xiaowei1234

featureU = FeatureUnion([
    ('transformations', DataFrameMapper([
        (trend_vars, ExpressionTransformer("X[:, 0] - X[:, 1]"))
        , (to_logs, make_pipeline(Imputer(strategy='median')
           , FunctionTransformer(np.log1p)
                                 ))
        ])
    )
 , ('identity', DataFrameMapper([(non_trans_vars, ContinuousDomain())]))
])

xiaowei1234

pl = PMMLPipeline([
    ('featureUnion', featureU)
    , ('impute', Imputer(strategy='median'))
    , ('standardize', StandardScaler())
#     , ('interactions', PolynomialFeatures(include_bias=False))
#     , ('clf', SGDClassifier(alpha=0.008, l1_ratio=0.13, max_iter=450,loss='log'
#                             ,penalty='elasticnet', n_iter=None, tol=None))# alpha = 0.8
    , ('clf', LogisticRegression(penalty='l2', max_iter=500, C=0.8))
])

xiaowei1234

pl = Pipeline([
    ('impute', Imputer(strategy='median'))
    , ('standardize', StandardScaler())
#     , ('interactions', PolynomialFeatures(include_bias=False))
    , ('clf', SGDClassifier())
])


alpha = [0.0001, 0.001, 0.1]

xiaowei1234

def cell_wrapper(df, func, field, drop=True, new_name=None):
    """
    decorator function for pandas pipe api
    takes func which applies function to one value in field
    returns modified dataframe
    df (pandas dataframe): the dataframe to apply transformation on
    func (function): function to apply to each value of field
    field (str): name of column in df
    drop (boolean): whether to drop 'field' after transformation
    new_name (str): whether to rename transformed 'field' column to new_name

xiaowei1234

import numpy as np
import pandas as pd
from scipy.stats import ks_2samp
from sklearn.metrics import make_scorer, roc_auc_score, log_loss
from sklearn.model_selection import GridSearchCV

def ks_stat(y, yhat):
    return ks_2samp(yhat[y==1], yhat[y!=1]).statistic

xiaowei1234

from sklearn2pmml.decoration import ContinuousDomain
from sklearn.impute import SimpleImputer


cont_d = ContinuousDomain(missing_value_replacement=350, missing_value_treatment='as_value'
                         , missing_values=[float("NaN"), -1], outlier_treatment='as_extreme_values'
                          , low_value=300, high_value=1500)

wrap = DataFrameMapper([
    ('amount', [cont_d, SimpleImputer(), FunctionTransformer(np.log1p, validate=False)])

xiaowei1234

from sklearn2pmml.preprocessing import CutTransformer
from sklearn.impute import SimpleImputer

bins = CutTransformer(bins=[0, 250, 2200], labels=[0.3, 0.4])

wrap = DataFrameMapper([
  ('amount', [SimpleImputer(), bins])
])

xiaowei1234

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder, LabelBinarizer
from sklearn_pandas import DataFrameMapper

col_trans2 = ColumnTransformer([('first', 'drop', [0])]
                               , remainder='passthrough', sparse_threshold=0.0)

mapper = DataFrameMapper([
  ('code1', [CategoricalDomain(missing_value_treatment = "as_value", missing_value_replacement = '!')
             ,LookupTransformer({'a': 'b', 'b': 'd', 'c': 'd'}, 'a')

xiaowei1234

from sklearn.pipeline import make_pipeline

wrap2 = DataFrameMapper([
    ('Status 1', LookupTransformer({203:1}, 0))
    , ('Status 2', LookupTransformer({203:1}, 0))
    , ('Status 3', LookupTransformer({203:1}, 0))
])

union = ExpressionTransformer("X[0]+X[1]+X[2]")

xiaowei1234

from sklearn.pipeline import make_pipeline
from sklearn2pmml.decoration import Alias
from sklearn.pipeline import FeatureUnion


wrap1 = DataFrameMapper([
    ('Status 1', Alias(LookupTransformer({202:1}, 0), 'status_1_202', prefit=True))
    , ('Status 2',  Alias(LookupTransformer({202:1}, 0), 'status_2_202', prefit=True))
    , ('Status 3',  Alias(LookupTransformer({202:1}, 0), 'status_3_202', prefit=True))
])