pandas-dev/pandas · error · TypeError

PeriodArray does not allow floating point in construction

Error message

PeriodArray does not allow floating point in construction

What it means

Raised by PeriodArray._from_sequence when the coerced input has float dtype and contains at least one non-NaN value. Period ordinals are integers by definition, so genuine floating-point data cannot be interpreted as periods; only all-NaN float arrays are tolerated (interpreted as all-NaT).

Source

Thrown at pandas/core/arrays/period.py:306

                return scalars.copy()
            return scalars

        if not isinstance(
            scalars, (np.ndarray, list, tuple, ABCSeries, ABCIndex, ExtensionArray)
        ):
            # test_constructor_empty_special has a case with an iter object
            scalars = list(scalars)

        if isinstance(scalars, ExtensionArray) and scalars.dtype.kind in "iu":
            # e.g. masked or arrow-backed integer array; np.asarray would cast
            #  integers-with-NA to float and raise a misleading "floating point"
            #  error below, so route through object dtype to keep the integers.
            scalars = scalars.to_numpy(dtype=object, na_value=NaT)

        arrdata = np.asarray(scalars)
        if arrdata.dtype.kind == "f" and len(arrdata) > 0:
            if not lib.all_nans(arrdata):
                raise TypeError(
                    "PeriodArray does not allow floating point in construction"
                )
            ordinals = np.full(arrdata.shape, iNaT, dtype=np.int64)
            return cls(ordinals, dtype=dtype)

        elif arrdata.dtype.kind in "iu":
            arr = arrdata.astype(np.int64, copy=False)
            ordinals = libperiod.from_calendar_ordinals(arr, dtype)  # type: ignore[arg-type]
            return cls(ordinals, dtype=dtype)
        else:
            periods = ensure_object(arrdata)

            if dtype is None:
                dtype_base = libperiod.extract_period_unit(periods)
                dtype = PeriodDtype(dtype_base)
            ordinals = libperiod.extract_ordinals(periods, dtype)  # type: ignore[arg-type]
            return cls(ordinals, dtype=dtype)

View on GitHub (pinned to 71959b8cb9)

Solutions

Cast to integer first: np.asarray(data, dtype='int64') (only safe if no fractional part).
Drop or NaN-out fractional values, then construct from strings/Period objects.
Build from Period objects or ISO strings via pd.period_array(['2020-01-01', ...]).

Example fix

# before
pd.period_array([2020.5, 2021.0], dtype=pd.PeriodDtype('Y'))
# after
pd.period_array([2020, 2021], dtype=pd.PeriodDtype('Y'))
# or from strings
pd.period_array(['2020','2021'], dtype=pd.PeriodDtype('Y'))

Defensive patterns

Strategy: validation

Validate before calling

import numpy as np
import pandas as pd

def to_period_safe(data, freq):
    arr = np.asarray(data)
    if arr.dtype.kind == 'f':
        if not pd.isna(arr).all() and not np.all(np.equal(np.mod(arr, 1), 0)):
            raise ValueError('data has fractional floats; cannot be periods')
        arr = arr.astype('int64', copy=False)
    return pd.period_array(arr, dtype=pd.PeriodDtype(freq))

Type guard

import numpy as np

def is_integer_or_all_nan_float(arr) -> bool:
    a = np.asarray(arr)
    if a.dtype.kind == 'i':
        return True
    if a.dtype.kind == 'f':
        return bool(np.all(np.isnan(a)) or np.all(np.equal(np.mod(a, 1), 0)))
    return False

Try / catch

try:
    pa = pd.period_array(data, dtype=pd.PeriodDtype(freq))
except TypeError:
    pa = pd.period_array(np.asarray(data, dtype='int64'), dtype=pd.PeriodDtype(freq))

Prevention

Coerce period-source columns to Int64 at load time, not Float64.
Reject fractional floats upstream rather than relying on the period constructor.
Prefer building periods from ISO strings or Period objects.

When it happens

Trigger: pd.period_array([1.5, 2.0], dtype=...), pd.PeriodIndex([3.14]), or feeding a float Series into a period dtype. Numeric columns with NaN that get cast to float before being treated as periods but contain real fractional values.

Common situations: Reading a CSV column of years as float (2020.0) and trying to build a PeriodIndex. Mixing NaN with integer-like data producing float dtype. Off-by-one casts from nullable Int64 to Float64.

Related errors

AI-assisted analysis of pandas-dev/pandas@71959b8cb9 (2026-08-07). Data as JSON: /api/errors/2bb63d31cfa9a852. Report an issue: GitHub.