pola-rs/polars · error

can not get dtype of Categorical AnyValue

Error message

can not get dtype of Categorical AnyValue

What it means

AnyValue::dtype() panics for Categorical/CategoricalOwned scalars. A Categorical AnyValue carries only the physical category index plus a borrowed reference to the categories; constructing a full DataType::Categorical would require deciding on and cloning the categories mapping, so the conversion is deliberately unimplemented.

Source

Thrown at crates/polars-core/src/datatypes/any_value.rs:268

            Float32(_) => DataType::Float32,
            Float64(_) => DataType::Float64,
            String(_) | StringOwned(_) => DataType::String,
            Binary(_) | BinaryOwned(_) => DataType::Binary,
            #[cfg(feature = "dtype-date")]
            Date(_) => DataType::Date,
            #[cfg(feature = "dtype-time")]
            Time(_) => DataType::Time,
            #[cfg(feature = "dtype-datetime")]
            Datetime(_, tu, tz) => DataType::Datetime(*tu, (*tz).cloned()),
            #[cfg(feature = "dtype-datetime")]
            DatetimeOwned(_, tu, tz) => {
                DataType::Datetime(*tu, tz.as_ref().map(|v| v.as_ref().clone()))
            },
            #[cfg(feature = "dtype-duration")]
            Duration(_, tu) => DataType::Duration(*tu),
            #[cfg(feature = "dtype-categorical")]
            Categorical(_, _) | CategoricalOwned(_, _) => {
                unimplemented!("can not get dtype of Categorical AnyValue")
            },
            #[cfg(feature = "dtype-categorical")]
            Enum(_, _) | EnumOwned(_, _) => unimplemented!("can not get dtype of Enum AnyValue"),
            List(s) => DataType::List(Box::new(s.dtype().clone())),
            #[cfg(feature = "dtype-array")]
            Array(s, size) => DataType::Array(Box::new(s.dtype().clone()), *size),
            #[cfg(feature = "dtype-struct")]
            Struct(_, _, fields) => DataType::Struct(fields.to_vec()),
            #[cfg(feature = "dtype-struct")]
            StructOwned(payload) => DataType::Struct(payload.1.clone()),
            #[cfg(feature = "dtype-decimal")]
            Decimal(_, p, s) => DataType::Decimal(*p, *s),
            #[cfg(feature = "object")]
            Object(o) => DataType::Object(o.type_name()),
            #[cfg(feature = "object")]
            ObjectOwned(o) => DataType::Object(o.0.type_name()),
        }
    }

View on GitHub (pinned to df599052da)

Solutions

1. Get the dtype from the Series/Column instead of the scalar: df["cat"].dtype
2. Extract the scalar as a string first if you only need its value: s.get(0).cast(pl.String) or str value accessors
3. If you truly need a per-scalar dtype, pattern-match the AnyValue and handle Categorical(_, ref_map) yourself

Example fix

# before
av = df["cat"].get(0)
dt = av.dtype()  # panics for Categorical AnyValue

# after
dt = df["cat"].dtype  # dtype from the column, never panics

# Ask the column, not the scalar
if isinstance(df["cat"].dtype, pl.Categorical):
    dtype = df["cat"].dtype  # safe
else:
    dtype = df["cat"].get(0).dtype()

def scalar_dtype_safe(series: pl.Series, idx: int = 0):
    if isinstance(series.dtype, (pl.Categorical, pl.Enum)):
        return series.dtype  # categorical/enum scalars cannot produce dtypes
    return series.get(idx).dtype()

When it happens

Trigger: Introspecting the dtype of a scalar extracted from a Categorical column: s = df["cat"]; av = s.get(0); av.dtype() - or any library code path that calls AnyValue::dtype() on a value taken from a categorical/enum Series (scalar coercion, schema inference from scalars).

Common situations: Building schemas from sample values, generic scalar-handling utilities, or passing a single categorical value around and querying its type instead of the column's type.

Related errors

• can not get dtype of Enum AnyValue
• comparing categoricals with different Categories is not supp
• not implemented
• not implemented
• horizontal_flatten not supported for data type {:?}

AI-assisted analysis of pola-rs/polars@df599052da (2026-08-16). Data as JSON: /api/errors/cd4607a189ed4928. Report an issue: GitHub.