Reference
Pandas syntax sheet
Each piece of syntax next to what it returns. Grows with every lesson.
Select columns · L01
| Syntax | Returns |
|---|---|
df["col"] | Series (one column) |
df[["a", "b"]] | DataFrame (list of names → table) |
Masks & filtering · L01
| Syntax | Returns |
|---|---|
df["col"] > 100 | Boolean Series (mask). Comparing with NaN gives False. |
> >= < <= == != | Comparison operators |
df[mask] | DataFrame of the True rows, original labels kept |
df[mask][["a"]] | Filter, then select columns |
Combine conditions · L01
| Syntax | Means |
|---|---|
(a) & (b) | AND |
(a) | (b) | OR |
~(a) | NOT |
Always put parentheses around each condition. Use the symbols, not the words and/or/not.
Aggregate · L01
| Syntax | Returns |
|---|---|
.sum() .mean() .min() .max() | One number, skipping NaN |
.count() | The number of non-missing values |
len(df) | The number of rows, missing values included |
.fillna(v) | A new Series with the gaps filled |
Sort & locate · L02
| Syntax | Returns |
|---|---|
df.sort_values("col") | DataFrame sorted ascending, NaN last |
…, ascending=False | Largest first |
.head(n) | The first n rows |
.iloc[0] / .iloc[-1] | The first or last row by position (a Series) |
row["col"] | One value |
df.loc[label, "col"] | One value, looked up by row label |
s.idxmax() / s.idxmin() | The label where the max or min is |
Group · L02
| Syntax | Returns |
|---|---|
df.groupby("key")["col"].sum() | A Series indexed by key, sorted by key |
….mean() .count() .max() | Any aggregation, applied per group |
….sum().idxmax() | The key with the largest total |
Summarise · L03
| Syntax | Returns |
|---|---|
s.value_counts() | A Series of counts indexed by value, most common first, NaN left out |
…(normalize=True) / …(dropna=False) | Shares instead of counts / include NaN |
….groupby("k")["c"].agg(["mean", "count"]) | A DataFrame: one row per key, one column per stat |
df["new"] = df["c"] <= 2 | Creates a True/False column |
df["new"].mean() | The share of True (True = 1, False = 0) |
s.sort_values(ascending=False) | A Series sorted by its values (no column name) |
Dates · L04
| Syntax | Returns |
|---|---|
df["d"] = pd.to_datetime(df["d"]) | Text column → real dates (do this first) |
df["d"].dt.month / .dt.year / .dt.day_name() | A part of each date |
df[df["d"] >= "2026-03-01"] | Rows on or after a date |
df["d"].dt.to_period("M") | Each date as its month (2026-03) |
df.groupby(df["d"].dt.to_period("M"))["c"].mean() | A Series per month, in time order: the trend |
s.sort_index() | A Series sorted by its labels (e.g. months after value_counts) |
s.rolling(3).mean() | Rolling average over 3 values; the first 2 are NaN |
Charts · L05
| Syntax | Draws |
|---|---|
import matplotlib.pyplot as plt | matplotlib, the drawing library pandas uses |
s.plot() | Line chart: index along the bottom, values up the side (use for time) |
s.plot.bar() / s.plot(kind="bar") | Upright bars, one per label, in the Series' order (use for groups) |
s.plot.barh() | Horizontal bars; the first label is at the bottom |
s.sort_values(ascending=False).plot.bar() | Ranked bars: sort first, the chart keeps the order |
.plot(title="…", xlabel="…", ylabel="…", ylim=(0, 5)) | Title, axis names, fixed y-axis range |
plt.title("…") / plt.ylabel("…") / plt.ylim(0, 5) | The same, on a line after the chart |
a.plot(label="a") + b.plot(label="b") + plt.legend() | Two lines on one chart, with a key |
ax = a.plot() then b.plot(ax=ax) | Explicitly draw onto the same chart |
plt.savefig("chart.png", bbox_inches="tight") | Saves the chart as an image file |
Load a CSV, from disk or S3 · L06
| Syntax | Returns |
|---|---|
pd.read_csv("file.csv") | DataFrame: header line → column names, empty values → NaN |
pd.read_csv(…, nrows=5000) | Only the first 5,000 rows |
pd.read_csv(…, usecols=["a", "b"]) | Only those columns, in the file's order |
df.shape | (rows, columns): a property, no brackets |
df.head(3) | The first 3 rows (5 by default) |
df.to_csv("copy.csv", index=False) | Writes a CSV; index=False leaves out the row labels |
| From S3 with boto3 | Returns |
|---|---|
s3 = boto3.Session(profile_name="NAME").client("s3") | A client logged in with that AWS profile |
r = s3.list_objects_v2(Bucket="b", Prefix="exports/") | Dict: r["KeyCount"], and r["Contents"] (one dict per object: Key, Size). No "Contents" when nothing matches. |
pd.DataFrame(r["Contents"]) | The listing as a table |
obj = s3.get_object(Bucket="b", Key="exports/x.csv") | Dict; obj["Body"] is the contents as a stream (read once) |
pd.read_csv(obj["Body"], nrows=5000) | DataFrame from the object |
s3.download_file("b", "exports/x.csv", "x.csv") | Saves the object to a local file |
pd.read_csv("s3://b/key.csv", storage_options={"profile": "NAME"}) | You'll also see this: the same in one line, via the extra s3fs package |
| In a notebook | Does |
|---|---|
| Shift+Enter | Runs the cell; variables stay until you restart |
%pip install boto3 | Installs into the kernel's own Python |
!aws sts get-caller-identity --profile NAME | ! runs a terminal command; this one says who that profile logs you in as |
Check and clean a CSV · L07
| Syntax | Returns |
|---|---|
pd.read_csv("file.csv", na_values=["No topic"]) | That text comes in as NaN too |
df.dtypes / df["a"].dtype | The type pandas guessed, per column / for one column |
df.info() | Prints each column's dtype and non-missing count |
df.isna().sum() / df.isna().mean() | Series: missing count / missing share, per column |
s.value_counts(dropna=False) | Counts per value, with a row for NaN |
df.dropna(subset=["a"]) | DataFrame without the rows where a is missing (no subset: any column) |
(df["topic"] == "No topic").sum() | Count a "nothing" written as text: isna() can't see it |
s.astype(int) | Whole numbers; fails if s has NaN |
s.str.lower() | Lower-case text, so spellings group together |
df.drop(columns=["a"]) | DataFrame without column a (without columns= it looks for a row) |
| In the file | dtype |
|---|---|
| text, and dates | str |
5.0, and a column empty on every row | float64 |
| True/False with gaps | object (no gaps: bool) |
Each fix returns something new: assign it back (df["a"] = …, df = df.drop(…)) to keep it.
Reading the API docs · L05
| Part of a page | What it tells you |
|---|---|
| API reference (pandas 3.0) | Every method, one page each; search for the method you're calling |
| Parameters | Every keyword argument: its name, the type it accepts, and its default |
**kwargs | Extra arguments passed on to another function: follow its link for the rest |
| Returns | What you get back (Series, DataFrame, Axes…) |
| See also / Examples | Related methods, and short runnable examples |
Checked against pandas 3.0.2, the version the sandbox runs · Python for Data Analysis, 3e · Pandas gotchas