解决Snowflake中保存的编码器无法用于推理新数据的问题

from snowflake.ml.modeling.preprocessing import OneHotEncoder,OrdinalEncoder
import numpy as np

# OneHotEncoder示例
ohe = OneHotEncoder(handle_unknown='ignore', input_cols='CATEGORY', output_cols='ROUTE_OHE')
ohe.fit(mock_df)

# OrdinalEncoder示例
categories = {
    "AIRPORT": np.array(['A', 'B', 'C'])
}
oe = OrdinalEncoder(
    handle_unknown='use_encoded_value', unknown_value=-1,
    encoded_missing_value=-1, input_cols='AIRPORT',
    output_cols='AIRPORT_ENCODE',
    categories=categories
)
oe.fit(mock_ohe)

# 将Snowpark编码器转换为scikit-learn编码器
ohe_obj = ohe.to_sklearn()
oe_obj = oe.to_sklearn()

# 在UDF中使用scikit-learn API进行转换
df_ohe = ohe_obj.transform(df[['ROUTE_CATEGORY_NAME']])

import cachetools
import pandas as pd
from snowflake.snowpark.types import PandasDataFrameType,PandasSeriesType,PandasDataFrame,PandasSeries
import snowflake.snowpark.functions as F

@cachetools.cached(cache={})
def read_file(filename):
    import sys
    import os
    import joblib
    # Get the "path" of where files added through iport are avalible
    import_dir = sys._xoptions.get("snowflake_import_directory")
    if import_dir:
        with open(os.path.join(import_dir, filename), 'rb') as file:
            m = joblib.load(file)
            return m

@F.udf(
        name='predict_package_mix_p',session=session,replace=True,
        is_permanent=True,stage_location='@AM_TEST_UDFS',
        input_type=PandasDataFrameType([IntegerType(),StringType(),StringType(),StringType(),StringType(),StringType(),IntegerType()], list(test_df.schema.names)),
        return_type=PandasSeriesType(FloatType())
)
def predict_package_mix_p(
    df:PandasDataFrame
) -> PandasSeries:
    import pandas as pd
    from joblib import load
    import sklearn
    import xgboost as xgb
    import json
    import snowflake.ml.modeling

    def transform_simple_target_encode_manual(
            df,transform_col,transform_df
        ):
        df = df.merge(transform_df, on=transform_col)
        return df

    def remove_space(df):
        cols = df.columns
        space_cols = [x for x in cols if ' ' in x]
        for c in space_cols:
            new_col = c.replace(" ","_")
            df = df.rename(columns={c:new_col})
        return df

    ohe = read_file('one_hot_encode.pkl')
    oe = read_file('ordinal_encode.pkl')
    te = pd.read_csv(import_dir + 'target_encoding.csv.gz')
    model = read_file('xgb_model.pkl.gz')
    print('loaded models')

    features = [
        "LS1_FLIGHT_ID","DEPARTURE_AIRPORT_CODE","ARRIVAL_AIRPORT_CODE",
        "ROUTE_CATEGORY_NAME","DEPARTURE_DATETIME_LOCAL",
        "ARRIVAL_DATETIME_LOCAL","CAPACITY"
    ]

    df.columns = features
    print('loaded dataframe')

    # transform data for one hot and ordinal encodings
    df_ohe = ohe.transform(df[['ROUTE_CATEGORY_NAME']])
    encoded_df = pd.DataFrame(df_ohe, columns=ohe.categories_)
    encoded_df.columns = encoded_df.columns.get_level_values(0)
    encoded_df = encoded_df.add_prefix('ROUTE_NAME_OHE_')
    df = pd.concat([df, encoded_df], axis=1)
    df['DEPART_CODE_ENCODE'] = oe.transform(df[['DEPARTURE_AIRPORT_CODE']])

    print('transformed via one hot and ordinal')
    # transform using pre-set target encoding
    df_te = transform_simple_target_encode_manual(df,'ARRIVAL_AIRPORT_CODE',te)
    df_final = remove_space(df_te)
    print('transformed via target encode')

    # change date cols to datetime
    df_final.loc[:,'DEPARTURE_DATETIME_LOCAL'] = pd.to_datetime(
        df_final.loc[:,'DEPARTURE_DATETIME_LOCAL'],format='%Y-%m-%d %H:%M:%S',yearfirst=True
    )
    df_final['ARRIVAL_DATETIME_LOCAL'] = pd.to_datetime(
        df_final['ARRIVAL_DATETIME_LOCAL'],format='%Y-%m-%d %H:%M:%S',yearfirst=True
    )
    print('transformed dates')

    df_final['DEPART_HOUR'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.hour
    # snowpark function goes from 1-7 whereas pandas goes from 0-6
    df_final['DEPART_WEEKDAY'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.day_of_week + 1
    df_final['DEPART_MONTHDAY'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.day
    df_final['DEPART_YEARDAY'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.day_of_year
    df_final['DEPART_MONTH'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.month
    df_final['DEPART_YEAR'] = df_final['DEPARTURE_DATETIME_LOCAL'].dt.year
    df_final['ARRIVE_HOUR'] = df_final['ARRIVAL_DATETIME_LOCAL'].dt.hour
    print('created features')

    pm = pd.Series(model.predict(df_final[
            ["DEPART_CODE_ENCODE","ROUTE_NAME_OHE_CITY","ROUTE_NAME_OHE_FAR_SUN",
            "ROUTE_NAME_OHE_SKI","ROUTE_NAME_OHE_SUN","CAPACITY",
            "ARRIVAL_AIRPORT_CODE_ENCODED","DEPART_HOUR",
            "DEPART_WEEKDAY","DEPART_MONTHDAY","DEPART_YEARDAY",
            "DEPART_MONTH","DEPART_YEAR","ARRIVE_HOUR"]
        ]))
    return pm