pandas .map vs .apply vs .applymap — Pick the Right One

Three pandas methods. Three different jobs. And one of them was removed in pandas 3.0 — which is why your copy-pasted Stack Overflow snippet just threw AttributeError. This tutorial walks through .map, .apply, and .applymap on a small prices DataFrame and shows exactly which method belongs where. Source code: https://github.com/GoCelesteAI/apply-vs-map-vs-applymap .map is Series-only. It takes a dict, a Series, or a function — and that dict overload is its superpower. If you're translating ticker symbols to sector names, country codes to country names, or any value-to-value lookup, .map is the one-liner that does it. .apply works on both Series and DataFrames, but always takes a function. On a Series it's basically a slower .map. On a DataFrame it's something else entirely — axis=0 hands your function each column as a Series, axis=1 hands it each row as a Series. That's how you compute per-column standard deviation in one line, or per-row max in another. .applymap was the old element-wise method — it ran your function on every cell of a DataFrame. As of pandas 2.1 it's deprecated, and in pandas 3.0 it's gone. The replacement is DataFrame.map, which works exactly the same way. What You'll Build: - pandas_methods.py — a single file that demonstrates all four cases (Series.map, Series.apply, DataFrame.apply axis=0 + axis=1, DataFrame.map) on a 3x3 prices DataFrame. Runs in two seconds. - The Series.map dict pattern — pass a dict of ticker→sector mappings and pandas does the lookup for every value at once. No loop. No merge. - The Series.apply function pattern — same shape, but takes a function. Useful when the transformation needs logic the dict can't express. - DataFrame.apply axis=0 — your function receives each column as a Series. Perfect for per-column reductions like std, mean, custom z-scores. - DataFrame.apply axis=1 — your function receives each row as a Series. The row-wise tool you reach for when no column expression works. - DataFrame.map element-wise — the modern, supported replacement for applymap. Format numbers, cast types, sanitize strings — anything per-cell. - The applymap gotcha — the demo tries to call df.applymap on pandas 3.0 and catches the AttributeError. If you're seeing that error, the fix is one rename: applymap → map. Timestamps: 0:00 - Intro — three methods, one decision tree 0:22 - Preview — when each one is the right tool 1:07 - Open pandas_methods.py in nvim 1:30 - Series.map — dict lookup, the killer feature 1:55 - Series.apply — function on a Series 2:15 - DataFrame.apply axis=0 — per-column 2:40 - DataFrame.apply axis=1 — per-row 3:05 - DataFrame.map — the new element-wise 3:30 - applymap — removed in pandas 3.0 3:50 - Run it 4:05 - End screen — recap Key Takeaways: 1. .map is Series-only — and the dict overload is what makes it special. tickers.map({"AAPL": "Tech", "MSFT": "Tech"}) does a vectorized lookup in one line. If you find yourself writing a for loop to translate values, .map is the fix. It also accepts functions, but the dict and Series overloads are why you reach for it first. 2. .apply takes a function and works on both Series and DataFrames. On a Series it's a slightly slower .map. On a DataFrame the axis argument is everything — axis=0 (the default) hands your function each column, axis=1 hands it each row. Per-column reductions go to axis=0, per-row computations go to axis=1. 3. .applymap was the element-wise method for DataFrames. It was deprecated in pandas 2.1 and REMOVED in pandas 3.0. If your code throws AttributeError: 'DataFrame' object has no attribute 'applymap', the fix is a one-line rename. df.applymap(f) → df.map(f). Same signature, same behavior, just the new name. 4. The decision tree is short. Series + dict lookup → .map. Series + function → .map (or .apply). DataFrame element-wise → .map. DataFrame per-column → .apply axis=0. DataFrame per-row → .apply axis=1. Anything that fits a vectorized numpy or pandas operation, do that instead — apply is still a Python loop in disguise. 5. Speed: .map is fastest on Series (especially with a dict), .apply on a DataFrame with axis=1 is the slowest of the bunch — it builds a Series for every row, just like iterrows. When the row computation can be rewritten as column arithmetic (df["A"] + df["B"]), the vectorized form is hundreds of times faster. 📺 Deeper dive: Pandas for Finance Ep 6 builds per-column returns and Sharpe with vectorized arithmetic — the case where you skip .apply entirely. https://www.youtube.com/watch?v=MNGU8wCzIQw This channel is run by Claude AI. Tutorials AI-produced; reviewed and published by Codegiz. Source code at codegiz.com. #Python #Pandas #DataAnalytics #PythonTutorial #LearnPython #DataFrame #pandas3 #ApplyVsMap --- Generated by Claude AI · part of the Common Questions in Python series