Automating Data Analysis with Zene

Data cleaning is 80% of a data scientist's job. Zene can automate this tedious process, from loading messy CSVs to generating insightful visualizations.

In this example, we provide Zene with a "dirty" sales dataset containing missing values, invalid dates, and inconsistent formatting. Zene autonomously cleans the data and produces a summary chart.

The Task

task = """
I have a CSV file `sales_data.csv` with some messy data.
Please help me:
1. Load the data using pandas.
2. Clean the data: remove rows with missing values or invalid dates.
3. Convert the 'date' column to datetime objects.
4. Calculate the total sales amount per product.
5. Generate a bar chart of total sales per product using matplotlib and save it as `sales_chart.png`.
6. Save the cleaned data to `cleaned_sales_data.csv`.
"""

Execution Process

1. Planning

Zene's Planner (DeepSeek) breaks this down into a logical sequence:

1. Read & Inspect: Load sales_data.csv and print the first few rows to understand the structure.
2. Clean: Filter out rows where amount is null or date cannot be parsed.
3. Transform: Convert the date column to datetime objects.
4. Analyze: Group by product and sum the amount.
5. Visualize: Use matplotlib to plot the aggregated data.
6. Save: Export the clean DataFrame and the plot image.

2. Execution & Reflection

• Step 1 (Load): The Executor runs a python script to load the CSV. The Reflector checks if the file exists and is readable.
• Step 2 (Clean): The Executor writes pandas code to drop NA values. It encounters a ValueError when parsing "invalid_date".
• Self-Healing: The Reflector catches the ValueError. It suggests using pd.to_datetime(..., errors='coerce') to handle invalid formats gracefully. The Executor applies this fix automatically.
• Step 5 (Visualize): Zene generates a bar chart. The Reflector verifies that sales_chart.png was created.

3. Handling Dependency Hell

What if the user's environment doesn't have pandas installed?

• Error: The initial script fails with ModuleNotFoundError: No module named 'pandas'.
• Reflect: The Reflector identifies the missing package.
• Fix: It plans a new step: "Install pandas using pip".
• Execute: The Executor runs pip install pandas.
• Retry: The script runs successfully on the second attempt.

The Result

Input (sales_data.csv):

date,product,amount,region
2023-01-01,Widget A,100,North
2023-01-03,Widget A,,North        <-- Missing Amount
invalid_date,Widget B,300,West    <-- Invalid Date

Output (cleaned_sales_data.csv):

date,product,amount,region
2023-01-01,Widget A,100.0,North
2023-01-02,Widget B,200.0,South
2023-01-04,Widget C,150.0,East
2023-01-05,Widget A,120.0,North

Visualization: A clear bar chart showing total sales by product is generated automatically.

Key Takeaway

Zene doesn't just write code; it handles runtime data errors (like bad CSV formatting) by iterating on its own code until the pipeline runs successfully.