Loading saved model using Pickle - getting error as fit_transform is done in loaded program

Question

I have created the first program to train the algorithm and save it.

Program 1

import numpy as np
from sklearn.model_selection import StratifiedShuffleSplit 
from sklearn.impute import SimpleImputer  
from sklearn.tree import DecisionTreeRegressor # import for Decision Tree Algorithm
import pickle
from sklearn.preprocessing import StandardScaler

SourceData=pd.read_excel("ASML Stock Predict.xlsx") # Load the data into Pandas DataFrame
SourceData["Nasdaq Category"]=pd.cut(SourceData["Adj Close Nasdaq 100"],
                                     bins=[0., 4500, 5500, 6500, 7500,8500, 9500, 10500, np.inf],
                                     labels=[1, 2, 3, 4,5,6,7,8])

""" Split the data source into test and train subset """
split = StratifiedShuffleSplit(n_splits=1, test_size=0.01, random_state=42)
for train_index, test_index in split.split(SourceData, SourceData["Nasdaq Category"]):
    strat_train_set = SourceData.loc[train_index]  # stratfied train dataset with all columns in original source data 
    strat_test_set = SourceData.loc[test_index] #stratified test dataset with all columns in original source data

""" Drop the new Nasdaq Category Cloumn from the data source after the train and test subset is prepared"""
for set_ in (strat_train_set, strat_test_set): 
    set_.drop("Nasdaq Category", axis=1, inplace=True)

DataSource_train_independent= strat_train_set.drop(["Date", "Adj Close ASML"], axis=1) # Drop depedent variable from training dataset
DataSource_train_dependent=strat_train_set["Adj Close ASML"].copy() #  New dataframe with only independent variable value for training dataset



imputer = SimpleImputer(strategy="median") # declated imputer to fill the blank values with Median value of the variable
imputer.fit(DataSource_train_independent) # calulate the median for different independent variables

""" Scale the independent variables training set. No need to scale the dependent variable """
sc_X = StandardScaler()
X=sc_X.fit_transform(DataSource_train_independent.values) # scale the independent variables
X_test=sc_X.transform(testdata.values) # scale the independent variables for test data
##sc_y = StandardScaler()
y=DataSource_train_dependent # scaling is not required for dependent variable


"""Decision Tree Regressor """

tree_reg = DecisionTreeRegressor()
tree_reg.fit(X,y)

filename = 'DecisionTree_TrainedModel.sav'
pickle.dump(tree_reg, open(filename, 'wb'))

Program 2

from sklearn.tree import DecisionTreeRegressor # import for Decision Tree Algorithm
from sklearn.preprocessing import StandardScaler
from sklearn.tree import DecisionTreeRegressor # import for Decision Tree Algorithm
import pandas as pd

testdata=pd.read_excel("ASML Test  Stock Predict.xlsx") # Load the test data

sc_X = StandardScaler()
X_test=sc_X.transform(testdata.values) # scale the independent variables for test data



loaded_model = pickle.load(open('DecisionTree_TrainedModel.sav', 'rb'))
decision_predictions = loaded_model.predict(X_test) # Predict the value of dependent variable
print("The prediction by Decision Treemodel is " , decision_predictions )

As I have "fit_transform" in program 1 and saved the model, hence in the second program after loading the model I have only transformed the independent variables.

I am getting the error message when running the second program "sklearn.exceptions.NotFittedError: This StandardScaler instance is not fitted yet. Call 'fit' with appropriate arguments before using this estimator."

Please suggest. As I understand that I only need to transform and not fit test independent variables.

Answer 1

You have to pickle trained StandardScaler also:

# train and pickle
sc = StandardScaler()
X = sc.fit_transform(DataSource_train_independent.values)

tree_reg = DecisionTreeRegressor()
tree_reg.fit(X, y)

pickle.dump(sc, open('StandardScaler.pk', 'wb'))
pickle.dump(tree_reg, open('DecisionTree.pk', 'wb'))

# load and predict
sc = pickle.load(open('StandardScaler.pk', 'rb'))
model = pickle.load(open('DecisionTree.pk', 'rb'))

X_test = sc.transform(testdata.values)
predictions = model.predict(X_test)

The better approach is to wrap all the steps in the single pipeline :

pipeline = Pipeline(steps=[('sc', StandardScaler()), 
                           ('tree_reg', DecisionTreeRegressor())])

pipeline.fit(X, y)
pipeline.predict(testdata.values)