Summarising a table

Every aggregation from the NumPy track works here. On a Series they give one number; on a DataFrame they run per column and hand back a Series indexed by column name:

orders["cups"].sum()          # one number
orders.sum(numeric_only=True) # one per numeric column
orders.mean(numeric_only=True)

numeric_only=True matters on a mixed table — without it, summing a text column either concatenates the strings or raises, depending on the operation.

describe as a checklist

orders.describe() runs eight summaries at once for every numeric column. Its value isn't the numbers themselves — it's that the rows form a checklist for problems you haven't thought to look for yet.

count counts non-missing values. Two columns with different counts means one has gaps, and you found them without going looking.

min and max are where nonsense lives. An age of -1, a price of 999999, a date in 1900 — sentinel values and parsing failures sit at the extremes, and this puts them in front of you.

mean against 50% is your skew check. When they diverge, the distribution has a tail and every average you report afterwards is misleading — that's the lesson-21 argument from the NumPy track, applied automatically to every column.

std of zero means a constant column, which is usually a bug or a column you can drop.

orders.describe()                  # numeric columns
orders.describe(include="all")     # text too: count, unique, top, freq
orders.describe(include="object")  # only the text columns

Counting categories

describe is for numbers. For a text or categorical column you want counts:

orders["city"].value_counts()                 # how many of each
orders["city"].value_counts(normalize=True)   # as proportions
orders["city"].value_counts(dropna=False)     # count the gaps too
orders["city"].nunique()                      # 3 — how many distinct

value_counts sorts by frequency descending, so the most common value is at the top. That's usually what you want — and it's how you spot a category that appears twice under two spellings, which is what lesson 11 was about.

Naming your own summaries

orders.agg(["min", "max", "mean"], numeric_only=True)
orders.agg({"cups": "sum", "revenue": "mean", "rating": "max"})

.agg() takes a list to apply several functions, or a dict to apply a different one per column. It also takes your own function, which is how you'd add a percentile or a custom metric to the same table.

All of these skip missing values by default — the same choice lesson 9 made explicit. If you want the gaps to propagate instead, do the arithmetic yourself.