how/test_relation_analyzer.py

"""
测试 relation_analyzer 模块
"""
import asyncio
from lib.relation_analyzer import analyze_relation


async def test_all_relations():
    """测试所有7种关系类型"""

    # 测试用例：每种关系类型的典型例子
    test_cases = [
        # 1. same（同义）
        {
            "phrase_a": "医生",
            "phrase_b": "大夫",
            "expected_relation": "same",
            "description": "完全同义"
        },
        {
            "phrase_a": "计算机",
            "phrase_b": "电脑",
            "expected_relation": "same",
            "description": "完全同义"
        },

        # 2. coordinate（同级）
        {
            "phrase_a": "轿车",
            "phrase_b": "SUV",
            "expected_relation": "coordinate",
            "description": "都是汽车的子类"
        },
        {
            "phrase_a": "苹果",
            "phrase_b": "香蕉",
            "expected_relation": "coordinate",
            "description": "都是水果"
        },

        # 3. contains（包含）
        {
            "phrase_a": "水果",
            "phrase_b": "苹果",
            "expected_relation": "contains",
            "description": "水果包含苹果"
        },
        {
            "phrase_a": "汽车",
            "phrase_b": "轿车",
            "expected_relation": "contains",
            "description": "汽车包含轿车"
        },

        # 4. contained_by（被包含）
        {
            "phrase_a": "苹果",
            "phrase_b": "水果",
            "expected_relation": "contained_by",
            "description": "苹果被水果包含"
        },
        {
            "phrase_a": "轿车",
            "phrase_b": "交通工具",
            "expected_relation": "contained_by",
            "description": "轿车被交通工具包含"
        },

        # 5. overlap（部分重叠）
        {
            "phrase_a": "红苹果",
            "phrase_b": "大苹果",
            "expected_relation": "overlap",
            "description": "有交集（又红又大的苹果）"
        },
        {
            "phrase_a": "学生",
            "phrase_b": "运动员",
            "expected_relation": "overlap",
            "description": "有交集（学生运动员）"
        },

        # 6. related（相关）
        {
            "phrase_a": "医生",
            "phrase_b": "医院",
            "expected_relation": "related",
            "description": "工作场所关系"
        },
        {
            "phrase_a": "阅读",
            "phrase_b": "书籍",
            "expected_relation": "related",
            "description": "动作-对象关系"
        },

        # 7. unrelated（无关）
        {
            "phrase_a": "医生",
            "phrase_b": "石头",
            "expected_relation": "unrelated",
            "description": "完全无关"
        },
        {
            "phrase_a": "苹果",
            "phrase_b": "数学",
            "expected_relation": "unrelated",
            "description": "完全无关"
        },
    ]

    # 模型选择（根据你的配置调整）
    model_name = "google/gemini-2.5-flash"  # 默认模型

    print(f"=" * 80)
    print(f"开始测试 relation_analyzer 模块")
    print(f"使用模型: {model_name}")
    print(f"测试用例数量: {len(test_cases)}")
    print(f"=" * 80)
    print()

    results = []

    for i, test_case in enumerate(test_cases, 1):
        phrase_a = test_case["phrase_a"]
        phrase_b = test_case["phrase_b"]
        expected = test_case["expected_relation"]
        description = test_case["description"]

        print(f"[{i}/{len(test_cases)}] 测试: \"{phrase_a}\" <-> \"{phrase_b}\"")
        print(f"     说明: {description}")
        print(f"     期望关系: {expected}")

        # 调用分析函数
        result = await analyze_relation(
            phrase_a=phrase_a,
            phrase_b=phrase_b,
            model_name=model_name
        )

        relation = result.get("relation", "unknown")
        score = result.get("score", 0.0)
        explanation = result.get("explanation", "")

        # 判断是否符合预期
        is_correct = (relation == expected)
        status = "✓" if is_correct else "✗"

        print(f"     实际关系: {relation} (score: {score:.2f}) {status}")
        print(f"     解释: {explanation}")
        print()

        results.append({
            "test_case": test_case,
            "result": result,
            "is_correct": is_correct
        })

    # 统计结果
    correct_count = sum(1 for r in results if r["is_correct"])
    total_count = len(results)
    accuracy = correct_count / total_count * 100

    print(f"=" * 80)
    print(f"测试完成")
    print(f"正确: {correct_count}/{total_count} ({accuracy:.1f}%)")
    print(f"=" * 80)

    # 显示错误的测试用例
    errors = [r for r in results if not r["is_correct"]]
    if errors:
        print()
        print("错误的测试用例：")
        for error in errors:
            tc = error["test_case"]
            result = error["result"]
            print(f"  - \"{tc['phrase_a']}\" <-> \"{tc['phrase_b']}\"")
            print(f"    期望: {tc['expected_relation']}, 实际: {result['relation']}")

    return results


async def test_single_example():
    """测试单个例子"""

    print("测试单个例子：")
    print()

    result = await analyze_relation(
        phrase_a="水果",
        phrase_b="苹果",
        model_name="google/gemini-2.5-flash"  # 默认模型
    )

    print(f"短语A: 水果")
    print(f"短语B: 苹果")
    print(f"关系: {result['relation']}")
    print(f"分数: {result['score']}")
    print(f"解释: {result['explanation']}")


if __name__ == "__main__":
    # 选择测试方式：

    # 方式1：测试单个例子（快速验证）
    # asyncio.run(test_single_example())

    # 方式2：测试所有关系类型（完整测试）
    asyncio.run(test_all_relations())