In this tutorial series, we are going to cover Logistic Regression using Pyspark. Logistic Regression is one of the basic ways to perform classification (don’t be confused by the word “regression”). Logistic Regression is a classification method. Some examples of classification are:
Loading Dataframe
We will be using the data for Titanic where I have columns PassengerId, Survived, Pclass, Name, Sex, Age, SibSp, Parch, Ticket, Fare, Cabin, and Embarked. We have to predict whether the passenger will survive or not using the Logistic Regression machine learning model. To get started, open a new notebook and follow the steps mentioned in the below code:
Python3
# Starting the Spark Session from pyspark.sql import SparkSession spark = SparkSession.builder.appName('Titanic').getOrCreate() # Reading the data df = spark.read.csv('Titanic.csv',inferSchema=True, header=True) # Showing the data df.show() |
Output:
Showing the data
df.printSchema()
Schema of the data
df.columns
Columns in the data
Removing NULL Values Columns
The next step includes removing the data having null values as shown in the above picture. We do not need the columns PassengerId, Name, Ticket, and Cabin as they are not required to train and test the model.
Python3
# Selecting the columns which are required # to train and test the model. rm_columns = df.select(['Survived','Pclass', 'Sex','Age','SibSp', 'Parch','Fare','Embarked']) # Drops the data having null values result = rm_columns.na.drop() # Again showing the data result.show() |
Output:
Convert String Column to Ordinal Columns
The next task is to convert the string columns (Sex and Embarked) to integral columns as without doing this, we cannot vectorize the data using VectorAssembler.
Python3
# Importing the required libraries from pyspark.ml.feature import VectorAssembler, StringIndexer, OneHotEncoder # Converting the Sex Column sexIdx = StringIndexer(inputCol='Sex', outputCol='SexIndex') sexEncode = OneHotEncoder(inputCol='SexIndex', outputCol='SexVec') # Converting the Embarked Column embarkIdx = StringIndexer(inputCol='Embarked', outputCol='EmbarkIndex') embarkEncode = OneHotEncoder(inputCol='EmbarkIndex', outputCol='EmbarkVec') # Vectorizing the data into a new column "features" # which will be our input/features class assembler = VectorAssembler(inputCols=['Pclass', 'SexVec','Age', 'SibSp','Parch', 'Fare','EmbarkVec'], outputCol='features') |
Now we need Pipeline to stack the tasks one by one and import and call the Logistic Regression Model.
Python3
# Importing Pipeline and Model from pyspark.ml import Pipeline from pyspark.ml.classification import LogisticRegression log_reg = LogisticRegression(featuresCol='features', labelCol='Survived') # Creating the pipeline pipe = Pipeline(stages=[sexIdx, embarkIdx, sexEncode, embarkEncode, assembler, log_reg]) |
After pipelining the tasks, we will split the data into training data and testing data to train and test the model.
Python3
# Splitting the data into train and test train_data, test_data = my_final_data.randomSplit([0.7, .3]) # Fitting the model on training data fit_model = pipeline.fit(train_data) # Storing the results on test data results = fit_model.transform(test_data) # Showing the results results.show() |
Output:
data.show()
Model evaluation using ROC-AUC
The results will add extra columns rawPrediction, probability, and prediction because we are transforming the results on our data. After getting the results, we will now find the AUC(Area under the ROC Curve) which will give the efficiency of the model. For this, we will use BinaryClassificationEvaluator as shown:
Python3
# Importing the evaluator from pyspark.ml.evaluation import BinaryClassificationEvaluator # Calling the evaluator res = BinaryClassificationEvaluator (rawPredictionCol='prediction',labelCol='Survived') # Evaluating the AUC on results ROC_AUC = res.evaluate(results) |
Output:
Note: In general, an AUC value above 0.7 is considered good, but it’s important to compare the value to the expected performance of the problem and the data to determine if it’s actually good.
ROC_AUC
