AgentSkillsCN

performing-eda

制定 Markdown 文件的规范与规则，确保格式统一、可读性高、渲染准确。

SKILL.md
--- frontmatter
name: performing-eda
description: Conducts Exploratory Data Analysis (EDA) on datasets. Use when the user asks to "explore", "clean", or "visualize" a new CSV or dataset.

Performing EDA (Exploratory Data Analysis)

When to use this skill

  • •User uploads a .csv or .json file and asks "what's in here?".
  • •User wants to "check for missing values" or "see distributions".
  • •Debugging model performance by analyzing training data.

Workflow

  • • Load: Read file into Pandas DataFrame.
  • • Inspect Structure: df.info(), df.head(), df.describe().
  • • Clean: Handle missing values (NaN), duplicates, and incorrect types.
  • • Univariate Analysis: Histograms/Boxplots for single variables.
  • • Bivariate Analysis: Correlation matrix, Scatter plots for relationships.
  • • Report: Summarize findings (Outliers, Trends, Data Quality).

Instructions

1. Standard Inspection Script (Python)

Use the model environment or a temporary script.

python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

def quick_eda(filepath):
    df = pd.read_csv(filepath)
    print("--- INFO ---")
    print(df.info())
    print("\n--- DESCRIBE ---")
    print(df.describe())
    
    # Check nulls
    nulls = df.isnull().sum()
    if nulls.sum() > 0:
        print("\n--- NULLS ---")
        print(nulls[nulls > 0])
        
    return df

2. Visualization Standards

  • •Use Seaborn for statistical plots (nicer defaults than matplotlib).
  • •Correlation Heatmap: Critical for finding redundant features.
  • •Pairplot: Useful for small feature sets (< 10 features).

3. Notebooks vs Scripts

  • •Only create .ipynb files if the user explicitly asks for a notebook or "interactive exploration".
  • •Otherwise, write a .py script that outputs text summaries and saves plot images to a plots/ folder.

Resources