Analyzing Statistics

Statistical Analysis & Data Science Mastery

This skill provides comprehensive agentic guidance for industrial-grade data science, statistical modeling, and automated insight generation within the Antigravity ecosystem.

Process

Follow these procedures to implement the capability:

When to Use

This skill should be used when completing tasks related to analyzing statistics.

🔬 Statistical Foundations

1. Correct Application of Statistics

• Descriptive: Use Pandas and DataManager.get_summary_stats for initial posture analysis (Mean vs. Median for skewed warehouse data).
• Inference: Apply Hypothesis Testing (T-tests, Chi-square) to verify if operation changes (e.g., new conveyor speed) are statistically significant using AnalysisManager.run_hypothesis_test.
• Distributions: Detect and handle non-normal distributions using Log/Box-Cox transformations via NumPy before spectral or linear modeling.

2. Outlier & Anomaly Management

• Use IQR (Interquartile Range) for robust outlier detection in noisy IoT streams.
• Use Z-score for detecting deviations in standardized performance metrics (e.g., UPH variance).

🤖 Data Science & Machine Learning

1. Predictive Modeling (Prognosis)

• Linear/Logistic Regression: Use for trend forecasting and probability modeling (e.g., probability of an SLA breach).
• Clustering (K-Means): Group warehouse bins or associates into "Performance Tiers" for targeted coaching or slotting optimization.
• Validation: Always report $R^2$, MSE, and P-values to ensure model reliability.

2. Tool Integration Patterns

• Pandas: Prioritize vectorized operations (df.apply, df.groupby) to handle building-scale datasets efficiently.
• Visuals:
  • Matplotlib: Deep-dive static plots for publication/reports.
  • Plotly/Altair: Interactive drill-down dashboards in Streamlit (e.g., Bin Density Heatmaps).

🧠 LLM-Augmented Analysis

• Narrative Generation: Extract statistical artifacts (coefficients, cluster centroids) and feed them to LLM APIs (OpenAI/Anthropic) to produce management-ready summaries.
• Automated Interpretation: Use LLMs to explain why a correlation might exist based on the business domain (e.g., "The correlation between shelf height and UPH suggests ergonomic fatigue").

Best Practices

• Workflow First: Start with /eda, move to /strategy-development or /warehouse-analytics, and finalize with an Executive Summary.
• Grounding: Ensure all AI-generated insights are anchored to specific calculated metrics ($R^2$, P-value) to avoid hallucinated trends.
• Reproducibility: Save all models and data snapshots to the Project database to allow analysts to audit the decision logic.

Prerequisites

• Access to relevant project documentation
• Environmental awareness of the target stack