Every ndarray has one dtype. Casting can overflow. Missing values need a float dtype because integers have no NaN.

Goal

Inspect dtypes, convert with astype, and see integer overflow and result_type.

Default dtypes

print(np.array([12, 7, 2]).dtype)
print(np.array([10.5, 22.0, 31.0]).dtype)
print(np.array([True, False, True]).dtype)

`astype`

units = np.array([[12, 7, 2], [9, 6, 0]], dtype=float)
as_int = units.astype(np.int64)
print(as_int)
print(as_int.dtype)

Overflow

big = np.array([200, 100, 50], dtype=np.int8)
print(big)
print(big.astype(np.int16))

int8 only holds -128…127. 200 wraps. Safer: start from int64 or float64, then narrow if you must.

Mixing types

units = np.array([12, 7, 2], dtype=np.int64)
prices = np.array([10.5, 22.0, 31.0])
print(np.result_type(units, prices))
print(units * prices)

The product is float64 — NumPy promotes rather than truncating prices.

Why NaN is float

vals = np.array([12, 7, 2], dtype=np.int64)
try:
    vals[1] = np.nan
except ValueError as err:
    print(type(err).__name__, err)
print()
as_float = vals.astype(float)
as_float[1] = np.nan
print(as_float)

Pitfall

astype(int) on an array that contains nan fails or becomes an arbitrary integer depending on the path. Convert to float before you store missing values.