Skip to main content

Selecting and Transforming Series

Select without ambiguity​

s.loc["Ada"]          # one label
s.loc[["Ada", "Sam"]] # several labels
s.iloc[0]             # first position
s.iloc[:3]            # first three positions
s[s.ge(80)]           # boolean filter

Use .loc for labels and .iloc for positions, especially when the index itself contains integers. Bare s[key] is concise for an unambiguous label, but explicit indexers communicate intent and survive index changes better.

Transform in this order​

  1. Use native vectorized arithmetic or string/datetime accessors.
  2. Use map for a scalar lookup or elementwise function.
  3. Use where, mask, replace, or fillna for conditional replacement.
  4. Use apply only when no clearer vectorized operation exists.
normalized = (s - s.mean()) / s.std()
bands = s.map({90: "A", 80: "B"})
clipped = s.clip(lower=0)

A dictionary passed to map turns unmatched values into missing values. Use replace when unmatched values should remain unchanged.

Combine Series​

Use pd.concat to stack or place objects side by side:

stacked = pd.concat([train, test], ignore_index=True)
table = pd.concat([actual.rename("actual"), predicted.rename("predicted")], axis=1)

Series.append is not the combination API. Before column-wise concatenation, check whether label alignment is intended.

Iteration​

If iteration is genuinely necessary, s.items() yields (label, value) pairs. Do not use iteration for arithmetic or routine filtering; vectorized expressions are clearer and usually faster.

Source​