In Perl there are two functions that users interacts in shell

perl train-truecaser.perl --model outputmodelfile --corpus inputfile1





perl truecaser.perl --model outputmodelfile < inputfile2 > outputfile2

from:

• https://github.com/moses-smt/mosesdecoder/blob/master/scripts/recaser/train-truecaser.perl


• https://github.com/moses-smt/mosesdecoder/blob/master/scripts/recaser/truecase.perl

In Python, if the two functions are going to be an object, it's a little hard to conceive how the user should interact with the function with and without shell interaction.

Assuming that there are two methods in the Truecaser() object in Python

• 
  train() that does what train-truecaser.perl do


• 
  truecase() that does what truecaser.perl do

Proposal 1

We could simply follow the perl style functions:

# Initialize an object.

tc = Truecaser()



# Train the model.

tc.train(inputfile1, save_to=outputmodelfile)



# Use the Truecase model from outputmodelfile

outputfile2  = tc.truecase(inputfile2, model=outputmodelfile)

In that case the train(...) returns nothing.
And truecase outputs the outputfile2

Proposal 2

But what if we make it less function and connect the models output without output-ing it to a file, e.g.

# Initialize the object.

tc = Truecaser()



# Still allow output-ing model to file 

model = tc.train(inputfile1, save_to=outputmodelfile)



# Apply the truecase 

outputfile2 = tc.truecase(inputfile2, model=model)

In that case, possible model has to be loaded as a specific object instead of a file or on top of loading a file.

Proposal 3

We can also implicitly load/override the model when train() is called, e.g.

# Initialize the object.

tc = Truecaser()



# Still allow output-ing model to file 

# But the output model is saved to tc.model variable.

tc.train(inputfile1, save_to=outputmodelfile)



# Implicitly calls the tc.model saves from the previous tc.train()

outputfile2 = tc.truecase(inputfile2)

I'm not sure which proposal for the Python interface is more Pythonic or more intuitive to users. Esp. when Perl interface focused on shell interactions while the goal of the Python interface is to encapsulate both the train() and truecase() methods in a single object.

Which is a preferred interface for the Python object + functions?

Are there examples of similar functions in other Python library?