Tableau vs Old Tools: What Changed for Python Developers

# The classic approach: a static chart saved as an image
# pip install pandas matplotlib
import matplotlib
matplotlib.use("Agg")            # draw without a screen
import matplotlib.pyplot as plt
import pandas as pd

sales = pd.DataFrame({
    "month": ["Jul", "Aug", "Sep", "Oct"],
    "north": [120, 135, 128, 150],
    "south": [100, 98, 115, 122],
})

ax = sales.plot(x="month", y=["north", "south"], marker="o", title="Monthly sales")
ax.set_ylabel("Units")
plt.tight_layout()
plt.savefig("monthly_sales.png")
print("Saved monthly_sales.png - next change request means editing this script")

The old request cycle

Day 1  Manager: "Show sales by month"
Day 1  Developer writes script, emails PNG
Day 2  Manager: "Can I see only the North region?"
Day 2  Developer edits script, emails again
Day 3  Manager: "And compare with last year?"
Day 4  Developer edits script, emails again

Result: developer time spent on small changes,
        manager waits for every new view.

# Python's new main job: clean, tidy data for Tableau
# pip install pandas
import pandas as pd

raw = pd.DataFrame({
    "Month": ["Jul", "Aug", "Sep", "Oct"],
    "North": [120, 135, 128, 150],
    "South": [100, 98, None, 122],
})

# Tableau works best with long ("tidy") data: one row per fact
tidy = raw.melt(id_vars="Month", var_name="Region", value_name="Units")
tidy["Units"] = tidy["Units"].fillna(0).astype(int)

# Basic data checks before publishing
assert tidy["Units"].ge(0).all(), "negative units found"
assert tidy["Region"].notna().all(), "missing region"

tidy.to_csv("sales_for_tableau.csv", index=False)
print(tidy.head())

Why Tableau prefers tidy (long) data

Wide (hard to use):
Month  North  South
Jul    120    100

Tidy (easy to use):
Month  Region  Units
Jul    North   120
Jul    South   100

With tidy data, users can:
- put Region on Color
- filter by Region
- add new regions without changing the dashboard

// Tableau calculated field using TabPy (TabPy server must be
// configured under Help > Settings and Performance >
// Manage Analytics Extension Connection)

// Add 18% tax to each mark, computed in Python
SCRIPT_REAL("
return [x * 1.18 for x in _arg1]
", SUM([Sales]))

// _arg1 receives the aggregated SUM([Sales]) values as a list.
// The Python code must return a list of the same length.

Modern workflow: Python + Tableau

1. Python pipeline extracts and cleans data (scheduled)
2. Automated checks: row counts, nulls, ranges
3. Clean tables written to the database
4. Tableau extract refreshes after checks pass
5. Analysts and managers explore dashboards
6. New business questions -> new fields or tables
   added upstream by the Python developer

Developer focus: correctness, freshness, performance

Data Handover Spec (Python -> Tableau)

Table / file name: _______________
Grain (one row = ?): _______________
Columns:
  name      type     description
  Month     date     first day of month
  Region    text     sales region
  Units     integer  units sold, never negative
Refresh: daily at ____
Owner: _______________
Known limitations: _______________

Before Each Refresh

[ ] Pipeline finished without errors
[ ] Row count within expected range
[ ] No nulls in key columns
[ ] No duplicate keys
[ ] Latest date is today / yesterday
[ ] Values in sensible ranges
[ ] Only then trigger the Tableau refresh

Dashboard Request Log

Date   Requester   Question                    Needs new data?  Status
____   Sales       Compare with last year      Y (history)      ____
____   Ops         Split by warehouse          N (filter)       ____
____   Finance     Margin by product           Y (cost data)    ____