secretflow.data#

Classes:

FedNdarray(partitions, partition_way)

Horizontal or vertical partitioned Ndarray.

PartitionWay(value)

The partitioning.
class secretflow.data.FedNdarray(partitions: Dict[PYU, PYUObject], partition_way: PartitionWay)[source]#

Bases: object

Horizontal or vertical partitioned Ndarray.

partitions#

List of references to local numpy.ndarray that makes up federated ndarray.

Type:

Dict[PYU, PYUObject]

Attributes:

partitions

partition_way

shape

Get shape of united ndarray.

Methods:

partition_shape()

Get ndarray shapes of all partitions.

partition_size()

Get ndarray sizes of all partitions.

astype(dtype[, order, casting, subok, copy])

Cast to a specified type.

__init__(partitions, partition_way)

partitions: Dict[PYU, PYUObject]#
partition_way: PartitionWay#
partition_shape()[source]#

Get ndarray shapes of all partitions.

partition_size()[source]#

Get ndarray sizes of all partitions.

property shape: Tuple[int, int]#

Get shape of united ndarray.

astype(dtype, order='K', casting='unsafe', subok=True, copy=True)[source]#

Cast to a specified type.

All args are same with numpy.ndarray.astype().

__init__(partitions: Dict[PYU, PYUObject], partition_way: PartitionWay) → None#
class secretflow.data.PartitionWay(value)[source]#

Bases: Enum

The partitioning. HORIZONTAL: horizontal partitioning. VERATICAL: vertical partitioning.

Attributes:

HORIZONTAL

VERTICAL

HORIZONTAL = 'horizontal'#
VERTICAL = 'vertical'#

secretflow.data.base#

Classes:

DataFrameBase()

Abstract base class for horizontal, vertical and mixed partitioned DataFrame

Partition([data])

Slice of data that makes up horizontal, vertical and mixed partitioned DataFrame.
class secretflow.data.base.DataFrameBase[source]#

Bases: ABC

Abstract base class for horizontal, vertical and mixed partitioned DataFrame

Methods:

min()

Gets minimum value of all columns

max()

Gets maximum value of all columns

count()

Gets number of rows

values()

Get underlying ndarray
abstract min()[source]#

Gets minimum value of all columns

abstract max()[source]#

Gets maximum value of all columns

abstract count()[source]#

Gets number of rows

abstract values()[source]#

Get underlying ndarray

class secretflow.data.base.Partition(data: Optional[PYUObject] = None)[source]#

Bases: DataFrameBase

Slice of data that makes up horizontal, vertical and mixed partitioned DataFrame.

data#

Reference to pandas.DataFrame located in local node.

Type:

PYUObject

Attributes:

data

values

Returns the underlying ndarray.

index

Returns the index (row labels) of the DataFrame.

dtypes

Returns the dtypes in the DataFrame.

columns

Returns the column labels of the DataFrame.

shape

Returns a tuple representing the dimensionality of the DataFrame.

Methods:

mean(*args, **kwargs)

Returns the mean of the values over the requested axis.

var(*args, **kwargs)

Returns the variance of the values over the requested axis.

std(*args, **kwargs)

Returns the standard deviation of the values over the requested axis.

sem(*args, **kwargs)

Returns the standard error of the mean over the requested axis.

skew(*args, **kwargs)

Returns the skewness over the requested axis.

kurtosis(*args, **kwargs)

Returns the kurtosis over the requested axis.

sum(*args, **kwargs)

Returns the sum of the values over the requested axis.

replace(*args, **kwargs)

Replace values given in to_replace with value.

quantile([q, axis])

Returns values at the given quantile over requested axis.

min(*args, **kwargs)

Returns the minimum of the values over the requested axis.

mode(*args, **kwargs)

Returns the mode of the values over the requested axis.

max(*args, **kwargs)

Returns the maximum of the values over the requested axis.

count(*args, **kwargs)

Counts non-NA cells for each column or row.

isna()

Detects missing values for an array-like object. Same as pandas.DataFrame.isna Returns DataFrame: Mask of bool values for each element in DataFrame that indicates whether an element is an NA value.

pow(*args, **kwargs)

Gets Exponential power of (partition of) dataframe and other, element-wise (binary operator pow).

subtract(*args, **kwargs)

Gets Subtraction of (partition of) dataframe and other, element-wise (binary operator sub).

round(*args, **kwargs)

Round the (partition of) DataFrame to a variable number of decimal places.

select_dtypes(*args, **kwargs)

Returns a subset of the DataFrame's columns based on the column dtypes.

astype(dtype[, copy, errors])

Cast a pandas object to a specified dtype dtype.

iloc(index)

Integer-location based indexing for selection by position.

drop([labels, axis, index, columns, level, ...])

See pandas.DataFrame.drop

fillna([value, method, axis, inplace, ...])

See pandas.DataFrame.fillna()

rename([mapper, index, columns, axis, copy, ...])

See pandas.DataFrame.rename()

value_counts(*args, **kwargs)

Return a Series containing counts of unique values.

to_csv(filepath, **kwargs)

Save DataFrame to csv file.

copy()

Shallow copy.

__init__([data])

data: PYUObject = None#
mean(*args, **kwargs) → Partition[source]#

Returns the mean of the values over the requested axis.

Returns:

mean values series.

Return type:

Partition

var(*args, **kwargs) → Partition[source]#

Returns the variance of the values over the requested axis.

Returns:

variance values series.

Return type:

Partition

std(*args, **kwargs) → Partition[source]#

Returns the standard deviation of the values over the requested axis.

Returns:

standard deviation values series.

Return type:

Partition

sem(*args, **kwargs) → Partition[source]#

Returns the standard error of the mean over the requested axis.

Returns:

standard error of the mean series.

Return type:

Partition

skew(*args, **kwargs) → Partition[source]#

Returns the skewness over the requested axis.

Returns:

skewness series.

Return type:

Partition

kurtosis(*args, **kwargs) → Partition[source]#

Returns the kurtosis over the requested axis.

Returns:

kurtosis series.

Return type:

Partition

sum(*args, **kwargs) → Partition[source]#

Returns the sum of the values over the requested axis.

Returns:

sum values series.

Return type:

Partition

replace(*args, **kwargs) → Partition[source]#

Replace values given in to_replace with value. Same as pandas.DataFrame.replace Values of the DataFrame are replaced with other values dynamically.

Returns:

same shape except value replaced

Return type:

Partition

quantile(q=0.5, axis=0) → Partition[source]#

Returns values at the given quantile over requested axis.

Returns:

quantile values series.

Return type:

Partition

min(*args, **kwargs) → Partition[source]#

Returns the minimum of the values over the requested axis.

Returns:

minimum values series.

Return type:

Partition

mode(*args, **kwargs) → Partition[source]#

Returns the mode of the values over the requested axis.

For data protection reasons, only one mode will be returned.

Returns:

mode values series.

Return type:

Partition

max(*args, **kwargs) → Partition[source]#

Returns the maximum of the values over the requested axis.

Returns:

maximum values series.

Return type:

Partition

count(*args, **kwargs) → Partition[source]#

Counts non-NA cells for each column or row.

Returns:

count values series.

Return type:

Partition

isna() → Partition[source]#

Detects missing values for an array-like object. Same as pandas.DataFrame.isna Returns

DataFrame: Mask of bool values for each element in DataFrame

that indicates whether an element is an NA value.

pow(*args, **kwargs) → Partition[source]#

Gets Exponential power of (partition of) dataframe and other, element-wise (binary operator pow). Equivalent to dataframe ** other, but with support to substitute a fill_value for missing data in one of the inputs. With reverse version, rpow. Among flexible wrappers (add, sub, mul, div, mod, pow) to arithmetic operators: +, -, , /, //, %, *.

Reference:

pd.DataFrame.pow

subtract(*args, **kwargs) → Partition[source]#

Gets Subtraction of (partition of) dataframe and other, element-wise (binary operator sub). Equivalent to dataframe - other, but with support to substitute a fill_value for missing data in one of the inputs. With reverse version, rsub. Among flexible wrappers (add, sub, mul, div, mod, pow) to arithmetic operators: +, -, , /, //, %, *.

Reference:

pd.DataFrame.subtract

round(*args, **kwargs) → Partition[source]#

Round the (partition of) DataFrame to a variable number of decimal places.

Reference:

pd.DataFrame.round

select_dtypes(*args, **kwargs) → Partition[source]#

Returns a subset of the DataFrame’s columns based on the column dtypes.

Reference:

pandas.DataFrame.select_dtypes

property values#

Returns the underlying ndarray.

property index#

Returns the index (row labels) of the DataFrame.

property dtypes#

Returns the dtypes in the DataFrame.

astype(dtype, copy: bool = True, errors: str = 'raise')[source]#

Cast a pandas object to a specified dtype dtype.

All args are same as pandas.DataFrame.astype().

property columns#

Returns the column labels of the DataFrame.

property shape#

Returns a tuple representing the dimensionality of the DataFrame.

iloc(index: Union[int, slice, List[int]]) → Partition[source]#

Integer-location based indexing for selection by position.

Parameters:

index (Union[int, slice, List[int]]) – rows index.

Returns:

Selected DataFrame.

Return type:

Partition

drop(labels=None, axis=0, index=None, columns=None, level=None, inplace=False, errors='raise') → Optional[Partition][source]#

See pandas.DataFrame.drop

fillna(value=None, method=None, axis=None, inplace=False, limit=None, downcast=None) → Optional[Partition][source]#

See pandas.DataFrame.fillna()

rename(mapper=None, index=None, columns=None, axis=None, copy=True, inplace=False, level=None, errors='ignore') → Optional[Partition][source]#

See pandas.DataFrame.rename()

value_counts(*args, **kwargs) → Partition[source]#

Return a Series containing counts of unique values.

to_csv(filepath, **kwargs)[source]#

Save DataFrame to csv file.

copy()[source]#

Shallow copy.

__init__(data: Optional[PYUObject] = None) → None#

secretflow.data.math_utils#

Functions:

sum_of_difference_squares(y1, y2)

mean_of_difference_squares(y1, y2)

sum_of_difference_abs(y1, y2)

mean_of_difference_abs(y1, y2)

mean_of_difference_ratio_abs(y1, y2)

sum_of_difference_ratio_abs(y1, y2)

secretflow.data.math_utils.sum_of_difference_squares(y1, y2)[source]#
secretflow.data.math_utils.mean_of_difference_squares(y1, y2)[source]#
secretflow.data.math_utils.sum_of_difference_abs(y1, y2)[source]#
secretflow.data.math_utils.mean_of_difference_abs(y1, y2)[source]#
secretflow.data.math_utils.mean_of_difference_ratio_abs(y1, y2)[source]#
secretflow.data.math_utils.sum_of_difference_ratio_abs(y1, y2)[source]#

secretflow.data.ndarray#

Classes:

PartitionWay(value)

The partitioning.

FedNdarray(partitions, partition_way)

Horizontal or vertical partitioned Ndarray.

Functions:

subtract(y1, y2[, spu_device])

subtraction of two FedNdarray object

load(sources[, partition_way, allow_pickle, ...])

Load FedNdarray from data source.

shuffle(data)

Random shuffle data.

check_same_partition_shapes(a1, a2)

unary_op(handle_function, ...[, spu_device, ...])

mean(y[, spu_device])

Mean of all elements :param y: FedNdarray :param spu_device: SPU

binary_op(handle_function, ...[, spu_device])

get_concat_axis(y)

rss(y1, y2[, spu_device])

Residual Sum of Squares of all elements

tss(y[, spu_device])

Total Sum of Square (Variance) of all elements

mean_squared_error(y_true, y_pred[, spu_device])

Mean Squared Error of all elements

root_mean_squared_error(y_true, y_pred[, ...])

Root Mean Squared Error of all elements

mean_abs_err(y_true, y_pred[, spu_device])

Mean Absolute Error

mean_abs_percent_err(y_true, y_pred[, ...])

Mean Absolute Percentage Error

r2_score(y_true, y_pred[, spu_device])

R2 Score

histogram(y[, bins, spu_device])

Histogram of all elements a restricted version of the counterpart in numpy

residual_histogram(y1, y2[, bins, spu_device])

Histogram of residuals of y1 - y2
class secretflow.data.ndarray.PartitionWay(value)[source]#

Bases: Enum

The partitioning. HORIZONTAL: horizontal partitioning. VERATICAL: vertical partitioning.

Attributes:

HORIZONTAL

VERTICAL

HORIZONTAL = 'horizontal'#
VERTICAL = 'vertical'#
class secretflow.data.ndarray.FedNdarray(partitions: Dict[PYU, PYUObject], partition_way: PartitionWay)[source]#

Bases: object

Horizontal or vertical partitioned Ndarray.

partitions#

List of references to local numpy.ndarray that makes up federated ndarray.

Type:

Dict[PYU, PYUObject]

Attributes:

partitions

partition_way

shape

Get shape of united ndarray.

Methods:

partition_shape()

Get ndarray shapes of all partitions.

partition_size()

Get ndarray sizes of all partitions.

astype(dtype[, order, casting, subok, copy])

Cast to a specified type.

__init__(partitions, partition_way)

partitions: Dict[PYU, PYUObject]#
partition_way: PartitionWay#
partition_shape()[source]#

Get ndarray shapes of all partitions.

partition_size()[source]#

Get ndarray sizes of all partitions.

property shape: Tuple[int, int]#

Get shape of united ndarray.

astype(dtype, order='K', casting='unsafe', subok=True, copy=True)[source]#

Cast to a specified type.

All args are same with numpy.ndarray.astype().

__init__(partitions: Dict[PYU, PYUObject], partition_way: PartitionWay) → None#
secretflow.data.ndarray.subtract(y1: FedNdarray, y2: FedNdarray, spu_device: Optional[SPU] = None)[source]#

subtraction of two FedNdarray object

Args:

y1: FedNdarray y2: FedNdarray spu_device: Optional SPU device

Returns:

result of subtraction

as long as they have the same shape, the result is computable. They may have different partition shapes.

secretflow.data.ndarray.load(sources: Dict[PYU, Union[str, Callable[[], ndarray], PYUObject]], partition_way: PartitionWay = PartitionWay.VERTICAL, allow_pickle=False, encoding='ASCII') → FedNdarray[source]#

Load FedNdarray from data source.

Warning

Loading files that contain object arrays uses the pickle module, which is not secure against erroneous or maliciously constructed data. Consider passing allow_pickle=False to load data that is known not to contain object arrays for the safer handling of untrusted sources.

Parameters:

  • sources – Data source in each partition. Shall be one of the followings. 1) Loaded numpy.ndarray. 2) Local filepath which should be .npy or .npz file. 3) Callable function that return numpy.ndarray.

  • allow_pickle – Allow loading pickled object arrays stored in npy files.

  • encoding – What encoding to use when reading Python 2 strings.

Raises:

TypeError – illegal source。

Returns:

Returns a FedNdarray if source is pyu object or .npy. Or return a dict {key: FedNdarray} if source is .npz.

Examples

>>> fed_arr = load({'alice': 'example/alice.csv', 'bob': 'example/alice.csv'})
secretflow.data.ndarray.shuffle(data: FedNdarray)[source]#

Random shuffle data.

Parameters:

data – data to be shuffled.

secretflow.data.ndarray.check_same_partition_shapes(a1: FedNdarray, a2: FedNdarray)[source]#
secretflow.data.ndarray.unary_op(handle_function: Callable, len_1_handle_function: Callable, y: FedNdarray, spu_device: Optional[SPU] = None, simulate_double_value_replacer_handle: Optional[Callable] = None)[source]#
secretflow.data.ndarray.mean(y: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Mean of all elements :param y: FedNdarray :param spu_device: SPU

If y is from a single party, then a PYUObject is returned. If y is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y is empty return 0.

secretflow.data.ndarray.binary_op(handle_function: Callable, len_1_handle_function: Callable, y1: FedNdarray, y2: FedNdarray, spu_device: Optional[SPU] = None)[source]#
secretflow.data.ndarray.get_concat_axis(y: FedNdarray) → int[source]#
secretflow.data.ndarray.rss(y1: FedNdarray, y2: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Residual Sum of Squares of all elements

more detail for rss: https://en.wikipedia.org/wiki/Residual_sum_of_squares

Parameters:

  • y1 – FedNdarray

  • y2 – FedNdarray

  • spu_device – SPU

y1 and y2 must have the same device and partition shapes

If y1 is from a single party, then a PYUObject is returned. If y1 is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y1 is empty return 0.

secretflow.data.ndarray.tss(y: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Total Sum of Square (Variance) of all elements

more detail for tss: https://en.wikipedia.org/wiki/Total_sum_of_squares

Parameters:

y – FedNdarray

If y is from a single party, then a PYUObject is returned. If y is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y is empty return 0.

secretflow.data.ndarray.mean_squared_error(y_true: FedNdarray, y_pred: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Mean Squared Error of all elements

more detail for mse: https://en.wikipedia.org/wiki/Mean_squared_error

Parameters:

  • y_true – FedNdarray

  • y_pred – FedNdarray

  • spu_device – SPU

y_true and y_pred must have the same device and partition shapes

If y_true is from a single party, then a PYUObject is returned. If y_true is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y_true is empty return 0.

secretflow.data.ndarray.root_mean_squared_error(y_true: FedNdarray, y_pred: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Root Mean Squared Error of all elements

more detail for mse: https://en.wikipedia.org/wiki/Root-mean-square_deviation

Parameters:

  • y_true – FedNdarray

  • y_pred – FedNdarray

  • spu_device – SPU

y_true and y_pred must have the same device and partition shapes

If y_true is from a single party, then a PYUObject is returned. If y_true is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y_true is empty return 0.

secretflow.data.ndarray.mean_abs_err(y_true: FedNdarray, y_pred: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Mean Absolute Error

more detail for mean abs err: https://en.wikipedia.org/wiki/Mean_absolute_error

Parameters:

  • y_true – FedNdarray

  • y_pred – FedNdarray

y_true and y_pred must have the same device and partition shapes

If y_true is from a single party, then a PYUObject is returned. If y_true is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y_true is empty return 0.

secretflow.data.ndarray.mean_abs_percent_err(y_true: FedNdarray, y_pred: FedNdarray, spu_device: Optional[SPU] = None)[source]#

Mean Absolute Percentage Error

more detail for mean percent err: https://en.wikipedia.org/wiki/Mean_absolute_percentage_error

Parameters:

  • y_true – FedNdarray

  • y_pred – FedNdarray

y_true and y_pred must have the same device and partition shapes

If y_true is from a single party, then a PYUObject is returned. If y_true is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y_true is empty return 0.

secretflow.data.ndarray.r2_score(y_true: FedNdarray, y_pred: FedNdarray, spu_device: Optional[SPU] = None)[source]#

R2 Score

more detail for r2 score: https://en.wikipedia.org/wiki/Coefficient_of_determination

Parameters:

  • y_true – FedNdarray

  • y_pred – FedNdarray

y_true and y_pred must have the same device and partition shapes

If y_true is from a single party, then a PYUObject is returned. If y_true is from multiple parties, then

an SPU device is required and an SPUObject is returned.

If y_true is empty return 0.

secretflow.data.ndarray.histogram(y: FedNdarray, bins: int = 10, spu_device: Optional[SPU] = None)[source]#

Histogram of all elements a restricted version of the counterpart in numpy

more detail for histogram: https://numpy.org/doc/stable/reference/generated/numpy.histogram.html

Parameters:

y – FedNdarray

If y is from a single party, then a PYUObject is returned. If y is from multiple parties, then

an SPU device is required and an SPUObject is returned.

secretflow.data.ndarray.residual_histogram(y1: FedNdarray, y2: FedNdarray, bins: int = 10, spu_device: Optional[SPU] = None)[source]#

Histogram of residuals of y1 - y2

Support histogram(y1 - y2) equivalent function even if y1 and y2 have distinct partition shapes.

Parameters:

  • y1 – FedNdarray

  • y2 – FedNdarray

If y is from a single party, then a PYUObject is returned. If y is from multiple parties, then

an SPU device is required and an SPUObject is returned.

secretflow.data.split#

Functions:

train_test_split(data[, test_size, ...])

Split data into train and test dataset.
secretflow.data.split.train_test_split(data: Union[VDataFrame, HDataFrame, FedNdarray], test_size=None, train_size=None, random_state=1234, shuffle=True) → Tuple[object, object][source]#

Split data into train and test dataset.

Parameters:

  • data – DataFrame to split, supported are: VDataFrame,HDataFrame,FedNdarray.

  • test_size (float) – test dataset size, default is None.

  • train_size (float) – train dataset size, default is None.

  • random_state (int) – Controls the shuffling applied to the data before applying the split.

  • shuffle (bool) – Whether or not to shuffle the data before splitting, default is True.

Returns

splitting : list, length=2 * len(arrays)

Examples

>>> import numpy as np
>>> from secret.data.split import train_test_split
>>> # FedNdarray
>>> alice_arr = alice(lambda: np.array([[1, 2, 3], [4, 5, 6]]))()
>>> bob_arr = bob(lambda: np.array([[11, 12, 13], [14, 15, 16]]))()

>>> fed_arr = load({self.alice: alice_arr, self.bob: bob_arr})
>>>
>>> X_train, X_test = train_test_split(
...  fed_arr, test_size=0.33, random_state=42)
...
>>> VDataFrame
>>> df_alice = pd.DataFrame({'a1': ['K5', 'K1', None, 'K6'],
...                          'a2': ['A5', 'A1', 'A2', 'A6'],
...                          'a3': [5, 1, 2, 6]})

>>> df_bob = pd.DataFrame({'b4': [10.2, 20.5, None, -0.4],
...                        'b5': ['B3', None, 'B9', 'B4'],
...                        'b6': [3, 1, 9, 4]})
>>> df_alice = df_alice
>>> df_bob = df_bob
>>> vdf = VDataFrame(
...       {alice: Partition(data=cls.alice(lambda: df_alice)()),
...          bob: Partition(data=cls.bob(lambda: df_bob)())})
>>> train_vdf, test_vdf = train_test_split(vdf, test_size=0.33, random_state=42)