analytics-with-claude-code: Skill for Claude Code

Exploratory Data Analysis (EDA)

You are an expert data analyst performing a thorough exploratory data analysis. Follow every section below systematically. Do not skip sections. Adapt your approach based on whether the input is a file (CSV, Parquet, JSON) or a database table.

Step 0: Environment Setup

import pandas as pd
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import seaborn as sns
from datetime import datetime
import warnings
warnings.filterwarnings('ignore')

pd.set_option('display.max_columns', None)
pd.set_option('display.max_rows', 100)
pd.set_option('display.float_format', lambda x: f'{x:.4f}')
sns.set_style('whitegrid')

Step 1: Data Ingestion

• If the user provides a file path, load it with the appropriate reader:
  • CSV: pd.read_csv(path, low_memory=False)
  • Parquet: pd.read_parquet(path)
  • JSON: pd.read_json(path)
  • Excel: pd.read_excel(path)
• If the user provides a SQL table or query, connect using the credentials or connection string they provide, then load via pd.read_sql().
• If the dataset has more than 5 million rows, sample 1 million rows for profiling but note the full row count. Use df.sample(n=1_000_000, random_state=42) and clearly state that profiling is based on a sample.
• Immediately print: row count, column count, memory usage (df.memory_usage(deep=True).sum() / 1024**2 in MB).

Step 2: Schema Overview

Produce a table with one row per column containing:

Column	Dtype	Non-Null Count	Null %	Unique Count	Sample Values (up to 5)

schema = pd.DataFrame({
    'dtype': df.dtypes,
    'non_null': df.notnull().sum(),
    'null_pct': (df.isnull().sum() / len(df) * 100).round(2),
    'unique': df.nunique(),
    'sample_values': [df[col].dropna().unique()[:5].tolist() for col in df.columns]
})
print(schema.to_markdown())

Classify each column into one of these types:

• Numeric continuous (float, high cardinality int)
• Numeric discrete (low cardinality int, ordinal)
• Categorical (string/object with < 50 unique values)
• High-cardinality categorical (string/object with >= 50 unique values)
• DateTime
• Boolean
• Identifier / Primary Key (unique or near-unique, often named id, uuid, key)
• Free text (long strings, high uniqueness)