[Feature] MXNet NumPy Standard APIs Implementation Checklist

[barry-jin]

As proposed in #20501, MXNet will support standardized Python Array APIs and I have collected most of unimplemented operators or methods to be tracked in this issue.

These operators or methods are splited into three parts. The part1 is the basic element-wise functions, which are prerequistes to implement the operators and methods in part2. Functions in part3 are some Linear Algebra Functions.

Part1: Element-wise Functions

• bitwise_left_shift ([API] Add bitwise_left/right_shift #20587)
• bitwise_invert : mx.np.invert ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• bitwise_right_shift ([API] Add bitwise_left/right_shift #20587)
• floor_divide ([API] Add floor_divide #20620)
• logaddexp ([API] Add logaddexp #20673)
• positive : (mx.np.copy [API] Add positive #20667)
• pow : mx.np.power ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)

Part2:

Operators:

• __pos__ : (positive(x) [API] Add positive #20667)
• __floordiv__ : floor_divide(x1, x2) ([API] Add floor_divide #20620)
• __lshift__ : bitwise_left_shift(x1, x2) ([API] Add bitwise_left/right_shift #20587)
• __rshift__ : bitwise_right_shift(x1, x2) ([API] Add bitwise_left/right_shift #20587)

In-place Operators:

• __ifloordiv__ : floor_divide(self, other, out=self) ([API] Add floor_divide #20620)
• __ipow__ : pow(self, other, out=self) ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• __ilshift__ : bitwise_left_shift(self, other, out=self) ([API] Add bitwise_left/right_shift #20587)
• __irshift__ : bitwise_right_shift(self, other, out=self) ([API] Add bitwise_left/right_shift #20587)

Reflected Operators:

• __radd__ : add(other, self) ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• __rfloordiv__ : floor_divide(other, self) ([API] Add floor_divide #20620)
• __rand__ : bitwise_and(self, other) ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• __ror__ : bitwise_or(self, other) ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• __rxor__ : bitwise_xor(self, other) ([API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• __rlshift__ : bitwise_left_shift(other, self) ([API] Add bitwise_left/right_shift #20587)
• __rrshift__ : bitwise_right_shift(other, self) ([API] Add bitwise_left/right_shift #20587)

Part3:

Linear Algebra Functions:

To be implemented:

• linalg.cross : (mx.np.cross [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)
• linalg.diagonal : (mx.np.diagonal [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)
• linalg.matrix_power : (fallback to onp.linalg.matrix_power)
• linalg.outer : (mx.np.outer [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)
• linalg.tensordot : (mx.np.tensordot [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)
• linalg.svdvals ([API] Add linalg.svdvals #20696)
• linalg.trace : (mx.np.trace [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)
• linalg.vecdot : (mx.np.vdot [API Standardization]Standardize MXNet NumPy Statistical & Linalg Functions #20592)
• linalg.vector_norm : (mx.np.linalg.norm [API NEW][LINALG] Add vector_norm, matrix_norm #20703)
• linalg.matrix_norm : (mx.np.linalg.norm [API NEW][LINALG] Add vector_norm, matrix_norm #20703)
• linalg.matrix_transpose : (mx.np.transpose [API Standardization] Add Linalg kernels: (diagonal, outer, tensordot, cross, trace, matrix_transpose) #20638)

To be standardized

• linalg.cholesky ([API STD][LINALG] Standardize sort & linalg operators #20694)
• linalg.matrix_rank ([API STD][LINALG] Standardize sort & linalg operators #20694)
• linalg.pinv ([API STD][LINALG] Standardize sort & linalg operators #20694)
• linalg.svd

Part4 (TODOs):

To be implemented:

• array.mT : ([API NEW][METHOD] Add mT, permute_dims #20688)
• permute_dims : ([API NEW][METHOD] Add mT, permute_dims #20688)
• __index__ ([API NEW][ARRAY METHOD] Add __Index__() and __array_namespace__() #20689)
• __array_namespace__ ([API NEW][ARRAY METHOD] Add __Index__() and __array_namespace__() #20689)
• can_cast ([WIP][API] Add data type functions #20690)
• finfo ([WIP][API] Add data type functions #20690)
• iinfo ([WIP][API] Add data type functions #20690)
• result_type ([WIP][API] Add data type functions #20690)
• unique_all ([API NEW][SET FUNC] Add set functions #20693)
• unique_inverse ([API NEW][SET FUNC] Add set functions #20693)
• unique_values ([API NEW][SET FUNC] Add set functions #20693)

To be standardized

• argmax ([API STD][SEARCH FUNC] Add keepdims=False to argmax/argmin #20692)
• argmin ([API STD][SEARCH FUNC] Add keepdims=False to argmax/argmin #20692)
• argsort ([API STD][LINALG] Standardize sort & linalg operators #20694)
• sort ([API STD][LINALG] Standardize sort & linalg operators #20694)