pandas

pandas is LINQ + DataTables, rebuilt for the real world. If you know SQL, you already know 70%.

▶ Watch this reel

What you'll learn

DataFrame & Series
Select, filter, sort
groupby & aggregate
merge / join
Missing data & dtypes
Memory & performance

Remember this

DataFrame = table, Series = column; boolean masks = WHERE; loc = SELECT…WHERE
groupby.agg = SQL GROUP BY; merge needs explicit on/how/validate — trust nothing
isna first, fillna as a documented claim, category dtype + vectorization for scale

DataFrame & Series

DataFrame = labeled table; Series = one column. Built on NumPy → vectorized.
First moves: read_* → .shape / .dtypes / .isna().sum().

Select / filter / sort

SQL	pandas
WHERE	`df[df["x"] > 1]` (boolean mask; `&` `	` need parens)
SELECT cols	`df.loc[mask, ["a", "b"]]`
ORDER BY … LIMIT	`.sort_values("x", ascending=False).head(10)`

groupby & aggregate

df.groupby("k")["v"].mean() · .agg(total=("v","sum"), n=("id","count")).

merge / join

df.merge(other, on="key", how="left", validate="m:1", indicator=True).
Defaults differ from SQL: inner join + shared-column keys → always be explicit.
Duplicated keys silently multiply rows → validate + row-count checks.

Missing data & dtypes

isna → know the holes · fillna = a documented claim · dropna.
pd.to_datetime / pd.to_numeric(errors="coerce") early; codify dtypes in a schema.

Memory & performance

astype("category") for low-cardinality strings (~10x).
Avoid .apply(f) — vectorize with masks / np.where / merges.
Bigger than RAM: read_csv(chunksize=…) — or Parquet + DuckDB (next reels).

Code: pandas for the SQL-brained — one realistic pipeline

import pandas as pd

# --- load + orient ------------------------------------------------
df = pd.read_csv("tickets.csv")
print(df.shape, df.dtypes, df.isna().sum(), sep="\n")

# --- clean (documented claims!) ------------------------------------
df["created"] = pd.to_datetime(df["created"], errors="coerce")
df["amount"] = pd.to_numeric(df["amount"], errors="coerce")
df["amount"] = df["amount"].fillna(0)          # claim: missing = free ticket

# --- WHERE + ORDER BY ----------------------------------------------
hot = (df[(df["priority"] == "P1") & (df["status"] == "open")]
         .sort_values("created"))

# --- GROUP BY (named aggs) -----------------------------------------
summary = (df.groupby(["priority", "status"])
             .agg(total=("amount", "sum"),
                  n=("id", "count"),
                  avg=("amount", "mean"))
             .reset_index())

# --- JOIN (explicit everything) ------------------------------------
agents = pd.read_csv("agents.csv")[["agent_id", "team"]]
report = (df.merge(agents, on="agent_id", how="left", validate="m:1")
           .groupby("team")["amount"].sum())

# --- memory trick ----------------------------------------------------
df["status"] = df["status"].astype("category")

report.to_csv("team_report.csv")