def _save_min_max_meta(self, name, namespace):
if self.scale_param.area == consts.ALL:
LOGGER.debug("save_min_max_meta with mode is all")
feat_upper = "None" if self.scale_param.feat_upper is None else str(self.scale_param.feat_upper)
feat_lower = "None" if self.scale_param.feat_lower is None else str(self.scale_param.feat_lower)
out_upper = "None" if self.scale_param.out_upper is None else str(self.scale_param.out_upper)
out_lower = "None" if self.scale_param.out_lower is None else str(self.scale_param.out_lower)
min_max_scale_meta = feature_scale_meta_pb2.MinMaxScaleMeta(feat_upper=feat_upper,
feat_lower=feat_lower,
out_upper=out_upper,
out_lower=out_lower)
minmax_scale_meta = {"0": min_max_scale_meta}
meta_protobuf_obj = feature_scale_meta_pb2.ScaleMeta(is_scale=True,
strategy=self.scale_param.method,
minmax_scale_meta=minmax_scale_meta)
else:
LOGGER.debug("save_min_max_meta with mode is {}".format(self.scale_param.mode))
meta_protobuf_obj = feature_scale_meta_pb2.ScaleMeta(is_scale=True)
buffer_type = "{}.meta".format(self.class_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def save_standard_scale_model(self, name, namespace):
"""
Save StandardScaler param and meta
Parameters
----------
name: str, save DTable table_name
namespace: str, DTable namespace
Returns
----------
tuple,
include meta string name and param string name
"""
meta_buffer_type = self._save_standard_scale_meta(name, namespace)
standard_scale_param_dict = {}
for i in range(len(self.header)):
if i in self.std_scale_column_idx:
mean = self.mean[i]
std = self.std[i]
param_obj = feature_scale_param_pb2.StandardScaleParam(mean=mean, scale=std)
standard_scale_param_dict[self.header[i]] = param_obj
param_protobuf_obj = feature_scale_param_pb2.ScaleParam(standard_scale_param=standard_scale_param_dict)
param_buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=param_protobuf_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def save_model(self, name, namespace):
meta_buffer_type = self._save_meta(name, namespace)
# In case arbiter has no header
header = self.header
weight_dict = {}
for idx, header_name in enumerate(header):
coef_i = self.coef_[idx]
weight_dict[header_name] = coef_i
param_protobuf_obj = lr_model_param_pb2.LRModelParam(
iters=self.n_iter_,
loss_history=self.loss_history,
is_converged=self.is_converged,
weight=weight_dict,
intercept=self.intercept_,
header=header)
buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=param_protobuf_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, buffer_type)]
def save_model(self, name, namespace):
meta_buffer_type = self._save_meta(name, namespace)
iv_attrs = []
for idx, iv_attr in enumerate(self.iv_attrs):
LOGGER.debug("{}th iv attr: {}".format(idx, iv_attr.__dict__))
iv_result = iv_attr.result_dict()
iv_object = feature_binning_param_pb2.IVParam(**iv_result)
iv_attrs.append(iv_object)
host_iv_attrs = []
if self.host_iv_attrs is not None:
for idx, iv_attr in enumerate(self.host_iv_attrs):
iv_result = iv_attr.result_dict()
iv_object = feature_binning_param_pb2.IVParam(**iv_result)
host_iv_attrs.append(iv_object)
result_obj = feature_binning_param_pb2.FeatureBinningParam(iv_result=iv_attrs,
host_iv_result=host_iv_attrs,
cols=self.cols)
param_buffer_type = "HeteroFeatureBinningGuest.param"
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=result_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def save_min_max_model(self, name, namespace):
meta_buffer_type = self._save_min_max_meta(name, namespace)
min_max_scale_param_dict = {}
if self.cols_scale_value is not None:
for i in range(len(self.header)):
feat_lower = self.cols_scale_value[i][0]
feat_upper = self.cols_scale_value[i][1]
out_lower = self.cols_scale_value[i][2]
out_upper = self.cols_scale_value[i][3]
param_obj = feature_scale_param_pb2.MinMaxScaleParam(
feat_upper=feat_upper,
feat_lower=feat_lower,
out_upper=out_upper,
out_lower=out_lower)
min_max_scale_param_dict[self.header[i]] = param_obj
param_protobuf_obj = feature_scale_param_pb2.ScaleParam(
minmax_scale_param=min_max_scale_param_dict)
param_buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=param_protobuf_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def _save_pipeline(self):
buffer_type = "Pipeline"
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=self.pipeline,
name=self.workflow_param.model_table,
namespace=self.workflow_param.model_namespace)
def save_outlier_model(outlier_replace=False,
outlier_replace_method=None,
outlier_impute=None,
outlier_replace_value=None,
header=None,
model_name="Outlier",
model_table=None,
model_namespace=None):
model_meta = OutlierMeta()
model_param = OutlierParam()
model_meta.is_outlier = outlier_replace
if outlier_replace:
if outlier_replace_method:
model_meta.strategy = str(outlier_replace_method)
if outlier_impute:
model_meta.outlier_value.extend(map(str, outlier_impute))
if outlier_replace_value:
outlier_value_dict = dict(
zip(header, map(str, outlier_replace_value)))
model_param.outlier_replace_value.update(outlier_value_dict)
manager.save_model(buffer_type=model_name + ".meta",
proto_buffer=model_meta,
name=model_table,
namespace=model_namespace)
manager.save_model(buffer_type=model_name + ".param",
proto_buffer=model_param,
name=model_table,
namespace=model_namespace)
def save_missing_imputer_model(missing_fill=False,
missing_replace_method=None,
missing_impute=None,
missing_fill_value=None,
header=None,
model_name="Imputer",
model_table=None,
model_namespace=None):
model_meta = ImputerMeta()
model_param = ImputerParam()
model_meta.is_imputer = missing_fill
if missing_fill:
if missing_replace_method:
model_meta.strategy = str(missing_replace_method)
if missing_impute is not None:
if missing_impute is not None:
model_meta.missing_value.extend(map(str, missing_impute))
if missing_fill_value is not None:
feature_value_dict = dict(zip(header, map(str,
missing_fill_value)))
model_param.missing_replace_value.update(feature_value_dict)
manager.save_model(buffer_type=model_name + ".meta",
proto_buffer=model_meta,
name=model_table,
namespace=model_namespace)
manager.save_model(buffer_type=model_name + ".param",
proto_buffer=model_param,
name=model_table,
namespace=model_namespace)
def save_data_io_model(input_format="dense",
delimitor=",",
data_type="str",
with_label=False,
label_idx=0,
label_type="int",
output_format="dense",
header=None,
model_name="DataIO",
model_table=None,
model_namespace=None):
model_meta = DataIOMeta()
model_param = DataIOParam()
model_meta.input_format = input_format
model_meta.delimitor = delimitor
model_meta.data_type = data_type
model_meta.with_label = with_label
model_meta.label_idx = label_idx
model_meta.label_type = label_type
model_meta.output_format = output_format
if header is not None:
model_param.header.extend(header)
manager.save_model(buffer_type=model_name + ".meta",
proto_buffer=model_meta,
name=model_table,
namespace=model_namespace)
manager.save_model(buffer_type=model_name + ".param",
proto_buffer=model_param,
name=model_table,
namespace=model_namespace)
def save_model(self, name, namespace):
"""
Save each classifier model of OneVsRest. It just include model_param but not model_meta now
"""
classifier_models = []
str_time = time.strftime("%Y%m%d%H%M%S", time.localtime())
for i, model in enumerate(self.models):
classifier_name = str_time + "_" + str(
i) + "_" + self.role + "_name"
model.save_model(classifier_name, namespace)
classifier_model = one_vs_rest_param_pb2.ClassifierModel(
name=classifier_name, namespace=namespace)
classifier_models.append(classifier_model)
LOGGER.info("finish save model_{}, role:{}".format(i, self.role))
str_classes = [str(c) for c in self.classes]
one_vs_rest_param_obj = one_vs_rest_param_pb2.OneVsRestParam(
classes=str_classes, classifier_models=classifier_models)
param_buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=one_vs_rest_param_obj,
name=name,
namespace=namespace)
meta_buffer_type = 'None'
LOGGER.info("finish OneVsRest save model.")
return [(meta_buffer_type, param_buffer_type)]
def _save_meta(self, name, namespace):
meta_protobuf_obj = onehot_meta_pb2.OneHotMeta(cols=self.cols)
buffer_type = "OneHotEncoder.meta"
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def save_model(self, name, namespace):
meta_buffer_type = self._save_meta(name, namespace)
result_obj = feature_selection_param_pb2.FeatureSelectionParam(
results=self.results)
param_buffer_type = "HeteroFeatureSelectionGuest.param"
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=result_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def save_model(self, model_table, model_namespace):
LOGGER.info("save model")
meta_name, meta_protobuf = self.get_model_meta()
param_name, param_protobuf = self.get_model_param()
manager.save_model(buffer_type=meta_name,
proto_buffer=meta_protobuf,
name=model_table,
namespace=model_namespace)
manager.save_model(buffer_type=param_name,
proto_buffer=param_protobuf,
name=model_table,
namespace=model_namespace)
return [(meta_name, param_name)]
def _save_standard_scale_meta(self, name, namespace):
with_mean = self.scale_param.with_mean
with_std = self.scale_param.with_std
standard_scale_meta = feature_scale_meta_pb2.StandardScaleMeta(with_mean=with_mean, with_std=with_std)
meta_protobuf_obj = feature_scale_meta_pb2.ScaleMeta(is_scale=True,
strategy=self.scale_param.method,
standard_scale_meta=standard_scale_meta)
buffer_type = "{}.meta".format(self.class_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def save_model(self, name, namespace):
meta_buffer_type = self._save_meta(name, namespace)
left_col_obj = feature_selection_param_pb2.LeftCols(
original_cols=self.cols, left_cols=self.left_cols)
result_obj = feature_selection_param_pb2.FeatureSelectionParam(
results=self.results, final_left_cols=left_col_obj)
param_buffer_type = "HeteroFeatureSelection{}.param".format(
self.party_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=result_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def _save_meta(self, name, namespace):
meta_protobuf_obj = feature_binning_meta_pb2.FeatureBinningMeta(
method=self.bin_param.method,
compress_thres=self.bin_param.compress_thres,
head_size=self.bin_param.head_size,
error=self.bin_param.error,
bin_num=self.bin_param.bin_num,
cols=self.cols,
adjustment_factor=self.bin_param.adjustment_factor,
local_only=self.bin_param.local_only)
buffer_type = "HeteroFeatureBinning{}.meta".format(self.party_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def _save_meta(self, name, namespace):
meta_protobuf_obj = feature_selection_meta_pb2.FeatureSelectionMeta(
filter_methods=self.filter_method,
local_only=self.params.local_only,
cols=self.cols,
unique_meta=self.unique_meta,
iv_value_meta=self.iv_value_meta,
iv_percentile_meta=self.iv_percentile_meta,
coe_meta=self.coe_meta,
outlier_meta=self.outlier_meta)
buffer_type = "HeteroFeatureSelection{}.meta".format(self.party_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def _save_meta(self, name, namespace):
meta_protobuf_obj = lr_model_meta_pb2.LRModelMeta(
penalty=self.param.penalty,
eps=self.eps,
alpha=self.alpha,
optimizer=self.param.optimizer,
party_weight=self.param.party_weight,
batch_size=self.batch_size,
learning_rate=self.learning_rate,
max_iter=self.max_iter,
converge_func=self.param.converge_func,
re_encrypt_batches=self.param.re_encrypt_batches)
buffer_type = "{}.meta".format(self.class_name)
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def save_model(self, name, namespace):
meta_buffer_type = self._save_meta(name, namespace)
pb_dict = {}
for col_name, value_dict in self.col_maps.items():
value_dict_obj = onehot_param_pb2.ColDict(encode_map=value_dict)
pb_dict[col_name] = value_dict_obj
result_obj = onehot_param_pb2.OneHotParam(col_map=pb_dict)
param_buffer_type = "OneHotEncoder.param"
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=result_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def save_standard_scale_model(self, name, namespace):
meta_buffer_type = self._save_standard_scale_meta(name, namespace)
standard_scale_param_dict = {}
for i in range(len(self.header)):
mean = self.mean[i]
std = self.std[i]
param_obj = feature_scale_param_pb2.StandardScaleParam(mean=mean, scale=std)
standard_scale_param_dict[self.header[i]] = param_obj
param_protobuf_obj = feature_scale_param_pb2.ScaleParam(standard_scale_param=standard_scale_param_dict)
param_buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=param_protobuf_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def _save_meta(self, name, namespace):
unique_param_dict = copy.deepcopy(self.params.unique_param.__dict__)
unique_param = feature_selection_meta_pb2.UniqueValueParam(
**unique_param_dict)
iv_dict = copy.deepcopy(self.params.iv_param.__dict__)
bin_dict = copy.deepcopy(self.params.iv_param.bin_param.__dict__)
del bin_dict['process_method']
del bin_dict['result_table']
del bin_dict['result_namespace']
del bin_dict['display_result']
if bin_dict['cols'] == -1:
bin_dict['cols'] = self.cols
bin_param = FeatureBinningMeta()
iv_dict["bin_param"] = bin_param
iv_param = feature_selection_meta_pb2.IVSelectionParam(**iv_dict)
coe_param_dict = copy.deepcopy(self.params.coe_param.__dict__)
coe_param = feature_selection_meta_pb2.CoeffOfVarSelectionParam(
**coe_param_dict)
outlier_param_dict = copy.deepcopy(self.params.outlier_param.__dict__)
outlier_param = feature_selection_meta_pb2.OutlierColsSelectionParam(
**outlier_param_dict)
meta_protobuf_obj = feature_selection_meta_pb2.FeatureSelectionMeta(
filter_methods=self.filter_method,
local_only=self.params.local_only,
select_cols=self.header,
unique_param=unique_param,
iv_param=iv_param,
coe_param=coe_param,
outlier_param=outlier_param)
buffer_type = "HeteroFeatureSelectionGuest.meta"
model_manager.save_model(buffer_type=buffer_type,
proto_buffer=meta_protobuf_obj,
name=name,
namespace=namespace)
return buffer_type
def save_min_max_model(self, name, namespace):
"""
Save MinMaxScaler param and meta
Parameters
----------
name: str, save DTable table_name
namespace: str, DTable namespace
Returns
----------
tuple,
include meta string name and param string name
"""
meta_buffer_type = self._save_min_max_meta(name, namespace)
cols_scale_value = self.cols_scale_res[0]
scale_column_idx = self.cols_scale_res[1]
min_max_scale_param_dict = {}
if cols_scale_value is not None:
for i in range(len(self.header)):
if i in scale_column_idx:
feat_lower = cols_scale_value[i][0]
feat_upper = cols_scale_value[i][1]
out_lower = cols_scale_value[i][2]
out_upper = cols_scale_value[i][3]
param_obj = feature_scale_param_pb2.MinMaxScaleParam(feat_upper=feat_upper,
feat_lower=feat_lower,
out_upper=out_upper,
out_lower=out_lower)
min_max_scale_param_dict[self.header[i]] = param_obj
param_protobuf_obj = feature_scale_param_pb2.ScaleParam(minmax_scale_param=min_max_scale_param_dict)
param_buffer_type = "{}.param".format(self.class_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=param_protobuf_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
def save_model(self, name, namespace, binning_result=None, host_results=None):
if binning_result is None:
binning_result = self.binning_result
if host_results is None:
host_results = self.host_results
meta_buffer_type = self._save_meta(name, namespace)
iv_attrs = {}
for col_name, iv_attr in binning_result.items():
iv_result = iv_attr.result_dict()
iv_object = feature_binning_param_pb2.IVParam(**iv_result)
iv_attrs[col_name] = iv_object
binning_result_obj = feature_binning_param_pb2.FeatureBinningResult(binning_result=iv_attrs)
final_host_results = {}
for host_id, this_host_results in host_results.items():
host_result = {}
for col_name, iv_attr in this_host_results.items():
iv_result = iv_attr.result_dict()
iv_object = feature_binning_param_pb2.IVParam(**iv_result)
host_result[col_name] = iv_object
final_host_results[host_id] = feature_binning_param_pb2.FeatureBinningResult(binning_result=host_result)
result_obj = feature_binning_param_pb2.FeatureBinningParam(binning_result=binning_result_obj,
host_results=final_host_results)
param_buffer_type = "HeteroFeatureBinning{}.param".format(self.party_name)
model_manager.save_model(buffer_type=param_buffer_type,
proto_buffer=result_obj,
name=name,
namespace=namespace)
return [(meta_buffer_type, param_buffer_type)]
