What causes "Cannot set float NaN to integer-backed IntervalArray" in pandas-dev/pandas?

Cast the array to a nullable-integer subtype: arr = arr.astype('interval[Int64]') or 'interval[float64]' before assigning NaN.

pandas-dev/pandas · error · TypeError

Cannot set float NaN to integer-backed IntervalArray

Error message

Cannot set float NaN to integer-backed IntervalArray

What it means

Raised by IntervalArray._validate_setitem_value when the scalar is a valid NA for the dtype but the underlying endpoint array is numpy integer-backed, which cannot store NaN. Pandas raises TypeError (not ValueError) intentionally to match the lossy-setitem contract for non-NA values on int arrays (GH#45484). Use a nullable subtype to allow NA.

Source

Thrown at pandas/core/arrays/interval.py:1199

            left = right = self.left._na_value
        else:
            raise TypeError(
                "can only insert Interval objects and NA into an IntervalArray"
            )
        return left, right

    def _validate_setitem_value(self, value):
        if is_list_like(value):
            return self._validate_listlike(value)

        left, right = self._validate_scalar(value)

        if is_valid_na_for_dtype(value, self.left.dtype):
            if is_integer_dtype(self.dtype.subtype):
                # can't set NaN on a numpy integer array
                # GH#45484 TypeError, not ValueError, matches what we get with
                #  non-NA un-holdable value.
                raise TypeError("Cannot set float NaN to integer-backed IntervalArray")

        return left, right

    # ---------------------------------------------------------------------
    # Rendering Methods

    def _formatter(self, boxed: bool = False) -> Callable[[object], str]:
        # returning 'str' here causes us to render as e.g. "(0, 1]" instead of
        #  "Interval(0, 1, closed='right')"
        return str

    # ---------------------------------------------------------------------
    # Vectorized Interval Properties/Attributes

    @property
    def left(self) -> Index:
        """
        Return the left endpoints of each Interval in the IntervalArray as an Index.

View on GitHub (pinned to 71959b8cb9)

Solutions

Cast the array to a nullable-integer subtype: arr = arr.astype('interval[Int64]') or 'interval[float64]' before assigning NaN.
Drop the row/index rather than setting NA if the subtype must stay numpy int.
Build the IntervalArray with a nullable dtype from the start via pd.array(..., dtype='interval[Int64]').

Example fix

# before
arr = pd.arrays.IntervalArray.from_tuples([(0, 1), (2, 3)])
arr[0] = np.nan

# after
arr = arr.astype('interval[float64]')
arr[0] = np.nan

Defensive patterns

Strategy: validation

Validate before calling

def can_hold_na(arr) -> bool:
    return not (np.issubdtype(arr.dtype.subtype, np.integer) and
                not isinstance(arr.dtype.subtype, pd.api.types.pandas_dtype('Int64').type))

# simpler: check the dtype string
def needs_nullable_for_na(arr):
    return 'int' in str(arr.dtype.subtype).lower() and 'Int' not in str(arr.dtype.subtype)

Type guard

def is_nullable_interval(arr) -> bool:
    sub = str(arr.dtype.subtype)
    return sub.startswith('Int') or sub.startswith('float') or sub.startswith('datetime')

Try / catch

try:
    arr[i] = np.nan
except TypeError as e:
    if 'integer-backed IntervalArray' in str(e):
        arr = arr.astype('interval[float64]')
        arr[i] = np.nan

Prevention

Build interval columns with nullable subtypes ('Int64','Float64') if NA is possible.
Drop rows instead of masking when subtype must be numpy int.
Document NA support per interval column in your schema.

When it happens

Trigger: Assigning np.nan or pd.NA into an IntervalArray with dtype 'interval[int64]', e.g. arr[i] = np.nan where arr.dtype.subtype is int64.

Common situations: Migrating code that worked on float-backed intervals to int-backed intervals; cleaning data pipelines that impute NaN into interval columns; reading parquet/CSV that yielded int intervals and then trying to mask out values.

Related errors

AI-assisted analysis of pandas-dev/pandas@71959b8cb9 (2026-08-07). Data as JSON: /api/errors/436ccf8bb5a548f3. Report an issue: GitHub.