🕯️LORE

Series

Properties2
tagspandas, python, data-structures
aliasespd.Series, Pandas Series

3 min read

Series

Definition

A Series is a one-dimensional, labeled array capable of holding any data type (integers, strings, floats, Python objects, etc.). It is the building block of a DataFrame , think of a DataFrame as a dict of Series sharing the same index.

Every Series has two components:

  • values : a NumPy array (or Extension Array) holding the data
  • index : the axis labels for the data

Syntax

pd.Series(data=None, index=None, dtype=None, name=None, copy=None)
ParameterDescription
dataarray-like, dict, scalar, or iterable
indexarray-like of labels; defaults to RangeIndex(0, n)
dtypeforce a specific dtype
namename of the Series (becomes column name if put into a DataFrame)
copycopy input data (default False for array-like)

Creating a Series

# From a list
s = pd.Series([10, 20, 30, 40])
 
# With a custom index
s = pd.Series([10, 20, 30], index=["a", "b", "c"])
 
# From a dict (keys become index)
s = pd.Series({"a": 1, "b": 2, "c": 3})
 
# From a scalar (broadcast across index)
s = pd.Series(5.0, index=["a", "b", "c"])
 
# With a name
s = pd.Series([1, 2, 3], name="scores")

Key Attributes

AttributeReturns
s.valuesunderlying NumPy/Extension array
s.indexthe Index object
s.dtypedata type of elements
s.shapetuple (n,)
s.sizenumber of elements
s.namename of the Series
s.emptyTrue if length is 0
s.nbytesmemory usage in bytes
s.is_uniqueTrue if all values distinct
s.hasnansTrue if any NaN present

Indexing & Access

s["a"]            # label-based access
s.iloc[0]          # position-based access
s.loc["a":"c"]     # label slice (inclusive of endpoint!)
s.iloc[0:2]        # position slice (exclusive of endpoint)
s[["a", "c"]]      # fancy indexing, multiple labels
s[s > 10]          # boolean filtering

Slicing endpoints differ

.loc slices are inclusive on both ends; .iloc slices behave like standard Python slicing (exclusive end).

Core Operations

s.sum(); s.mean(); s.median(); s.std(); s.var()
s.min(); s.max(); s.idxmin(); s.idxmax()
s.cumsum(); s.cumprod(); s.cummax(); s.cummin()
s.sort_values(ascending=True)
s.sort_index()
s.rank()
s.unique()          # array of unique values
s.nunique()         # count of unique values
s.value_counts()    # frequency table, sorted descending
s.astype("float64") # cast dtype
s.round(2)
s.apply(lambda x: x * 2)
s.map({1: "one", 2: "two"})   # element-wise mapping

Vectorized Arithmetic (auto-aligned by index)

s1 = pd.Series([1, 2, 3], index=["a", "b", "c"])
s2 = pd.Series([10, 20, 30], index=["b", "c", "d"])
 
s1 + s2
# a     NaN   (no match)
# b    12.0
# c    32.0
# d     NaN   (no match)

Fill missing alignment

Use s1.add(s2, fill_value=0) to treat unmatched labels as 0 instead of producing NaN.

Boolean & Membership

s.isin([10, 20])          # element-wise membership test
s.between(10, 30)         # inclusive range test
s.isna() / s.notna()      # null checks
s.all() / s.any()         # for boolean series

Converting

s.to_list()
s.to_dict()
s.to_frame(name="col")    # convert Series -> single-column DataFrame
s.to_numpy()
pd.to_numeric(s, errors="coerce")

Notes & Gotchas

Series is index-aware, not position-aware by default

Arithmetic and comparisons align on the index, not position. Mismatched indices silently produce NaN , always check .index when combining Series from different sources.

dtype=object is a catch-all

If a Series mixes types (e.g., strings and numbers), pandas falls back to object dtype, which is slower and loses vectorized numeric operations. Use pd.to_numeric(s, errors="coerce") to clean it up.

Graph View

Backlinks