Compatibility with numpy functions — Dask documentation (original) (raw)
Toggle navigation sidebar
Compatibility with numpy functions¶
The following table describes the compatibilities between numpy and dask.arrayfunctions. Please be aware that some inconsistencies with the two versions may exist.
This table has been compiled manually and may not reflect the current Dask state. Update contributions are welcome.
- A blank entry indicates that the function is not implemented in Dask.
- Direct implementation are direct calls to numpy functions.
- Element-wise implementations are derived from numpy but applied element-wise: the argument should be a dask array.
- Dask equivalent are Dask implementations, which may lack or add parameters with respect to the numpy function.
A more in-depth comparison in the framework of the Array APIis available via the Array API Comparison repository.
Footnotes
21(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15)
where
parameter not supported.
initial
parameter not supported.
Input must be a dask array.
order
parameter not supported.
Sort operations are notoriously difficult to do in parallel. Parallel-friendly alternatives sort the k largest elements.
out
parameter not supported.
Use of numpy.matrix is discouraged in NumPy and thus there is no need to add it.
mode
parameter not supported.
keepdims
parameter not supported.
fweights
, aweights
, dtype
parameters not supported.
like
parameter not supported. Callable functions not supported.
Not implemented with more than one output.
edge_order
parameter not supported.
Chunking of the input data (sample) is only allowed along the 0th (row) axis.
Only implemented for monotonic obj
arguments.
overwrite_input
parameter not supported.
copy
parameter not supported.
Dask implementation introduces an additional parameter method
.
axis
parameter not supported.
casting
parameter not supported.