How to convert a NumPy array dtype

Text-based data sources can leave measurements in a NumPy string array, where arithmetic and numeric comparisons do not use numeric semantics. A numeric dtype makes those values usable by NumPy calculations without changing the array's shape.

The ndarray.astype() method casts each element to the requested dtype and returns a converted array. Naming np.float64 explicitly gives the result a fixed 64-bit floating-point representation, while the source string array remains available for inspection.

Every source value must be valid decimal text for this conversion to succeed; an invalid token raises ValueError. Dtype and value assertions stop execution if the result differs from the expected floating-point array.

1. Create array-convert-dtype.py with decimal text stored in a NumPy array.

   array-convert-dtype.py

   import numpy as np
    
   raw = np.array(["12.5", "18.0", "21.75"])

2. Append the float64 conversion to array-convert-dtype.py.

   array-convert-dtype.py

   converted = raw.astype(np.float64)

   astype() defaults to casting=“unsafe”. Narrowing a numeric dtype can therefore round, truncate, or overflow values, while the assertions in the next step detect changes for this decimal-string conversion.

3. Append dtype and value assertions to array-convert-dtype.py.

   array-convert-dtype.py

   assert converted.dtype == np.dtype("float64")
   np.testing.assert_allclose(converted, [12.5, 18.0, 21.75])
    
   print("source dtype:", raw.dtype)
   print("converted dtype:", converted.dtype)
   print("converted values:", converted)

4. Run array-convert-dtype.py to verify the converted dtype and values.

   $ python3 array-convert-dtype.py
   source dtype: <U5
   converted dtype: float64
   converted values: [12.5  18.   21.75]