secretflow.preprocessing.binning.kernels

secretflow.preprocessing.binning.kernels.base_binning

Classes:

BaseBinning(bin_names, bin_indexes, bin_num, ...)

class secretflow.preprocessing.binning.kernels.base_binning.BaseBinning(bin_names: List, bin_indexes: List, bin_num: int, abnormal_list: List)

基类：ABC

Methods:

__init__(bin_names, bin_indexes, bin_num, ...)
fit_split_points(data)

Attributes:

split_points

__init__(bin_names: List, bin_indexes: List, bin_num: int, abnormal_list: List)

property split_points

abstract fit_split_points(data)

secretflow.preprocessing.binning.kernels.quantile_binning

Classes:

QuantileBinning([bin_num, compress_thres, ...])

Use QuantileSummary algorithm for constant frequency binning

class secretflow.preprocessing.binning.kernels.quantile_binning.QuantileBinning(bin_num: int = 10, compress_thres: int = 10000, head_size: int = 10000, error: float = 0.0001, bin_indexes: List[int] = [], bin_names: List[str] = [], local_only: bool = False, abnormal_list: Optional[List[str]] = None, allow_duplicate: bool = False)

基类：BaseBinning

Use QuantileSummary algorithm for constant frequency binning

bin_num

the num of buckets

compress_thres

if size of summary greater than compress_thres, do compress operation

cols_dict

mapping of value to index. {key: col_name , value: index}.

head_size

buffer size

error

0 <= error < 1 default: 0.001,error tolerance, floor((p - 2 * error) * N) <= rank(x) <= ceil((p + 2 * error) * N)

abnormal_list

list of anomaly features.

summary_dict

a dict store summary of each features

col_name_maps

a dict store column index to name

bin_idx_name

a dict store index to name

allow_duplicate

Whether duplication is allowed

Methods:

__init__([bin_num, compress_thres, ...])
fit_split_points(data_frame)	calculate bin split points base on QuantileSummary algorithm
feature_summary(data_frame, compress_thres, ...)	calculate summary

__init__(bin_num: int = 10, compress_thres: int = 10000, head_size: int = 10000, error: float = 0.0001, bin_indexes: List[int] = [], bin_names: List[str] = [], local_only: bool = False, abnormal_list: Optional[List[str]] = None, allow_duplicate: bool = False)

fit_split_points(data_frame: DataFrame) → DataFrame

calculate bin split points base on QuantileSummary algorithm

参数:

data_frame – input data

返回:

bin result returned as dataframe

返回类型:

bin_result

static feature_summary(data_frame: DataFrame, compress_thres: int, head_size: int, error: float, bin_dict: Dict[str, int], abnormal_list: List[str]) → Dict

calculate summary

参数:

• data_frame – pandas.DataFrame, input data

• compress_thres – int,

• head_size – int, buffer size, when

• error – float, error tolerance

• bin_dict – a dict store col name to index

• abnormal_list – list of anomaly features

secretflow.preprocessing.binning.kernels.quantile_summaries

Classes:

Stats(value, w, delta)	store information for each item in the summary
QuantileSummaries([compress_thres, ...])	QuantileSummary

class secretflow.preprocessing.binning.kernels.quantile_summaries.Stats(value: float, w: int, delta: int)

基类：object

store information for each item in the summary

value

value of this stat

Type:

float

w

weight of this stat

Type:

int

delta

delta = rmax - rmin

Type:

int

Attributes:

value
w
delta

Methods:

__init__(value, w, delta)

value: float

w: int

delta: int

__init__(value: float, w: int, delta: int) → None

class secretflow.preprocessing.binning.kernels.quantile_summaries.QuantileSummaries(compress_thres: int = 10000, head_size: int = 10000, error: float = 0.0001, abnormal_list: Optional[List] = None)

基类：object

QuantileSummary

insert: insert data to summary merge: merge summaries fast_init: A fast version implementation creates the summary with little performance loss compress: compress summary to some size

compress_thres

if num of stats greater than compress_thres, do compress

head_size

buffer size for insert data, when samples come to head_size do create summary

error

0 <= error < 1 default: 0.001, error tolerance for binning. floor((p - 2 * error) * N) <= rank(x) <= ceil((p + 2 * error) * N)

abnormal_list

List of abnormal feature, will not participate in binning

Methods:

__init__([compress_thres, head_size, error, ...])
fast_init(col_data)
compress()	compress the summary, summary.sample will under compress_thres
query(quantile)	Use to query the value that specifies the quantile location
value_to_rank(value)
batch_query_value(values)	batch query function

__init__(compress_thres: int = 10000, head_size: int = 10000, error: float = 0.0001, abnormal_list: Optional[List] = None)

fast_init(col_data: ndarray)

compress()

compress the summary, summary.sample will under compress_thres

query(quantile: float) → float

Use to query the value that specifies the quantile location

参数:

quantile – float [0.0, 1.0]

返回:

float, the value of the quantile location

value_to_rank(value: Union[float, int]) → int

batch_query_value(values: List[float]) → List[int]

batch query function

参数:

values – List sorted_list of value. eg:[13, 56, 79]

返回:

output ranks of each query

返回类型:

List