flox.Aggregation¶
- class flox.Aggregation(name, *, numpy=None, chunk, combine, preprocess=None, finalize=None, fill_value=None, final_fill_value=<NA>, dtypes=None, final_dtype=None, reduction_type='reduce', new_dims_func=None, preserves_dtype=False)[source]¶
- Attributes:
- new_dims
- num_new_vector_dims
- __init__(name, *, numpy=None, chunk, combine, preprocess=None, finalize=None, fill_value=None, final_fill_value=<NA>, dtypes=None, final_dtype=None, reduction_type='reduce', new_dims_func=None, preserves_dtype=False)[source]¶
Blueprint for computing grouped aggregations.
See aggregations.py for examples on how to specify reductions.
- Attributes:
- namestr
Name of reduction.
- numpystr or callable, optional
Reduction function applied to numpy inputs. This function should compute the grouped reduction and must have a specific signature. If string, these must be “native” reductions implemented by the backend engines (numpy_groupies, flox, numbagg). If None, will be set to
name
.- chunkNone or str or tuple of str or callable or tuple of callable
For dask inputs only. Either a single function or a list of functions to be applied blockwise on the input dask array. If None, will raise an error for dask inputs.
- combineNone or str or tuple of str or callbe or tuple of callable
For dask inputs only. Functions applied when combining intermediate results from the blockwise stage (see
chunk
). If None, will raise an error for dask inputs.- finalizecallable
For dask inputs only. Function that combines intermediate results to compute final result.
- preprocesscallable
For dask inputs only. Preprocess inputs before
chunk
stage.- reduction_type{“reduce”, “argreduce”}
Type of reduction.
- fill_valuenumber or tuple(number), optional
Value to use when a group has no members. If single value will be converted to tuple of same length as chunk. If appropriate, provide a different fill_value per reduction in
chunk
as a tuple.- final_fill_valueoptional
fill_value for final result.
- dtypesDType or tuple(DType), optional
dtypes for intermediate results. If single value, will be converted to a tuple of same length as chunk. If appropriate, provide a different fill_value per reduction in
chunk
as a tuple.- final_dtypeDType, optional
DType for output. By default, uses dtype of array being reduced.
- new_dims_func: Callable
Function that receives finalize_kwargs and returns a tupleof sizes of any new dimensions added by the reduction. For e.g. quantile for q=(0.5, 0.85) adds a new dimension of size 2, so returns (2,)
- preserves_dtype: bool,
Whether a function preserves the dtype on return E.g. min, max, first, last, mode
Methods
__init__
(name, *[, numpy, preprocess, ...])Blueprint for computing grouped aggregations.
Attributes
new_dims
num_new_vector_dims