统计假设检验框架

# Clone the repository
git clone https://github.com/yourusername/mcp-hypothesis-tester.git
cd mcp-hypothesis-tester

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Generate example datasets
python generate_datasets.py

# Test the implementation
python test_implementation.py

# Start MCP server
python -m src.mcp_server

# Or use FastMCP CLI
fastmcp run

# Step 1: Auto-suggest appropriate test
result = suggest_test(
    dataset_name="sales_example",
    test_params={"mu0": 5000}
)
# Suggests: one_sample_t_test

# Step 2: Run the test
result = run_hypothesis_test(
    test_type="one_sample_t_test",
    dataset_name="sales_example",
    alpha=0.05,
    test_params={"mu0": 5000}
)

# Result:
{
  "summary": {
    "decision": "reject",
    "p_value": 0.0441,
    "significant": true,
    "interpretation": "The sample mean (5150.00) is statistically
                      significantly different from the hypothesized
                      value (5000). This represents a 3.0% change.",
    "effect_size": 0.309,
    "recommendation": "Evidence suggests a significant effect.
                      Investigate what drove this improvement!"
  },
  "cache_id": "abc123...",
  "note": "Full details cached. Use get_cached_result() to retrieve."
}

# Auto-suggest recognizes paired data
result = suggest_test("website_redesign")
# Suggests: paired_t_test (before/after comparison)

# Run test
result = run_hypothesis_test(
    test_type="paired_t_test",
    dataset_name="website_redesign"
)

# If significant:
# "Visit time increased from 5.0 to 5.5 minutes (p=0.02).
#  Analyze WHERE users spend extra time:
#  - Product pages? GOOD (considering purchases)
#  - Help pages? BAD (confused, frustrated)"

# 1. List available datasets
datasets = list_available_datasets()

# 2. Get test suggestion
suggestion = suggest_test("sales_example", {"mu0": 5000})
print(f"Recommended: {suggestion['test_name']}")
print(f"Reason: {suggestion['reason']}")

# 3. Check assumptions
assumptions = check_assumptions("sales_example", "one_sample_t_test")
if not assumptions['all_met']:
    print(f"Warnings: {assumptions['violations']}")
    print(f"Recommendation: {assumptions['recommendation']}")

# 4. Run test
result = run_hypothesis_test(
    suggestion['primary_test'],
    "sales_example",
    test_params=suggestion['parameters']
)

# 5. Get full details if needed
full_result = get_cached_result(result['cache_id'])

mcp-hypothesis-tester/
├── datasets/                    # 6 documentation-aligned datasets
│   ├── sales_example.json      # Case 1: One-sample t-test
│   ├── coffee_shop_traffic.json # Case 2: Fail to reject
│   ├── website_redesign.json   # Case 3: Paired t-test
│   ├── coffee_machine_ratings.json
│   ├── sales_three_stores.json # ANOVA example
│   └── normality_test_example.json
│
├── src/                         # Source code
│   ├── core/
│   │   ├── base_test.py        # Abstract base class
│   │   └── cache_manager.py    # Temp JSON caching
│   ├── tests/
│   │   ├── parametric/         # T-tests, ANOVA
│   │   ├── non_parametric/     # Mann-Whitney, Wilcoxon, etc.
│   │   ├── normality/          # Shapiro-Wilk, K-S, etc.
│   │   └── variance/           # Levene's, Bartlett's, F-test
│   ├── datasets/
│   │   └── loader.py           # Dataset management
│   ├── utils/
│   │   └── test_suggester.py  # Auto test suggestion
│   └── mcp_server.py           # FastMCP integration
│
├── Documentation/
│   ├── README.md               # This file
│   ├── USAGE_GUIDE.md          # Comprehensive usage guide
│   ├── IMPLEMENTATION-1.md     # Implementation plan
│   ├── IMPLEMENTATION_COMPLETE.md
│   ├── PROJECT_STRUCTURE.md    # Detailed structure
│   ├── LLM_USAGE_GUIDE.md      # For LLM users
│   ├── hypothesis-tester-for-megginners.md
│   └── sample-cases-and-explanation.md
│
├── Scripts/
│   ├── generate_datasets.py    # Generate example datasets
│   └── test_implementation.py  # Test suite
│
└── Configuration/
    ├── fastmcp.json
    ├── requirements.txt
    └── pyproject.toml

class BaseHypothesisTest(ABC):
    def run(self, *args):
        # 1. Validate input
        # 2. Check assumptions
        # 3. Compute statistic
        # 4. Calculate effect size
        # 5. Interpret result
        # 6. Cache details
        # 7. Return TestResult

@dataclass
class TestResult:
    test_name: str
    statistic: float
    p_value: float
    decision: str
    interpretation: str
    effect_size: float
    assumptions_met: Dict[str, bool]
    cache_id: str

python test_implementation.py

fastmcp >= 2.12.0      # MCP server framework
numpy >= 1.24.0        # Array operations
scipy >= 1.10.0        # Statistical tests
pandas >= 2.0.0        # Data handling
matplotlib >= 3.7.0    # Plotting (optional)
seaborn >= 0.12.0      # Visualization (optional)
statsmodels >= 0.14.0  # Advanced statistics