numarray User's Manual

Previous: 4.3 Creating arrays with Up: 4. Array Basics Next: 4.5 Operating on Arrays

---

4.4 Coercion and Casting

We've mentioned the types of arrays, and how to create arrays with the right type. But what happens when arrays with different types interact? For some operations, the behavior of numarray is significantly different from Numeric.

4.4.1 Automatic Coercions and Binary Operations

In numarray (in contrast to Numeric), there is now a distinction between how coercion is treated in two basic cases: array/scalar operations and array/array operations. In the array/array case, the coercion rules are nearly identical to those of Numeric, the only difference being combining signed and unsigned integers of the same size. The array/array result types are enumerated in table 4.2.

Scalars, however, are treated differently. If the scalar is of the same ``kind'' as the array (for example, the array and scalar are both integer types) then the output is the type of the array, even if it is of a normally ``lower'' type than the scalar. Adding an Int16 array with an integer scalar results in an Int16 array, not an Int32 array as is the case in Numeric. Likewise adding a Float32 array to a float scalar results in a Float32 array rather than a Float64 array as is the case with Numeric. Adding an Int16 array and a float scalar will result in a Float64 array, however, since the scalar is of a higher kind than the array. Finally, when scalars and arrays are operated on together, the scalar is converted to a rank-0 array first. Thus, adding a ``small'' integer to a ``large'' floating point array is equivalent to first casting the integer ``up'' to the type of the array.

>>> print (array ((1, 2, 3), type=Int16) * 2).type()
numarray type: Int16
>>> arange(0, 1.0, .1) + 12
array([ 12. , 12.1, 12.2, 12.3, 12.4, 12.5, 12.6, 12.7, 12.8, 12.9]

The results of array/scalar operations are enumerated in table 4.3. Entries marked with " are identical to their neighbors on the same row.

4.4.2 The type value table

The type identifiers (Float32, etc.) are NumericType instances. The mapping between type and the equivalent C variable is machine dependent. The correspondences between types and C variables for 32-bit architectures are shown in Table 4.4.

4.4.3 Long: the platform relative type

4.4.4 Deliberate casts (potentially down)

astype(

type)

You may also force numarray to cast any number array to another number array. For example, to take an array of any numeric type (IntX or FloatX or ComplexX or UIntX) and convert it to a 64-bit float, one can do:

>>> floatarray = otherarray.astype(Float64)

The type can be any of the number types, ``larger'' or ``smaller''. If it is larger, this is a cast-up. If it is smaller, the standard casting rules of the underlying language (C) are used, which means that truncation or integer wrap-around can occur:

>>> print x
[   0.     0.4    0.8    1.2  300.6]
>>> print x.astype(Int32)
[  0   0   0   1 300]
>>> print x.astype(Int8)      # wrap-around
[ 0  0  0  1 44]

If the type used with astype is the original array's type, then a copy of the original array is returned.

Footnotes

...

Float64

...

Complex64

---

numarray User's Manual

Previous: 4.3 Creating arrays with Up: 4. Array Basics Next: 4.5 Operating on Arrays

---

Release 1.5, documentation updated on November 2, 2005.