[Python] Why does NumPy assign different string dtypes when mixing types in np.array()?

I'm trying to understand how NumPy determines the dtype when creating an array with mixed types. I noticed that the inferred dtype for strings can vary significantly depending on the order and type of elements in the list.

print(np.array([1.0, True, 'is']))
# Output: array(['1.0', 'True', 'is'], dtype='<U32')

print(np.array(['1.0', True, 'is']))
# Output: array(['1.0', 'True', 'is'], dtype='<U5')

print(np.array(['1.0', 'True', 'is']))
# Output: array(['1.0', 'True', 'is'], dtype='<U4')


I understand that NumPy upcasts everything to a common type ... usually the most general one.., and that strings tend to dominate. But why does the resulting dtype (<U32, <U5, <U4) differ so much when the content looks almost the same?

Specifically:

Why does np.array([1.0, True, 'is']) result in ?

What determines the exact length in the dtype (e.g., vs )?

Is there a consistent rule for how NumPy infers the dtype and string length in such cases?

Continue reading...