Scikit-learn/ Sklearn Objects – fit() vs transform() vs fit_transform() vs predict()

Introduction

Sklearn or scikit-learn is no doubt the most useful library for machine learning in Python. The Sklearn library contains endless efficient tools for Machine Learning and Statistical modeling which includes Classification, Regression, Clustering, and Dimensionality reduction.

In this article, we will learn different types of objects that are present in Sklearn. This article mainly deals with a clear understanding of the functions present in the objects and gives a clear idea about when to use which function in our Machine Learning Pipeline.

Sklearn Objects

Primarily, there are three types of objects in scikit-learn design:

1. Estimator

2. Predictor

3. Transformer

Now, Let’s see the usage of some important methods fit(), transform(), fit_transform() and predict().

Estimators

fit()

– It is used for calculating the initial parameters on the training data and later saves them as internal objects state.

– This method calculates the parameters μ(mean) and σ(standard deviation) and saves them as internal objects.

– A black box which only does the computation and gives nothing.

Transformers

transform()

– Use the initial above calculated values and return modified training data as output.

– Using these same parameters, using this method we can transform a particular dataset.

– Used for pre-processing before modeling.

fit_transform()

– It is a conglomerate above two steps. Internally, it first calls fit() and then transform() on the same data.

– It joins the fit() and transform() method for the transformation of the dataset.

– It is used on the training data so that we can scale the training data and also learn the scaling parameters. Here, the model built will learn the mean and variance of the features of the training set. These learned parameters are then further used to scale our test data.

Predictors

fit()

– It calculates the parameters or weights on the training data (e.g. parameters returned by coef() in case of Linear Regression) and saves them as an internal object state.

predict()

– Use the above-calculated weights on the test data to make the predictions.

Difference between fit(), transform(), and fit_transform() methods in scikit-learn

Let’s try to understand the difference with a given example:

Suppose you have an array arr = [1,2,3,y,5] and you have a sklearn class FillMyArray that filled your array.

When you declare an instance of your class:

my_filler = FillMyArray()

We have the in hand methods fit(), transform() and fit_transform().

fit(): my_filler.fit(arr) will compute the value to assign to x to fill out the array and store it in our instance my_filler.

transform(): After the value is computed and stored during the previous .fit() stage we can call my_filler.transform(arr) which will return the filled array [1,2,3,4,5].

fit_transform(): If we perform my_filler.fit_transform(arr) we can skip one line of code and have the value calculated along with assigned to the filled array that is directly returned in only one stage.

Let’s answer a crucial question

Question: We are aware that we call fit_transform() on our training dataset, while the transform() method on our test dataset. But the question is why do we do so?

The real deal here is “Data leakage”.

fit_transform do certain calculations followed by transformation, Let’s assume calculating the mean or average of columns from certain data and then replacing the missing values according to it. For a training set, we need to both calculate followed by transformation.

However, for the testing set, Machine learning applies predictions based on the learning during the training set, due to which it doesn’t need to perform calculations and perform just the transformation.

If we perform the fit() method even on test data, we will compute a new mean and variance that will be a Naive scale for each feature and will allow the model to learn on the test data too. However, we will no longer be able to keep it as a surprise to our model and it wouldn’t be able to give us a good estimate on model performance on the unseen data, which is certainly our ultimate aim.

It is the general procedure to scale the data when building a machine learning model. So that the model is not biased to a specific feature and prevents our model to learn the trends of our test data at the same time.

Implementation in Python

Here we try to implement all the functions which we studied in the above part of the article.

Step-1: Import necessary python libraries and then read and load the “TITANIC” Dataset.

Step-2: Calculate the number of missing values per column.

df.isnull().sum()

Step-3: Fill the missing value of the “Age” column with their respective median.

df['Age'].fillna(df.Age.median(),inplace=True)

Step-4: Now, again check if there are missing values present or not in any column.

df.isnull().sum()

Step-5: Define our Independent(predictor) and Dependent(response) variables.

X=df.iloc[:,1:]
y=df.iloc[:,0]

Step-6: Split our dataset into train and test subsets.

from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.20,random_state=42)

Step-7: Now using standard scaler we first fit and then transform our dataset.

from sklearn.preprocessing import StandardScaler
scaler=StandardScaler()
X_train_fit=scaler.fit(X_train)
X_train_scaled=scaler.transform(X_train)
pd.DataFrame(X_train_scaled)

Step-8: Use fit_transform() function directly and verify the results.

X_train_scaled=scaler.fit_transform(X_train)
pd.DataFrame(X_train_scaled)

Step-9: Transform our test data.

X_test_scaled=scaler.transform(X_test)
pd.DataFrame(X_test_scaled)

Conclusion

– Here we observe that the fit_transform() function gives the same result as the function fit() and the transform() function gives separately by combining the results.

– Remember fit_transform() function only acts on training data, transform() acts on test data, and predict() acts on test data.

– In summary, fit() performs or completes the training step, transform() changes the data in the pipeline to pass it on to the next stage in the pipeline, and fit_transform() does both the fitting and the transforming in one possibly short step.

End Notes

Thanks for reading!

If you liked this and want to know more, go visit my other articles on Data Science and Machine Learning by clicking on the Link

Please feel free to contact me on Linkedin, Email.

Something not mentioned or want to share your thoughts? Feel free to comment below And I’ll get back to you.

Till then Stay Home, Stay Safe to prevent the spread of COVID-19, and Keep Learning!

About the Author

Chirag Goyal

Currently, I pursuing my Bachelor of Technology (B.Tech) in Computer Science and Engineering from the Indian Institute of Technology Jodhpur(IITJ). I am very enthusiastic about Machine learning, Deep Learning, and Artificial Intelligence.

CHIRAG GOYAL 26 Aug 2022

I am currently pursuing my Bachelor of Technology (B.Tech) in Computer Science and Engineering from the Indian Institute of Technology Jodhpur(IITJ). I am very enthusiastic about Machine learning, Deep Learning, and Artificial Intelligence. Feel free to connect with me on Linkedin.