Create an API from EDA or ML outcome?

Question

I have the following sample dataset (the actual dataset is over 10 million records)

     Passenger    Trip
0       Mark     London
1       Mike     Girona
2    Michael      Paris
3        Max     Sydney
4     Martin  Amsterdam
5     Martin  Barcelona
6     Martin  Barcelona
7       Mark     London
8       Mark      Paris
9     Martin   New york
10       Max     Sydney
11       Max      Paris
12       Max     Sydney
...
...
...

And I wanted to get the destination frequently travelled by a passenger !

I was playing around in Jupyter and got the expected data with the following approach

series_px = df_px_dest.groupby('Passenger')['Trip'].apply(lambda x: x.value_counts().head(1))

df_px = series_px.to_frame() 

df_px.index = df_px.index.set_names(['UName', 'DEST'])

df_px.reset_index(inplace=True) 

def getNextPossibleDestByUser(pxname,df=df_px): 
    return df.query('UName==@pxname')['DEST'].to_string(index=False)

While the response is fine. I have few doubts now

1) What's the best way to expose the method (say in this case getNextPossibleDestByUser) as a API (pass customer name as input and get the destination as output) ?

2) Whenever the API is being called , does that mean all the 10 million records gets processed each time ? Are there anyway to optimise that ?

3) Rather than dataframe (pandas) query approach can I consider some ml models or utility functions from say scikit to solve the same problem ?

Answer 1

What's the best way to expose the method (say in this case getNextPossibleDestByUser) as a API (pass customer name as input and get the destination as output) ?

Use flask. Pretty easy to build an API.

Whenever the API is being called , does that mean all the 10 million records gets processed each time ? Are there anyway to optimise that ?

It depends on the exact application but for you case you can take the passenger as an input from the call to your API as POST body or in the arguments and only iterate over the subset of data for that passenger.

Rather than dataframe (pandas) query approach can I consider some ml models or utility functions from say scikit to solve the same problem ?

If the problem is -- as you illustrated -- simple frequency calculations then better to stick to pandas, in my opinion. If your problem is solved by the simple approach it would not make sense to go for a complicated approach.

Answer 2

Try Djangorestframework example : https://www.django-rest-framework.org/#example

In django views you have to write all code and can define parameters there.