Part 4 of 12

NumPy and pandas Basics with a Bot: Cheat Sheet

Cheat sheetBy Bot Tutorials · Published Nov 3, 2026 · Facts last verified Nov 3, 2026

ML libraries series complete. "Review Your Training Code with a Bot: Checklist" was the finale. See every part in the series hub.

Part 4: shapes, dtypes, broadcasting, missing values and groupby, with real output from NumPy 2.5 and pandas 3.1, and prompts that keep a bot honest.

Video: NumPy and pandas Basics with a Bot: Cheat Sheet (4:00). Narrated from this guide, with the real code and real output we ran. Captions included (caption file). Prefer reading? Everything in the video is in the guide below.

Almost every machine learning library sits on top of NumPy arrays, and most real data arrives as a pandas DataFrame. You don't need to know everything about them, but you do need a handful of ideas: shapes, dtypes, broadcasting, missing values and grouping. This cheat sheet covers those with a short script and its real output, and shows how to get a bot to explain your own data without making things up.

What the bot can and can't do here

It can:

Explain why two arrays won't combine, given their shapes.
Translate "I want the average price per part" into a pandas line.
Read a traceback and find the column or shape at fault.

It can't:

See your DataFrame. Paste df.head(), df.dtypes and df.shape, not a description.
Remember which defaults changed between versions. pandas 3 changed some, as the dtype line below shows.
Decide what a missing value means in your data. That's your call.

Check first, transform second

Before you change data, look at it: shape, dtypes and missing counts. Most bugs are a wrong assumption about one of those three.

The script

import numpy as np, pandas as pd
rng = np.random.default_rng(42)
a = rng.normal(size=(3, 4))
print("shape", a.shape, "dtype", a.dtype)
print("column means", np.round(a.mean(axis=0), 3))
b = np.arange(4)
print("broadcast add row:", (a + b).shape)
df = pd.DataFrame({"part": ["gpu", "cpu", "ram", "gpu"], "price": [499.0, 219.0, None, 649.0]})
print(df.dtypes.to_dict())
print("missing prices:", int(df["price"].isna().sum()))
print(df.groupby("part")["price"].mean().round(1).to_dict())
print(df.dropna().query("price > 300")[["part", "price"]].to_string(index=False))

Real output on our box with NumPy 2.5.4 and pandas 3.1.0:

shape (3, 4) dtype float64
column means [-0.554 -1.065  0.586  0.467]
broadcast add row: (3, 4)
{'part': <StringDtype(storage='python', na_value=nan)>, 'price': dtype('float64')}
missing prices: 1
{'cpu': 219.0, 'gpu': 574.0, 'ram': nan}
part  price
 gpu  499.0
 gpu  649.0

NumPy in five lines

Shape and dtype first. a.shape is (3, 4): three rows, four columns. a.dtype is float64, NumPy's default for these random numbers.
Axis means direction. a.mean(axis=0) averages down the rows and gives one number per column, so four numbers.
Broadcasting. Adding b with shape (4,) to a with shape (3, 4) works because NumPy stretches b across each row. The result keeps shape (3, 4). If the last dimensions don't match, you get an error instead.
Random numbers with a Generator. NumPy's random docs recommend np.random.default_rng(seed) for new code. A fixed seed makes the numbers repeatable on the same version.
Vectorize. Whole-array operations are usually much faster than Python loops. Part 11 measures this with a profiler.

pandas in five lines

Check dtypes. On our box the part column came back as a pandas StringDtype, not the older object type, because we're on pandas 3.1. That's a good example of why bots that learned on older versions can be wrong about defaults.
Count missing values. df["price"].isna().sum() gave 1.
Group and summarize. groupby("part")["price"].mean() gave 574.0 for gpu, 219.0 for cpu and nan for ram, because its only price was missing.
Drop or fill on purpose. dropna() removed the ram row. Whether that's right depends on your data.
Filter with query. query("price > 300") kept the two GPU rows.

Prompts that work

Here are df.shape, df.dtypes and df.head(10) (pasted). Write pandas code to get the mean of price for each part. Don't assume columns I haven't shown. Tell me what happens to missing prices.

These two arrays have shapes (3, 4) and (3,). Why does adding them fail, and what are two ways to fix it? Show the resulting shapes.

The second prompt is a good test of a bot. The right answer explains that broadcasting aligns trailing dimensions, so (3,) doesn't line up with the 4 columns, and suggests reshaping to (3, 1) or using the matching axis.

Common mistakes to watch for

Chained assignment. Code like df[df.x > 0]["y"] = 1 may not change df. Use df.loc[mask, "y"] = 1. The pandas user guide explains indexing with loc.
Silent float conversion. An integer column with a missing value becomes float in many cases. Check dtypes after loading.
Wrong axis. If a mean gives you the wrong number of values, the axis is probably swapped.
Copying data into a bot. A sample of rows is usually enough. Don't paste private data.

Words you'll see

ndarray: NumPy's array type. Every element has the same dtype.
dtype: the data type, such as float64, int64 or a string type.
axis: the direction an operation runs. Axis 0 is down the rows, axis 1 across the columns.
Series and DataFrame: pandas' one-column and table types. Each column of a DataFrame is a Series.
NaN: "not a number", the usual marker for a missing float value.

Knowing these words makes bot answers easier to check, because you can ask exactly what you mean.

Tested vs documented

Tested by us: every line of the script and every output shown, on NumPy 2.5.4 and pandas 3.1.0.
Documented, not tested by us: the chained-assignment advice and the Generator recommendation, from the pandas and NumPy docs.

Where to go next

The next part today uses these arrays to train a first scikit-learn model.

Safety, licensing and privacy: read this before you run anything

This series is about learning and small personal projects with machine learning libraries. A bot is a helper, not an authority.

Check bot advice against official docs. Version numbers, install commands and compatibility tables change often. Compare every answer with the PyTorch, TensorFlow, JAX, scikit-learn, NVIDIA, Hugging Face or ONNX Runtime pages we link. If the bot and the docs disagree, the docs win.
Work in a virtual environment. Don't install or remove packages in your system Python, and don't run sudo pip.
Run untrusted files carefully. Pickle-based model files can run code when loaded. Only load models from sources you trust, and prefer safer formats like safetensors where the library supports them.
Respect licenses. Datasets and model weights carry their own licenses and terms. Read the model card and license before you use, share or sell anything built on them. We don't give legal advice; ask a qualified person if it matters.
Protect personal data. Don't paste private, customer or medical data into a bot or a training set unless you have the right to, and check the bot's own privacy settings first.
No hardware claims you didn't measure. Treat any speed figure from a bot as a guess until you've timed it on your own machine.
No medical, legal or financial decisions from a toy model. These guides are for learning only.

FAQ

Why does the text column show a StringDtype?

That's what pandas 3.1 reported on our box for a column of strings. Older pandas versions showed object. If a bot tells you to expect object, check your version.

Should I use np.random.seed?

NumPy's docs recommend the Generator interface, created with np.random.default_rng, for new code. That's what our example uses.

Was this tested?

Yes. Every output on this page came from running the script on our box.