It is not obvious why you should put the if __name__ == "__main__": line in your script. Indeed in a lot of cases, putting it or not won't change anything to how your code runs. But in specific settings with multiple scripts importing from each pother, not putting it in can quickly lead to a nightmare. To give you an insight of how and why it is useful, here is an example (if you don't want to read or if you want complementary explanations, here is a nice youtube video about it).

Suppose you have a script to fit a Ridge model on provided data, judiciously named fit_Ridge.py, which looks like this :

#!/usr/bin/env python
import argparse
import pickle  # pickle is a librairie to save and load python objects.
import numpy as np
from sklearn.linear_model import Ridge

def fit_Ridge_model(X, Y):
    model = Ridge()
    model.fit(X, Y)
    return model

parser = argparse.ArgumentParser()
parser.add_argument("--X_data_path", type=str)
parser.add_argument("--Y_data_path", type=str)
parser.add_argument("--output_path", type=str)
args = parser.parse_args()

X = np.load(args.X_data_path)
Y = np.load(args.Y_data_path)
model = fit_Ridge_model(X, Y)
pickle.dump(model, open(args.output_path, 'wb'))

This script allows the user to provide the paths to two numpy files as data to fit a Ridge model, and to save the model to the provided path with a command like :

python fit_Ridge.py --X_data_path data_folder/X.npy --Y_data_path data_folder/Y.npy --output_path models/Ridge.pk

There is no if __name__ == "__main__": to be seen but, used on its own, the script works fine.

But later, you write an other script compare_to_Lasso.py that compare Ridge and Lasso models on the same data, so you need to fit a Ridge model again. Eager to apply the good practices of programming, you judiciously decide not to duplicate the code for fitting a ridge model, but to import the fit_Ridge_model function from the fit_Ridge.py. Thus your second script looks like that :

#!/usr/bin/env python
import numpy as np
import argparse
from sklearn.linear_model import Lasso
from fit_Ridge import fit_Ridge_model

parser = argparse.ArgumentParser()
parser.add_argument("--X_data_path", type=str)
parser.add_argument("--Y_data_path", type=str)
args = parser.parse_args()

X = np.load(args.X_data_path)
Y = np.load(args.Y_data_path)

ridge_model = fit_Ridge_model(X, Y)
lasso_model = Lasso()
lasso_model.fit(X, Y)

ridge_score = ridge_model.score(X, Y)
lasso_score = lasso_model.score(X, Y)

if Ridge_score > lasso_score:
    print("Ridge model is better.")
else:
    print("Lasso model is better.")

It seems fine but here when you try to call

python compare_to_Lasso.py --X_data_path data_folder/x.npy --Y_data_path data_folder/Y.npy

you get an error :

Traceback (most recent call last):
  File "compare_lasso_ridge.py", line 5, in <module>
    from fit_Ridge import fit_Ridge_model
  File "/Users/francois/scratch/fit_Ridge.py", line 21, in <module>
    pickle.dump(model, open(args.output_path, 'wb'))
TypeError: expected str, bytes or os.PathLike object, not NoneType

The error shows that the script tried to save a model to the path args.output_path, which was not defined so it was set to None and raised a TypeError. But our compare_to_Lasso.py script never tries to save a model ! Indeed looking at the other lines of the error message, we see that it comes from the import. In fact what happens is that when we try to import the fit_Ridge_model function from the fit_Ridge.py file, python will read the entire file and execute everything that is written in it, so it will try to fit a Ridge model and to save it. But we don't want python to execute everything, we just want it to read the definition of the fit_Ridge_model function. That is why here we absolutely need the if __name__ == "__main__":, so we modify the fit_Ridge.py script like that :

#!/usr/bin/env python
import argparse
import pickle  # pickle is a librairie to save and load python objects.
import numpy as np
from sklearn.linear_model import Ridge

def fit_Ridge_model(X, Y):
    model = Ridge()
    model.fit(X, Y)
    return model

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--X_data_path", type=str)
    parser.add_argument("--Y_data_path", type=str)
    parser.add_argument("--output_path", type=str)
    args = parser.parse_args()

    X = np.load(args.X_data_path)
    Y = np.load(args.Y_data_path)
    model = fit_Ridge_model(X, Y)
    pickle.dump(model, open(args.output_path, 'wb'))

Now when importing from this script, python will read the definition of the function, but after that it will not execute the rest, since during the import the variable __name__ is not set to "__main__" but to "fit_Ridge".

In the end using if __name__ == "__main__": is the only way to safely import functions from our script, and since you never know for sure that you won't have to import something from a script in the future, putting it in all of your script by default is not a bad idea.