Attempt at a more generic wrapper class

Idea is to Wrap ndarray, xr.DataArray, and xr.Dataset objects…

Classes:

MomParamsT	MomParams type variable
GenArrayT	Generic array type variable
CentralMomentsABC(obj, *, mom_params[, fastpath])	Wrapper to calculate central moments.

class cmomy.wrapper._wrapper_abc.MomParamsT

MomParams type variable

alias of TypeVar(‘MomParamsT’, MomParamsArray, MomParamsXArray)

class cmomy.wrapper._wrapper_abc.GenArrayT

Generic array type variable

alias of TypeVar(‘GenArrayT’, NDArray[Any], xr.DataArray, xr.Dataset)

class cmomy.wrapper._wrapper_abc.CentralMomentsABC(obj, *, mom_params, fastpath=False)

Bases: ABC, Generic[GenArrayT, MomParamsT]

Wrapper to calculate central moments.

Parameters:

• obj (numpy.ndarray or xarray.DataArray or xarray.Dataset) – Central moments array.

• mom_params (MomParams or MomParamsDict or dict, optional) – Moment parameters. You can set moment parameters axes and dims using this option. For example, passing mom_params={"dim": ("a", "b")} is equivalent to passing mom_dims=("a", "b"). You can also pass as a MomParams object with mom_params=cmomy.MomParams(dims=("a", "b")).

• fastpath (bool) – For internal use.

Notes

Base data has the form

\[\begin{split}{\rm data}[..., i, j] = \begin{cases} \text{weight} & i = j = 0 \\ \langle x \rangle & i = 1, j = 0 \\ \langle (x - \langle x \rangle^i) (y - \langle y \rangle^j) \rangle & i + j > 0 \end{cases}\end{split}\]

See also

cmomy.wrap, cmomy.wrap_reduce_vals, cmomy.wrap_resample_vals, cmomy.wrap_raw

Attributes:

obj	Underlying object.
mom_params	Moments parameters object.
mom_ndim	Number of moment dimensions.
mom_shape	Shape of moments dimensions.
mom	Moments tuple.
dtype	DType of wrapped object.
shape	Shape of wrapped object.
val_shape	Shape of values dimensions.
ndim	Total number of dimensions.
val_ndim	Number of values dimensions (ndim - mom_ndim).

Methods:

to_numpy()	Coerce wrapped data to ndarray if possible.
new_like([obj, verify, copy, dtype, order, ...])	Create new object like self, with new data.
astype(dtype, *[, order, casting, subok, copy])	Underlying data cast to specified type
zeros_like()	Create new empty object like self.
copy()	Create a new object with copy of data.
push_data(data, *[, casting, parallel, scale])	Push data object to moments.
push_datas(datas, *[, axis, mom_axes, ...])	Push and reduce multiple average central moments.
push_val(x, *y[, weight, casting, parallel])	Push single sample to central moments.
push_vals(x, *y[, axis, weight, casting, ...])	Push multiple samples to central moments.
__iadd__(other)	Self adder.
__add__(other)	Add objects to new object.
__isub__(other)	Inplace subtraction.
__sub__(other)	Subtract objects.
__mul__(scale)	New object with weight scaled by scale.
__imul__(scale)	Inplace multiply.
pipe(func_or_method, *args[, _reorder, ...])	Apply func_or_method to underlying data and wrap results in new wrapped object.
moveaxis([axis, dest, dim, dest_dim, ...])	Generalized moveaxis
select_moment(name, *[, squeeze, ...])	Select specific moments.
assign_moment([moment, squeeze, copy, ...])	Create object with update weight, average, etc.
cumulative(*[, axis, dim, out, dtype, ...])	Convert to cumulative moments.
moments_to_comoments(*, mom[, mom_dims_out, ...])	Convert moments (mom_ndim=1) to comoments (mom_ndim=2).
resample_and_reduce(*, sampler[, axis, dim, ...])	Bootstrap resample and reduce.
jackknife_and_reduce(*[, axis, dim, ...])	Jackknife resample and reduce
reduce()	Create new object reduce along axis.
weight()	Weight data.
mean([squeeze])	Mean (first moment).
var([squeeze])	Variance (second central moment).
std([squeeze])	Standard deviation (ddof=0).
cov()	Covariance (or variance if mom_ndim==1).
cmom()	Central moments.
to_raw(*[, weight])	Raw moments accumulation array.
rmom()	Raw moments.
zeros(*, mom)	Create a new base object.

property obj

Underlying object.

property mom_params

Moments parameters object.

property mom_ndim

Number of moment dimensions.

property mom_shape

Shape of moments dimensions.

property mom

Moments tuple.

property dtype

DType of wrapped object.

property shape

Shape of wrapped object.

property val_shape

Shape of values dimensions.

property ndim

Total number of dimensions.

property val_ndim

Number of values dimensions (ndim - mom_ndim).

to_numpy()

Coerce wrapped data to ndarray if possible.

abstractmethod new_like(obj=None, *, verify=False, copy=None, dtype=None, order=None, fastpath=False)

Create new object like self, with new data.

Parameters:

• obj (numpy.ndarray or xarray.DataArray or xarray.Dataset) – Data for new object. Must be conformable to self.obj.

• verify (bool) – If True, make sure data is c-contiguous.

• copy (bool, optional) – If True, copy the data. If None or False, attempt to use view. Note that False values will be converted to None for numpy versions >2.0. This will be changed to reflect the new behavior of the copy parameter to numpy.array() when the minimum numpy version >2.0.

• dtype (dtype) – Optional dtype for output data.

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• fastpath (bool) – Internal variable.

Returns:

object – New object object with data set to zero.

astype(dtype, *, order=None, casting=None, subok=None, copy=False)

Underlying data cast to specified type

Parameters:

• dtype (str or dtype) – Type code of data-type to cast the array data. Note that a value of None will cast to np.float64. This is the same behavior as asarray().

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• subok (bool, optional) – If True, then sub-classes will be passed-through, otherwise the returned array will be forced to be a base-class array.

• copy (bool, optional) – By default, astype always returns a newly allocated array. If this is set to False and the dtype requirement is satisfied, the input array is returned instead of a copy.

Notes

Only numpy.float32 and numpy.float64 dtype are supported.

See also

numpy.ndarray.astype

zeros_like()

Create new empty object like self.

Returns:

output (object) – Object with same attributes as caller, but with data set to zero.

See also

new_like

copy()

Create a new object with copy of data.

Returns:

output (object) – Same type as calling class. Object with same attributes as caller, but with new underlying data.

See also

new_like, zeros_like

abstractmethod push_data(data, *, casting='same_kind', parallel=False, scale=None)

Push data object to moments.

Parameters:

• data (array-like or numpy.ndarray or xarray.DataArray or xarray.Dataset) – Accumulation array conformable to self.obj.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

• scale (array-like) – Scaling to apply to weights of data. Optional.

Returns:

output (object) – Same object with pushed data.

abstractmethod push_datas(datas, *, axis=-1, mom_axes=None, casting='same_kind', parallel=None)

Push and reduce multiple average central moments.

Parameters:

• datas (array-like or numpy.ndarray or xarray.DataArray or xarray.Dataset) – Collection of accumulation arrays to push onto self.

• mom_axes (int or tuple of int, optional) – Location of the moment dimensions. Default to (-mom_ndim, -mom_ndim+1, ...). If specified and mom_ndim is None, set mom_ndim to len(mom_axes). Note that if mom_axes is specified, negative values are relative to the end of the array. This is also the case for axes if mom_axes is specified.

• axis (int, optional) – Axis to reduce/sample along. Note that negative values are relative to data.ndim - mom_ndim. It is assumed that the last dimensions are for moments. For example, if data.shape == (1,2,3) with mom_ndim=1, axis = -1 `` would be equivalent to ``axis = 1. Defaults to axis=-1.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

Returns:

output (object) – Same object with pushed data.

abstractmethod push_val(x, *y, weight=None, casting='same_kind', parallel=False)

Push single sample to central moments.

Parameters:

• x (array-like or numpy.ndarray or xarray.DataArray or xarray.Dataset) – Values to push onto self.

• *y (array-like or numpy.ndarray or xarray.DataArray or xarray.Dataset) – Additional values (needed if mom_ndim > 1)

• weight (int, float, array-like or numpy.ndarray or xarray.DataArray or xarray.Dataset) – Weight of each sample. If scalar, broadcast w.shape to x0.shape.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

Returns:

output (object) – Same object with pushed data.

Notes

Array x0 should have same shape as self.val_shape.

abstractmethod push_vals(x, *y, axis=-1, weight=None, casting='same_kind', parallel=None)

Push multiple samples to central moments.

Parameters:

• x (array) – Value to reduce.

• *y (array-like) – Additional array (if self.mom_ndim == 2).

• weight (int, float, array-like, optional) – Weight of each sample. If scalar, broadcast to x0.shape

• axis (int) – Axis to reduce/sample along.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

Returns:

output (object) – Same object with pushed data.

__iadd__(other)

Self adder.

__add__(other)

Add objects to new object.

__isub__(other)

Inplace subtraction.

__sub__(other)

Subtract objects.

__mul__(scale)

New object with weight scaled by scale.

__imul__(scale)

Inplace multiply.

pipe(func_or_method, *args, _reorder=True, _copy=None, _verify=False, _fastpath=False, **kwargs)

Apply func_or_method to underlying data and wrap results in new wrapped object.

This is useful for calling any not implemented methods on numpy.ndarray or xarray.DataArray data.

Note that a ValueError will be raised if last dimension(s) do not have the same shape as self.mom_shape.

Parameters:

• func_or_method (str or callable()) – If callable, then apply values = func_or_method(self.to_values(), *args, **kwargs). If string is passed, then values = getattr(self.to_values(), func_or_method)(*args, **kwargs).

• *args (Any) – Extra positional arguments to func_or_method

• _reorder (bool, default True) – If True, reorder the data such that mom_dims are last. Only applicable for DataArray like underlying data.

• _copy (bool, default False) – If True, copy the resulting data. Otherwise, try to use a view.

• _verify (bool, default False)

• **kwargs (Any) – Extra keyword arguments to func_or_method

Returns:

output (object) – New object after func_or_method is applies to self.to_values()

Notes

Use leading underscore for (_copy, etc) to avoid possible name clashes with func_or_method.

moveaxis(axis=MISSING, dest=MISSING, *, dim=MISSING, dest_dim=MISSING, axes_to_end=False)

Generalized moveaxis

Parameters:

• axis (int or sequence of int) – Original positions of axes to move.

• dest (int or sequence of int) – Destination positions for each original axes.

• dim (str or sequence of str) – Original dimensions.

• dest_dim (str or sequence of str) – Destination position of dimensions.

Returns:

output (object) – Object with moved axes. This is a view of the original data.

See also

utils.moveaxis, numpy.moveaxis

select_moment(name, *, squeeze=True, dim_combined='variable', coords_combined=None, keep_attrs=None, apply_ufunc_kwargs=None)

Select specific moments.

Parameters:

• moment ({"weight", "ave", "var", "cov", "xave", "xvar", "yave", "yvar", "xmom_0", "xmom_1", "ymom_0", "ymom_1"}) –

  Name of moment(s) to select.

  • "weight" : weights

  • "ave" : Averages.

  • "var": Variance.

  • "cov": Covariance if mom_ndim == 2, or variance if mom_ndim == 1.

  • "xave": Average of first variable.

  • "xvar": Variance of first variable.

  • "yave": Average of second variable (if mom_ndim == 2).

  • "yvar": Variance of second variable (if mom_ndim == 2).

  • "all": All values.

  Names "weight", "xave", "yave", "xvar", "yvar", "cov" imply shape data.shape[:-mom_ndim]. Names "ave", "var" imply shape (*data.shape[:-mom_ndim], mom_ndim), unless mom_ndim == 1 and squeeze = True.

• squeeze (bool, default False) – If True, squeeze last dimension if name is one of ave or var and mom_ndim == 1.

• dim_combined (str, optional) – Name of dimension for options that produce multiple values (e.g., name="ave").

• coords_combined (str or sequence of str, optional) – Coordinates to assign to dim_combined. Defaults to names of moments dimension(s)

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

Returns:

output (ndarray or DataArray or Dataset) – Same type as self.obj. If name is ave or var, the last dimensions of output has shape mom_ndim with each element corresponding to the ith variable. If squeeze=True and mom_ndim==1, this last dimension is removed. For all other name options, output has shape of input with moment dimensions removed.

See also

utils.select_moment

assign_moment(moment=None, *, squeeze=True, copy=True, dim_combined=None, keep_attrs=None, apply_ufunc_kwargs=None, **moment_kwargs)

Create object with update weight, average, etc.

Parameters:

• moment (mapping of str to array-like) –

  Mapping from moment name to new value. Allowed moment names are:

  • "weight" : weights

  • "ave" : Averages.

  • "var": Variance.

  • "cov": Covariance if mom_ndim == 2, or variance if mom_ndim == 1.

  • "xave": Average of first variable.

  • "xvar": Variance of first variable.

  • "yave": Average of second variable (if mom_ndim == 2).

  • "yvar": Variance of second variable (if mom_ndim == 2).

  • "xmom_n", "ymom_n": All values with first (second) variable moment == n.

  • "all": All values.

  Names "weight", "xave", "yave", "xvar", "yvar", "cov" imply shape data.shape[:-mom_ndim]. Names "ave", "var" imply shape (*data.shape[:-mom_ndim], mom_ndim), unless mom_ndim == 1 and squeeze = True.

• squeeze (bool, default False) – If True, squeeze last dimension if name is one of ave or var and mom_ndim == 1.

• copy (bool, default True) – If True (the default), return new array with updated weights. Otherwise, return the original array with weights updated inplace. Note that a copy is always created for a dask backed object.

• dim_combined (str, optional) – Name of dimensions for multiple values. Must supply if passing in multiple values for name="ave" etc.

• mom_dims (hashable or tuple of hashable) – Name of moment dimensions. If specified, infer mom_ndim from mom_dims. If also pass mom_ndim, check that mom_dims is consistent with mom_dims. If not specified, defaults to data.dims[-mom_ndim:]. This is primarily used if data is a Dataset, or if mom_dims are not the last dimensions.

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

• **moment_kwargs (Union[Buffer, _SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], complex, bytes, str, _NestedSequence[complex | bytes | str], DataArray, Dataset]) – Keyword argument form of moment. Must provide either moment or moment_kwargs.

Returns:

output (object) – Same type as self with updated data.

See also

utils.assign_moment

cumulative(*, axis=MISSING, dim=MISSING, out=None, dtype=None, casting='same_kind', order=None, parallel=None, axes_to_end=False, keep_attrs=None, apply_ufunc_kwargs=None)

Convert to cumulative moments.

Parameters:

• axis (int, optional) – Axis to reduce/sample along. Note that negative values are relative to data.ndim - mom_ndim. It is assumed that the last dimensions are for moments. For example, if data.shape == (1,2,3) with mom_ndim=1, axis = -1 `` would be equivalent to ``axis = 1. Defaults to axis=-1.

• axes_to_end (bool) – If True, place sampled dimension (if exists in output) and moment dimensions at end of output. Otherwise, place sampled dimension (if exists in output) at same position as input axis and moment dimensions at same position as input (if input does not contain moment dimensions, place them at end of array).

• out (ndarray) – Optional output array. If specified, output will be a reference to this array. Note that if the output if method returns a Dataset, then this option is ignored.

• dtype (dtype) – Optional dtype for output data.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• mom_dims (hashable or tuple of hashable) – Name of moment dimensions. If specified, infer mom_ndim from mom_dims. If also pass mom_ndim, check that mom_dims is consistent with mom_dims. If not specified, defaults to data.dims[-mom_ndim:]. This is primarily used if data is a Dataset, or if mom_dims are not the last dimensions.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

Returns:

output (numpy.ndarray or xarray.DataArray or xarray.Dataset) – Same type as self.obj, with moments accumulated over axis.

See also

convert.cumulative

moments_to_comoments(*, mom, mom_dims_out=None, dtype=None, order=None, keep_attrs=None, apply_ufunc_kwargs=None)

Convert moments (mom_ndim=1) to comoments (mom_ndim=2).

Parameters:

• mom (tuple of int) –

  Moments for comoments array. Pass a negative value for one of the moments to fill all available moments for that dimensions. For example, if original array has moments m (i.e., values.shape=(..., m + 1)), and pass in mom = (2, -1), then this will be transformed to

  mom = (2, m - 2).

• mom_dims_out (tuple of str) – Moments dimensions for output (mom_ndim=2) data. Defaults to ("mom_0", "mom_1").

• dtype (dtype) – Optional dtype for output data.

• order ({"C", "F"}, optional) – Order argument. See numpy.zeros().

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

Returns:

output (object) – Same type as self with mom_ndim=2.

See also

convert.moments_to_comoments

resample_and_reduce(*, sampler, axis=MISSING, dim=MISSING, rep_dim='rep', axes_to_end=False, out=None, dtype=None, casting='same_kind', order=None, parallel=None, keep_attrs=None, apply_ufunc_kwargs=None)

Bootstrap resample and reduce.

Parameters:

• sampler (int or array-like or IndexSampler or mapping) – Passed through cmomy.resample.factory_sampler() to create an IndexSampler. Value can either be nrep (the number of replicates), freq (frequency array), a IndexSampler object, or a mapping of parameters. The mapping can have form of FactoryIndexSamplerKwargs. Allowable keys are freq, indices, ndat, nrep, nsamp, paired, rng, replace, shuffle.

• axis (int, optional) – Axis to reduce/sample along. Note that negative values are relative to data.ndim - mom_ndim. It is assumed that the last dimensions are for moments. For example, if data.shape == (1,2,3) with mom_ndim=1, axis = -1 `` would be equivalent to ``axis = 1. Defaults to axis=-1.

• rep_dim (hashable) – Name of new ‘replicated’ dimension:

• axes_to_end (bool) – If True, place sampled dimension (if exists in output) and moment dimensions at end of output. Otherwise, place sampled dimension (if exists in output) at same position as input axis and moment dimensions at same position as input (if input does not contain moment dimensions, place them at end of array).

• out (ndarray) – Optional output array. If specified, output will be a reference to this array. Note that if the output if method returns a Dataset, then this option is ignored.

• dtype (dtype) – Optional dtype for output data.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

Returns:

output (object) – Instance of calling class. Note that new object will have (...,shape[axis-1], nrep, shape[axis+1], ...), where nrep is the number of replicates.

See also

reduce, factory_sampler

resample_data

method to perform resampling

Examples

>>> import cmomy
>>> rng = cmomy.default_rng(0)
>>> da = cmomy.wrap_reduce_vals(
...     rng.random((10, 3)),
...     mom=3,
...     axis=0,
... ).to_x(dims="rec")
>>> da
<CentralMomentsData(mom_ndim=1)>
<xarray.DataArray (rec: 3, mom_0: 4)> Size: 96B
array([[ 1.0000e+01,  5.2485e-01,  1.1057e-01, -4.6282e-03],
       [ 1.0000e+01,  5.6877e-01,  6.8876e-02, -1.2745e-02],
       [ 1.0000e+01,  5.0944e-01,  1.1978e-01, -1.4644e-02]])
Dimensions without coordinates: rec, mom_0

>>> sampler = cmomy.resample.factory_sampler(data=da.obj, dim="rec", nrep=5)
>>> da_resamp = da.resample_and_reduce(
...     dim="rec",
...     sampler=sampler,
... )
>>> da_resamp
<CentralMomentsData(mom_ndim=1)>
<xarray.DataArray (rep: 5, mom_0: 4)> Size: 160B
array([[ 3.0000e+01,  5.0944e-01,  1.1978e-01, -1.4644e-02],
       [ 3.0000e+01,  5.3435e-01,  1.0038e-01, -1.2329e-02],
       [ 3.0000e+01,  5.2922e-01,  1.0360e-01, -1.6009e-02],
       [ 3.0000e+01,  5.5413e-01,  8.3204e-02, -1.1267e-02],
       [ 3.0000e+01,  5.4899e-01,  8.6627e-02, -1.5407e-02]])
Dimensions without coordinates: rep, mom_0

jackknife_and_reduce(*, axis=MISSING, dim=MISSING, data_reduced=None, rep_dim='rep', axes_to_end=False, out=None, dtype=None, casting='same_kind', order=None, parallel=None, keep_attrs=None, apply_ufunc_kwargs=None)

Jackknife resample and reduce

Parameters:

• axis (int, optional) – Axis to reduce/sample along. Note that negative values are relative to data.ndim - mom_ndim. It is assumed that the last dimensions are for moments. For example, if data.shape == (1,2,3) with mom_ndim=1, axis = -1 `` would be equivalent to ``axis = 1. Defaults to axis=-1.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

• data_reduced (array or object) – Data reduced along axis. Array of same type as self.obj or same type as self.

• rep_dim (hashable) – Name of new ‘replicated’ dimension:

• axes_to_end (bool) – If True, place sampled dimension (if exists in output) and moment dimensions at end of output. Otherwise, place sampled dimension (if exists in output) at same position as input axis and moment dimensions at same position as input (if input does not contain moment dimensions, place them at end of array).

• out (ndarray) – Optional output array. If specified, output will be a reference to this array. Note that if the output if method returns a Dataset, then this option is ignored.

• dtype (dtype) – Optional dtype for output data.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• parallel – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

• keep_attrs ({"drop", "identical", "no_conflicts", "drop_conflicts", "override"} or bool, optional) –

  • ‘drop’ or False: empty attrs on returned xarray object.

  • ’identical’: all attrs must be the same on every object.

  • ’no_conflicts’: attrs from all objects are combined, any that have the same name must also have the same value.

  • ’drop_conflicts’: attrs from all objects are combined, any that have the same name but different values are dropped.

  • ’override’ or True: skip comparing and copy attrs from the first object to the result.

• apply_ufunc_kwargs (dict-like) – Extra parameters to xarray.apply_ufunc(). One useful option is on_missing_core_dim, which can take the value "copy" (the default), "raise", or "drop" and controls what to do with variables of a Dataset missing core dimensions. Other options are join, dataset_join, dataset_fill_value, and dask_gufunc_kwargs. Unlisted options are handled internally.

Returns:

object – Instance of calling class with jackknife resampling along axis.

See also

jackknife_data

abstractmethod reduce()

Create new object reduce along axis.

Parameters:

• axis (int, optional) – Axis to reduce/sample along. Note that negative values are relative to data.ndim - mom_ndim. It is assumed that the last dimensions are for moments. For example, if data.shape == (1,2,3) with mom_ndim=1, axis = -1 `` would be equivalent to ``axis = 1. Defaults to axis=-1.

• by (array-like of int) – Group by values of same length as data along sampled dimension. Negative values indicate no group (i.e., skip this index).

• block (int, optional) – If specified, perform block average reduction with blocks of this size. Negative values are transformed to all data.

• axes_to_end (bool) – If True, place sampled dimension (if exists in output) and moment dimensions at end of output. Otherwise, place sampled dimension (if exists in output) at same position as input axis and moment dimensions at same position as input (if input does not contain moment dimensions, place them at end of array).

• out (ndarray) – Optional output array. If specified, output will be a reference to this array. Note that if the output if method returns a Dataset, then this option is ignored.

• dtype (dtype) – Optional dtype for output data.

• casting ({'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional) –

  Controls what kind of data casting may occur.

  • ’no’ means the data types should not be cast at all.

  • ’equiv’ means only byte-order changes are allowed.

  • ’safe’ means only casts which can preserve values are allowed.

  • ’same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.

  • ’unsafe’ (default) means any data conversions may be done.

• order ({"C", "F", "A", "K"}, optional) – Order argument. See numpy.asarray().

• keepdims (bool) – If this is set to True, the axes which are reduced are left in the result as dimensions with size one. With this option, the result will broadcast correctly against the input array.

• parallel (bool, optional) – If True, use parallel numba numba.njit or numba.guvectorized code if possible. If None, use a heuristic to determine if should attempt to use parallel method.

Returns:

output (object) – If by is None, reduce all samples along axis, optionally keeping axis with size 1 if keepdims=True. Otherwise, reduce for each unique value of by. In this case, output will have shape (..., shape[axis-1], ngroup, shape[axis+1], ...) where ngroups = np.max(by) + 1 is the number of unique positive values in by.

See also

reduce_data, reduce_data_grouped, reduce_data_indexed, block_by

weight()

Weight data.

mean(squeeze=True)

Mean (first moment).

var(squeeze=True)

Variance (second central moment).

std(squeeze=True)

Standard deviation (ddof=0).

cov()

Covariance (or variance if mom_ndim==1).

cmom()

Central moments.

Strict central moments of the form

\[\text{cmom[..., n, m]} = \langle (x - \langle x \rangle)^n (y - \langle y \rangle)^m \rangle\]

where

\[\langle x \rangle = \sum_i w_i x_i / \sum_i w_i\]

Returns:

output (ndarray or DataArray)

to_raw(*, weight=None)

Raw moments accumulation array.

\[\begin{split}\text{raw[..., n, m]} = \begin{cases} \text{weight} & n = m = 0 \\ \langle x^n y ^m \rangle & \text{otherwise} \end{cases}\end{split}\]

where

\[\langle x \rangle = \sum_i w_i x_i / \sum_i w_i\]

Returns:

raw (ndarray or DataArray)

See also

wrap_raw, convert.moments_type

rmom()

Raw moments.

\[\text{rmom[..., n, m]} = \langle x^n y^m \rangle\]

where

\[\langle x \rangle = \sum_i w_i x_i / \sum_i w_i\]

Returns:

raw_moments (ndarray or DataArray)

See also

to_raw, convert.moments_type, cmom

abstractmethod classmethod zeros(*, mom)

Create a new base object.

Parameters:

mom (int or tuple of int) – Order or moments. If integer or length one tuple, then moments are for a single variable. If length 2 tuple, then comoments of two variables

Returns:

output (object) – New instance with zero values.

See also

numpy.zeros