和目标值。 """ + self.SMILESGenerator = SMILESGenerator(min_len=self.min_len, max_len=self.max_len, seed=None) return self.SMILESGenerator.generate_n_valid_smiles(1)[0] def prompt_func(self, SMILES) -> str: diff --git a/internbootcamp/bootcamp/__init__.py b/internbootcamp/bootcamp/__init__.py index 7a73937..9c69795 100755 --- a/internbootcamp/bootcamp/__init__.py +++ b/internbootcamp/bootcamp/__init__.py @@ -32,11 +32,14 @@ from .bbeh_boolean_expressions.bbeh_boolean_expressions import Bbehbooleanexpres from .ChemStructure2Property.InChI2logPBootCamp import InChI2logPbootcamp -from .ChemStructure2Property.InChI2MRBootCamp import InChI2MRBootCamp -from .ChemStructure2Property.SMILES2logPBootCamp import SMILES2logPBootCamp -from .ChemStructure2Property.SMILES2MRBootCamp import SMILES2MRBootCamp +from .ChemStructure2Property.InChI2MRBootCamp import InChI2MRbootCamp +from .ChemStructure2Property.SMILES2logPBootCamp import SMILES2logPbootcamp +from .ChemStructure2Property.SMILES2MRBootCamp import SMILES2MRbootCamp +from .earth_dew2humidity.earth_dew2humidity import Earthdew2humiditybootcamp +from .linearODE.linearODE import LinearODEbootcamp + from .kakurasu.kakurasu import Kakurasubootcamp from .nonograms.nonograms import Nonogramsbootcamp from .hitori.hitori import Hitoribootcamp diff --git a/internbootcamp/bootcamp/earth_dew2humidity/earth_dew2humidity.py b/internbootcamp/bootcamp/earth_dew2humidity/earth_dew2humidity.py index e07aadf..bf803fd 100644 --- a/internbootcamp/bootcamp/earth_dew2humidity/earth_dew2humidity.py +++ b/internbootcamp/bootcamp/earth_dew2humidity/earth_dew2humidity.py @@ -4,7 +4,7 @@ import numpy as np from internbootcamp.bootcamp.base import Basebootcamp -class earth_dew2humidity(Basebootcamp): +class Earthdew2humiditybootcamp(Basebootcamp): def __init__( self, temperature_range=(-20, 40), @@ -37,49 +37,71 @@ class earth_dew2humidity(Basebootcamp): "e = 611.2 * np.exp(17.67 * (dewpoint - 273.15) / (dewpoint - 29.65))\n" "e_s = 611.2 * np.exp(17.67 * (temperature - 273.15) / (temperature - 29.65))\n" "relative humidity = e / e_s * 100\n" - "只需返回 “relative humidity = ”。" + "以\\boxed{relative humidity = ？%} 格式输出你的最终答案，例如\\boxed{relative humidity = your answer%}。" ) @staticmethod - def extract_output(output: str) -> str: - # 用正则提取“relative humidity = …”右侧的表达式 - m = re.search(r"relative humidity\s*=\s*([^\n\r]+)", output) - return m.group(1).strip() if m else None + def extract_output(output): + boxed_start_index = output.find('\\boxed{') + boxed_end_index = output.rfind('}', boxed_start_index) + boxed = output[boxed_start_index + 7:boxed_end_index] + # print(boxed) + # 提取数字（含小数点） + number_match = re.findall(r'\d+(?:\.\d+)?', boxed)[-1] + if number_match: + try: + return float(number_match) + except ValueError: + return None + return None @classmethod def _verify_correction(cls, solution: str, identity: dict) -> bool: # 解析 LLM 给出的系数 c，形如 “c*x” - solution = solution.replace(" ", "") try: c = float(solution) except: return False # print(c) # 验证 c ≈ k - return abs(c - identity["humidity"]) < 1e-2 + return abs(c - identity["humidity"]) < 1e-1 if __name__ == "__main__": - bootcamp = earth_dew2humidity(seed=123) - # 生成几个样例 - examples = [bootcamp.case_generator() for _ in range(3)] - print(examples) - print(bootcamp.prompt_func(examples[0])) - print(bootcamp.extract_output("xxxxx relative humidity = 111222 ")) - solution = bootcamp.extract_output("xxxxx relative humidity = 84.79 ") - print(bootcamp._verify_correction(solution, examples[0])) - solution = bootcamp.extract_output("xxxxx relative humidity = 83.79 ") - print(bootcamp._verify_correction(solution, examples[0])) + # bootcamp = Earthdew2humiditybootcamp(seed=123) + # # 生成几个样例 + # examples = [bootcamp.case_generator() for _ in range(3)] + # print(examples) + # print(bootcamp.prompt_func(examples[0])) + # print(bootcamp.extract_output("xxxxx relative humidity = 111222 ")) + # solution = bootcamp.extract_output("xxxxx relative humidity = 84.79 ") + # print(bootcamp._verify_correction(solution, examples[0])) + # solution = bootcamp.extract_output("[answer]relative humidity = relative humidity = 83.79%[/answer]") + # print(bootcamp._verify_correction(solution, examples[0])) - for identity in examples: - # 构造“模型”返回答案，模拟 LLM 的输出 - humidity = identity["humidity"] - sol = f"{humidity:.4f}" - # 调用 Basebootcamp 提供的 verify_score 接口进行验证 - score = bootcamp.verify_score(sol, identity, short_threshold=1e-2) - # 打印结果 - print(json.dumps({ - "identity": identity, - "solution": sol, - "verify_score": score - }, ensure_ascii=False, indent=2)) \ No newline at end of file + # for identity in examples: + # # 构造“模型”返回答案，模拟 LLM 的输出 + # humidity = identity["humidity"] + # sol = f"\\boxed{{relative humidity = {humidity:.4f}%}}" + # # 调用 Basebootcamp 提供的 verify_score 接口进行验证 + # score = bootcamp.verify_score(sol, identity, short_penalty=False,format_penalty=False) + # # 打印结果 + # print(json.dumps({ + # "identity": identity, + # "solution": sol, + # "extract_output": Earthdew2humiditybootcamp.extract_output(sol), + # "verify_score": score + # }, ensure_ascii=False, indent=2)) + case = { + "id": 49, + "prompt": "下面给出露点温度（dewpoint）=13.225845924995145 (摄氏度)\n温度（temperature）=13.427933720529904 (摄氏度)\n请计算湿度，计算公式为：\ndewpoint = dewpoint + 273.15，temperature = temperature + 273.15\ne = 611.2 * np.exp(17.67 * (dewpoint - 273.15) / (dewpoint - 29.65))\ne_s = 611.2 * np.exp(17.67 * (temperature - 273.15) / (temperature - 29.65))\nrelative humidity = e / e_s * 100\n以 \\boxed{relative humidity = ？%} 格式输出你的最终答案，例如 \\boxed{relative humidity = your answer%}。", + "output_len": 130, + "score": 0, + "ground_truth": { + "dewpoint": 13.225845924995145, + "temperature": 13.427933720529904, + "humidity": 98.69041013549538 + }, + "output": "首先，将露点温度和温度转换为绝对温度。露点温度为13.225845924995145摄氏度，加上273.15得到露点温度的绝对温度，即为286.37584592499516 K。同样地，温度13.427933720529904摄氏度加上273.15，得到绝对温度286.5779337205299 K。\n\n接下来，计算露点温度对应的饱和水蒸气压。使用公式e = 611.2 * np.exp(17.67 * (dewpoint - 273.15) / (dewpoint - 29.65))，其中dewpoint为286.37584592499516 K。计算得到e的值约为654.54 hPa。\n\n然后，计算温度对应的饱和水蒸气压。使用公式e_s = 611.2 * np.exp(17.67 * (temperature - 273.15) / (temperature - 29.65))，其中temperature为286.5779337205299 K。计算得到e_s的值约为656.53 hPa。\n\n最后，计算相对湿度。相对湿度等于e除以e_s，再乘以100，即(654.54 / 656.53) * 100 ≈ 99.69%。\n\n\n好的，我们按照给定的公式计算相对湿度。\n\n**步骤如下：**\n\n1. **将露点温度和温度转换为绝对温度：**\n \n \\[\n \\text{dewpoint} = 13.225845924995145\\,^\\circ\\text{C} + 273.15 = 286.37584592499516\\,\\text{K}\n \\]\n \n \\[\n \\text{temperature} = 13.427933720529904\\,^\\circ\\text{C} + 273.15 = 286.5779337205299\\,\\text{K}\n \\]\n\n2. **计算露点温度对应的饱和水蒸气压 \\( e \\)：**\n \n \\[\n e = 611.2 \\times \\exp\\left(\\frac{17.67 \\times (286.37584592499516 - 273.15)}{286.37584592499516 - 29.65}\\right) \\approx 654.54\\,\\text{hPa}\n \\]\n\n3. **计算温度对应的饱和水蒸气压 \\( e_s \\)：**\n \n \\[\n e_s = 611.2 \\times \\exp\\left(\\frac{17.67 \\times (286.5779337205299 - 273.15)}{286.5779337205299 - 29.65}\\right) \\approx 656.53\\,\\text{hPa}\n \\]\n\n4. **计算相对湿度：**\n \n \\[\n \\text{relative humidity} = \\left( \\frac{e}{e_s} \\right) \\times 100 = \\left( \\frac{654.54}{656.53} \\right) \\times 100 \\approx 99.69\\%\n \\]\n\n**最终答案：**\n\n\\[\n\\boxed{\\text{relative humidity} = 99.69\\%}\n\\]" +} + print(Earthdew2humiditybootcamp.extract_output(case['output'])) \ No newline at end of file diff --git a/internbootcamp/bootcamp/linearODE/linearODE.py b/internbootcamp/bootcamp/linearODE/linearODE.py index 4afff31..4eaccd3 100644 --- a/internbootcamp/bootcamp/linearODE/linearODE.py +++ b/internbootcamp/bootcamp/linearODE/linearODE.py @@ -5,7 +5,7 @@ from scipy.integrate import odeint from internbootcamp.bootcamp.base import Basebootcamp -class LinearODEBootcamp(Basebootcamp): +class LinearODEbootcamp(Basebootcamp): def __init__( self, k_range=(0.1, 1.0), @@ -38,41 +38,70 @@ class LinearODEBootcamp(Basebootcamp): return ( f"下面给出变量 x(t) 的观测数据点：\n{points}\n\n" "请找出其满足的微分方程，形式为：dx/dt = f(x)。\n" - "只需返回 “dx/dt = <表达式>”。" + "以dx/dt = <表达式>格式表示你的答案。" + "并且使用[answer]标签包裹你的最终答案, 例如[answer]dx/dt = <表达式>[/answer]." ) @staticmethod - def extract_output(output: str) -> str: - # 用正则提取“dx/dt = …”右侧的表达式 - m = re.search(r"dx/dt\s*=\s*([^\n\r]+)", output) - return m.group(1).strip() if m else None + def extract_output(output): + import re + matches = re.findall(r'\[answer\](.*?)\[/answer\]', output, re.DOTALL) + if not matches: + return None + last_match = matches[-1].strip() + try: + raw_expr = last_match.replace('dx/dt = ', '').strip() + expr = raw_expr.strip() + pattern = re.fullmatch( + r""" + ([+-]?\s* # 可选的正负号，后可带空格 + (?:\d+(?:\.\d*)? # 整数或小数点后数字 + |\.\d+)? # 或只有小数部分 + (?:[eE][+-]?\d+)? # 可选的科学计数部分 + )? # 整个系数是可选的（允许直接 x 或 -x） + \s*\*?\s* # 可选乘号，前后允许空格 + [xX] # x 或 X + """, + expr, + re.VERBOSE + ) + + if pattern: + raw = pattern.group(1) + if raw is None or raw.strip() == '': + return 1.0 + elif raw.strip() in ['+', '+1']: + return 1.0 + elif raw.strip() in ['-', '-1']: + return -1.0 + else: + return float(raw) + else: + return None + except ValueError: + return None @classmethod - def _verify_correction(cls, solution: str, identity: dict) -> bool: - # 解析 LLM 给出的系数 c，形如 “c*x” - sol = solution.replace(" ", "") - match = re.fullmatch(r"([\-0-9\.eE]+)\*x", sol) - if not match: - return False - c = float(match.group(1)) - # 验证 c ≈ -k - return abs(c + identity["k"]) < 1e-2 + def _verify_correction(cls, solution: float, identity: dict) -> bool: + delta = abs(solution + identity["k"]) + return delta < 1e-2 if __name__ == "__main__": - bootcamp = LinearODEBootcamp(seed=123) + bootcamp = LinearODEbootcamp(seed=123) # 生成几个样例 examples = [bootcamp.case_generator() for _ in range(3)] for identity in examples: # 构造“模型”返回答案，模拟 LLM 的输出 coeff = -identity["k"] - sol = f"{coeff:.4f}*x" + sol = f"[answer]dx/dt = {coeff:.2f}*x[/answer]" # 调用 Basebootcamp 提供的 verify_score 接口进行验证 - score = bootcamp.verify_score(sol, identity, short_threshold=1e-2) + score = bootcamp.verify_score(sol, identity, short_penalty=False,format_penalty=False) # 打印结果 print(json.dumps({ "identity": identity, "solution": sol, + "extract_output": LinearODEbootcamp.extract_output(sol), "verify_score": score }, ensure_ascii=False, indent=2)) \ No newline at end of file diff --git a/internbootcamp/bootcamp/med_calculator/med_calculator.py b/internbootcamp/bootcamp/med_calculator/med_calculator.py index 887417c..21a4566 100755 --- a/internbootcamp/bootcamp/med_calculator/med_calculator.py +++ b/internbootcamp/bootcamp/med_calculator/med_calculator.py @@ -154,7 +154,7 @@ class Medcalculatorbootcamp(Basebootcamp): # 医学计算器类 return cases if __name__ == '__main__': - bootcamp = MedCalculatorbootcamp(seed=42) + bootcamp = Medcalculatorbootcamp(seed=42) identity = bootcamp.case_generator() print(f'{identity = }') diff --git a/setup.py b/setup.py index 033418b..497d81e 100755 --- a/setup.py +++ b/setup.py @@ -19,7 +19,8 @@ setuptools.setup( "transformers", "langdetect", "pympler", - "shortuuid" + "shortuuid", + "rdkit" ], package_data={