secretflow.data.vertical package

Submodules

secretflow.data.vertical.dataframe module

Classes:

VDataFrame(partitions[, aligned])

Federated dataframe holds vertical partitioned data.

class secretflow.data.vertical.dataframe.VDataFrame(partitions: Dict[PYU, Partition], aligned: bool = True)

Bases: DataFrameBase

Federated dataframe holds vertical partitioned data.

This dataframe is design to provide a federated pandas dataframe and just same as using pandas. The original data is still stored locally in the data holder and is not transmitted out of the domain during all the methods execution.

The method with a prefix partition_ will return a dict {pyu of partition: result of partition}.

partitions

a dict of pyu and partition.

Type

Dict[secretflow.device.device.pyu.PYU, secretflow.data.base.Partition]

aligned

a boolean indicating whether the data is

Type

bool

Examples

>>> from secretflow.data.vertical import read_csv
>>> from secretflow import PYU
>>> alice = PYU('alice')
>>> bob = PYU('bob')
>>> v_df = read_csv({alice: 'alice.csv', bob: 'bob.csv'})
>>> v_df.columns
Index(['sepal_length', 'sepal_width', 'petal_length', 'petal_width', 'class'], dtype='object')
>>> v_df.mean(numeric_only=True)
sepal_length    5.827693
sepal_width     3.054000
petal_length    3.730000
petal_width     1.198667
dtype: float64
>>> v_df.min(numeric_only=True)
sepal_length    4.3
sepal_width     2.0
petal_length    1.0
petal_width     0.1
dtype: float64
>>> v_df.max(numeric_only=True)
sepal_length    7.9
sepal_width     4.4
petal_length    6.9
petal_width     2.5
dtype: float64
>>> v_df.count()
sepal_length    130
sepal_width     150
petal_length    120
petal_width     150
class           150
dtype: int64
>>> v_df.fillna({'sepal_length': 2})

Attributes:

partitions
aligned
dtypes	Return the dtypes in the DataFrame.
columns	The column labels of the DataFrame.
shape	Return a tuple representing the dimensionality of the DataFrame.
values	Return a federated Numpy representation of the DataFrame.
partition_columns	Returns columns of each partition.

Methods:

mode([numeric_only, dropna])	Return the mode of the values over the axis 0.
sum([numeric_only])	Return the sum of the values over the axis 0.
min([numeric_only])	Return the min of the values over the axis 0.
max([numeric_only])	Return the max of the values over the axis 0.
pow(*args, **kwargs)	Gets Exponential power of dataframe and other, element-wise (binary operator pow).
round(*args, **kwargs)	Round the 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.
replace(*args, **kwargs)	Replace values given in to_replace with value.
subtract(*args, **kwargs)	Gets Subtraction of dataframe and other, element-wise (binary operator sub).
astype(dtype[, copy, errors])	Cast object to a specified dtype dtype.
mean([numeric_only])	Return the mean of the values over the axis 0.
var([numeric_only])	Return the var of the values over the axis 0.
std([numeric_only])	Return the std of the values over the axis 0.
sem([numeric_only])	Return the standard error of the mean over the axis 0.
skew([numeric_only])	Return the skewness over the axis 0.
kurtosis([numeric_only])	Return the kurtosis over the requested axis.
quantile([q])	Returns values at the given quantile over axis 0.
count([numeric_only])	Count non-NA cells for each column.
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.
copy()	Shallow copy of this dataframe.
drop([labels, axis, index, columns, level, ...])	Drop specified labels from rows or columns.
fillna([value, method, axis, inplace, ...])	Fill NA/NaN values using the specified method.
to_csv(fileuris, **kwargs)	Write object to a comma-separated values (csv) file.
partition_shape()	Return shapes of each partition.
__init__(partitions[, aligned])

partitions: Dict[PYU, Partition]

aligned: bool = True

mode(numeric_only=False, dropna=True) → Series

Return the mode of the values over the axis 0. The mode of a set of values is the value that appears most often. Restrict mode on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

sum(numeric_only=False) → Series

Return the sum of the values over the axis 0.

Restrict sum on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

min(numeric_only=False) → Series

Return the min of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict min on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

max(numeric_only=False) → Series

Return the max of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict max on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

pow(*args, **kwargs) → VDataFrame

Gets Exponential power 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: +, -, , /, //, %, *.

Returns

VDataFrame

Reference:

pd.DataFrame.pow

round(*args, **kwargs) → VDataFrame

Round the DataFrame to a variable number of decimal places.

Returns

same shape except value rounded

Return type

VDataFrame

Reference:

pd.DataFrame.round

select_dtypes(*args, **kwargs) → VDataFrame

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

Reference:

pandas.DataFrame.select_dtypes

replace(*args, **kwargs) → VDataFrame

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

VDataFrame

subtract(*args, **kwargs) → VDataFrame

Gets Subtraction 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: +, -, , /, //, %, *.

Note each part only will contains its own columns.

Reference:

pd.DataFrame.subtract

property dtypes: Series

Return the dtypes in the DataFrame.

Returns

the data type of each column.

Return type

pd.Series

astype(dtype, copy: bool = True, errors: str = 'raise')

Cast object to a specified dtype dtype.

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

property columns

The column labels of the DataFrame.

property shape

Return a tuple representing the dimensionality of the DataFrame.

mean(numeric_only=False) → Series

Return the mean of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict mean on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

var(numeric_only=False) → Series

Return the var of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict var on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

std(numeric_only=False) → Series

Return the std of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict std on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

sem(numeric_only=False) → Series

Return the standard error of the mean over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict sem on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

skew(numeric_only=False) → Series

Return the skewness over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict skew on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

kurtosis(numeric_only=False) → Series

Return the kurtosis over the requested axis.

Note columns containing None values are ignored. Fill before proceed.

Restrict kurtosis on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

quantile(q=0.5) → Series

Returns values at the given quantile over axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict quantile on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

count(numeric_only=False) → Series

Count non-NA cells for each column.

Restrict count on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

isna() → VDataFrame

“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.

Returns

VDataFrame

Reference:

pd.DataFrame.isna

property values

Return a federated Numpy representation of the DataFrame.

Returns

FedNdarray.

copy() → VDataFrame

Shallow copy of this dataframe.

Returns

VDataFrame.

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

Drop specified labels from rows or columns.

All arguments are same with pandas.DataFrame.drop().

Returns

VDataFrame without the removed index or column labels or None if inplace=True.

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

Fill NA/NaN values using the specified method.

All arguments are same with pandas.DataFrame.fillna().

Returns

VDataFrame with missing values filled or None if inplace=True.

to_csv(fileuris: Dict[PYU, str], **kwargs)

Write object to a comma-separated values (csv) file.

Parameters

• fileuris – a dict of file uris specifying file for each PYU.

• kwargs – other arguments are same with pandas.DataFrame.to_csv().

Returns

Returns a list of PYUObjects whose value is none. You can use secretflow.wait to wait for the save to complete.

partition_shape()

Return shapes of each partition.

Returns

shape}

Return type

a dict of {pyu

property partition_columns

Returns columns of each partition.

Returns

columns}

Return type

a dict of {pyu

__init__(partitions: Dict[PYU, Partition], aligned: bool = True) → None

secretflow.data.vertical.io module

Functions:

read_csv(filepath[, delimiter, dtypes, spu, ...])	Read a comma-separated values (csv) file into VDataFrame.
to_csv(df, file_uris, **kwargs)	Write object to a comma-separated values (csv) file.

secretflow.data.vertical.io.read_csv(filepath: Dict[PYU, str], delimiter=',', dtypes: Optional[Dict[PYU, Dict[str, type]]] = None, spu: Optional[SPU] = None, keys: Optional[Union[str, List[str], Dict[Device, List[str]]]] = None, drop_keys: Optional[Union[str, List[str], Dict[Device, List[str]]]] = None, psi_protocl=None) → VDataFrame

Read a comma-separated values (csv) file into VDataFrame.

When specifying spu and keys, the fields specified by keys are used for PSI alignment.Fields used for alignment must be common to all parties, and other fields cannot be repeated across parties. The data for each party is supposed pre-aligned if not specifying spu and keys.

Parameters

• filepath –

  The file path of each party. It can be a local file with a relative or absolute path, or a remote file starting with oss://, http(s)://, E.g.

  {
      PYU('alice'): 'oss://bucket/data/alice.csv',
      PYU('bob'): 'oss://bucket/data/bob.csv'
  }
  

• delimiter – the file separator.

• dtypes –

  Participant field type. It will be inferred from the file if not specified, E.g.

  {
      PYU('alice'): {'uid': np.str, 'age': np.int32},
      PYU('bob'): {'uid': np.str, 'score': np.float32}
  }
  

• spu – SPU device, used for PSI data alignment. The data of all parties are supposed pre-aligned if not specified.

• keys – The field used for psi, which can be single or multiple fields. This parameter is required when spu is specified.

• drop_keys – keys to removed, which can be single or multiple fields. This parameter is required when spu is specified since VDataFrame doesn’t allow duplicate column names.

• psi_protocl – Specified protocol for PSI. Default ‘KKRT_PSI_2PC’ for 2 parties, ‘ECDH_PSI_3PC’ for 3 parties.

Returns

A aligned VDataFrame.

secretflow.data.vertical.io.to_csv(df: VDataFrame, file_uris: Dict[PYU, str], **kwargs)

Write object to a comma-separated values (csv) file.

Parameters

• df – the VDataFrame to save.

• file_uris – the file path of each PYU.

• kwargs – all other arguments are same with pandas.DataFrame.to_csv().

Module contents

Classes:

VDataFrame(partitions[, aligned])

Federated dataframe holds vertical partitioned data.

Functions:

read_csv(filepath[, delimiter, dtypes, spu, ...])	Read a comma-separated values (csv) file into VDataFrame.
to_csv(df, file_uris, **kwargs)	Write object to a comma-separated values (csv) file.

class secretflow.data.vertical.VDataFrame(partitions: Dict[PYU, Partition], aligned: bool = True)

Bases: DataFrameBase

Federated dataframe holds vertical partitioned data.

This dataframe is design to provide a federated pandas dataframe and just same as using pandas. The original data is still stored locally in the data holder and is not transmitted out of the domain during all the methods execution.

The method with a prefix partition_ will return a dict {pyu of partition: result of partition}.

partitions

a dict of pyu and partition.

Type

Dict[secretflow.device.device.pyu.PYU, secretflow.data.base.Partition]

aligned

a boolean indicating whether the data is

Type

bool

Examples

>>> from secretflow.data.vertical import read_csv
>>> from secretflow import PYU
>>> alice = PYU('alice')
>>> bob = PYU('bob')
>>> v_df = read_csv({alice: 'alice.csv', bob: 'bob.csv'})
>>> v_df.columns
Index(['sepal_length', 'sepal_width', 'petal_length', 'petal_width', 'class'], dtype='object')
>>> v_df.mean(numeric_only=True)
sepal_length    5.827693
sepal_width     3.054000
petal_length    3.730000
petal_width     1.198667
dtype: float64
>>> v_df.min(numeric_only=True)
sepal_length    4.3
sepal_width     2.0
petal_length    1.0
petal_width     0.1
dtype: float64
>>> v_df.max(numeric_only=True)
sepal_length    7.9
sepal_width     4.4
petal_length    6.9
petal_width     2.5
dtype: float64
>>> v_df.count()
sepal_length    130
sepal_width     150
petal_length    120
petal_width     150
class           150
dtype: int64
>>> v_df.fillna({'sepal_length': 2})

Attributes:

partitions
aligned
dtypes	Return the dtypes in the DataFrame.
columns	The column labels of the DataFrame.
shape	Return a tuple representing the dimensionality of the DataFrame.
values	Return a federated Numpy representation of the DataFrame.
partition_columns	Returns columns of each partition.

Methods:

mode([numeric_only, dropna])	Return the mode of the values over the axis 0.
sum([numeric_only])	Return the sum of the values over the axis 0.
min([numeric_only])	Return the min of the values over the axis 0.
max([numeric_only])	Return the max of the values over the axis 0.
pow(*args, **kwargs)	Gets Exponential power of dataframe and other, element-wise (binary operator pow).
round(*args, **kwargs)	Round the 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.
replace(*args, **kwargs)	Replace values given in to_replace with value.
subtract(*args, **kwargs)	Gets Subtraction of dataframe and other, element-wise (binary operator sub).
astype(dtype[, copy, errors])	Cast object to a specified dtype dtype.
mean([numeric_only])	Return the mean of the values over the axis 0.
var([numeric_only])	Return the var of the values over the axis 0.
std([numeric_only])	Return the std of the values over the axis 0.
sem([numeric_only])	Return the standard error of the mean over the axis 0.
skew([numeric_only])	Return the skewness over the axis 0.
kurtosis([numeric_only])	Return the kurtosis over the requested axis.
quantile([q])	Returns values at the given quantile over axis 0.
count([numeric_only])	Count non-NA cells for each column.
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.
copy()	Shallow copy of this dataframe.
drop([labels, axis, index, columns, level, ...])	Drop specified labels from rows or columns.
fillna([value, method, axis, inplace, ...])	Fill NA/NaN values using the specified method.
to_csv(fileuris, **kwargs)	Write object to a comma-separated values (csv) file.
partition_shape()	Return shapes of each partition.
__init__(partitions[, aligned])

partitions: Dict[PYU, Partition]

aligned: bool = True

mode(numeric_only=False, dropna=True) → Series

Return the mode of the values over the axis 0. The mode of a set of values is the value that appears most often. Restrict mode on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

sum(numeric_only=False) → Series

Return the sum of the values over the axis 0.

Restrict sum on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

min(numeric_only=False) → Series

Return the min of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict min on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

max(numeric_only=False) → Series

Return the max of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict max on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

pow(*args, **kwargs) → VDataFrame

Gets Exponential power 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: +, -, , /, //, %, *.

Returns

VDataFrame

Reference:

pd.DataFrame.pow

round(*args, **kwargs) → VDataFrame

Round the DataFrame to a variable number of decimal places.

Returns

same shape except value rounded

Return type

VDataFrame

Reference:

pd.DataFrame.round

select_dtypes(*args, **kwargs) → VDataFrame

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

Reference:

pandas.DataFrame.select_dtypes

replace(*args, **kwargs) → VDataFrame

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

VDataFrame

subtract(*args, **kwargs) → VDataFrame

Gets Subtraction 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: +, -, , /, //, %, *.

Note each part only will contains its own columns.

Reference:

pd.DataFrame.subtract

property dtypes: Series

Return the dtypes in the DataFrame.

Returns

the data type of each column.

Return type

pd.Series

astype(dtype, copy: bool = True, errors: str = 'raise')

Cast object to a specified dtype dtype.

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

property columns

The column labels of the DataFrame.

property shape

Return a tuple representing the dimensionality of the DataFrame.

mean(numeric_only=False) → Series

Return the mean of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict mean on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

var(numeric_only=False) → Series

Return the var of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict var on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

std(numeric_only=False) → Series

Return the std of the values over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict std on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

sem(numeric_only=False) → Series

Return the standard error of the mean over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict sem on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

skew(numeric_only=False) → Series

Return the skewness over the axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict skew on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

kurtosis(numeric_only=False) → Series

Return the kurtosis over the requested axis.

Note columns containing None values are ignored. Fill before proceed.

Restrict kurtosis on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

quantile(q=0.5) → Series

Returns values at the given quantile over axis 0.

Note columns containing None values are ignored. Fill before proceed.

Restrict quantile on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

count(numeric_only=False) → Series

Count non-NA cells for each column.

Restrict count on axis 0 in VDataFrame for data protection reasons.

Returns

pd.Series

isna() → VDataFrame

“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.

Returns

VDataFrame

Reference:

pd.DataFrame.isna

property values

Return a federated Numpy representation of the DataFrame.

Returns

FedNdarray.

copy() → VDataFrame

Shallow copy of this dataframe.

Returns

VDataFrame.

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

Drop specified labels from rows or columns.

All arguments are same with pandas.DataFrame.drop().

Returns

VDataFrame without the removed index or column labels or None if inplace=True.

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

Fill NA/NaN values using the specified method.

All arguments are same with pandas.DataFrame.fillna().

Returns

VDataFrame with missing values filled or None if inplace=True.

to_csv(fileuris: Dict[PYU, str], **kwargs)

Write object to a comma-separated values (csv) file.

Parameters

• fileuris – a dict of file uris specifying file for each PYU.

• kwargs – other arguments are same with pandas.DataFrame.to_csv().

Returns

Returns a list of PYUObjects whose value is none. You can use secretflow.wait to wait for the save to complete.

partition_shape()

Return shapes of each partition.

Returns

shape}

Return type

a dict of {pyu

property partition_columns

Returns columns of each partition.

Returns

columns}

Return type

a dict of {pyu

__init__(partitions: Dict[PYU, Partition], aligned: bool = True) → None

secretflow.data.vertical.read_csv(filepath: Dict[PYU, str], delimiter=',', dtypes: Optional[Dict[PYU, Dict[str, type]]] = None, spu: Optional[SPU] = None, keys: Optional[Union[str, List[str], Dict[Device, List[str]]]] = None, drop_keys: Optional[Union[str, List[str], Dict[Device, List[str]]]] = None, psi_protocl=None) → VDataFrame

Read a comma-separated values (csv) file into VDataFrame.

When specifying spu and keys, the fields specified by keys are used for PSI alignment.Fields used for alignment must be common to all parties, and other fields cannot be repeated across parties. The data for each party is supposed pre-aligned if not specifying spu and keys.

Parameters

• filepath –

  The file path of each party. It can be a local file with a relative or absolute path, or a remote file starting with oss://, http(s)://, E.g.

  {
      PYU('alice'): 'oss://bucket/data/alice.csv',
      PYU('bob'): 'oss://bucket/data/bob.csv'
  }
  

• delimiter – the file separator.

• dtypes –

  Participant field type. It will be inferred from the file if not specified, E.g.

  {
      PYU('alice'): {'uid': np.str, 'age': np.int32},
      PYU('bob'): {'uid': np.str, 'score': np.float32}
  }
  

• spu – SPU device, used for PSI data alignment. The data of all parties are supposed pre-aligned if not specified.

• keys – The field used for psi, which can be single or multiple fields. This parameter is required when spu is specified.

• drop_keys – keys to removed, which can be single or multiple fields. This parameter is required when spu is specified since VDataFrame doesn’t allow duplicate column names.

• psi_protocl – Specified protocol for PSI. Default ‘KKRT_PSI_2PC’ for 2 parties, ‘ECDH_PSI_3PC’ for 3 parties.

Returns

A aligned VDataFrame.

secretflow.data.vertical.to_csv(df: VDataFrame, file_uris: Dict[PYU, str], **kwargs)

Write object to a comma-separated values (csv) file.

Parameters

• df – the VDataFrame to save.

• file_uris – the file path of each PYU.

• kwargs – all other arguments are same with pandas.DataFrame.to_csv().