numpy is intended to replace numarray and Numeric. It comes with a script to convert most adjustments for numarray (except for strings and records).

To have an overview of (some) changes needed to convert a numarray based program, this page will document the list of necessary changes that must be made to your code .

Python Changes

Numarray	NumPy
`import numarray`	`import numpy`
`numarray.array(shape=s,type=t)`	`numpy.empty(shape=s,dtype=t)`
`numarray.abs(x)`	`numpy.absolute(x)`
`.type()`	`.dtype.type` or `.dtype` or `.dtype.name` or ...
`.rank`	`.ndim`
`.itemsize()`	`.itemsize`
`._byteorder == sys.byteorder`	`.dtype.byteorder == '='`
`NumArray`	`ndarray`
`ComplexType`	`complexfloating`
`NumericType`	`number`
`type=`	`dtype=`
`UInt16` (etc)	`uint16` (etc)
`Complex32`	`complex64` (!)
`Complex64`	`complex128` (!)
`.iscontiguous()`	`.flags.contiguous`
`.linear_algebra.inverse`	`.linalg.inv`
`.linear_algebra.generalized_inverse`	`.linalg.pinv`

Some module functions and/or array methods have changed bahaviour

in numpy: arr.transpose() method returns a transposed view of arr - i.e. it shares the memory with the original
in numarray: array.transpose() returns nothing
the transpose() function in numarray is different from the method - but in numpy they are the same.

Extension module Changes

Numarray provided two separate APIs to write C extensions. One was inherently "new" (numarray specific) while the other was intended for easier transition from Numeric. (note/check: if I'm not mistaken, this second one was more feature limited)

1) The numarray-specific C-API is supported by numpy. Use

#include "numpy/libnumarray.h" instead of #include "numarray/libnumarray.h"

2) If you used the Numeric-compatible C-API, then it is also easy. Full

backward-compatible support is available using #include "numpy/oldnumeric.h" instead of "include "numarray/libnumeric.h"