Pass prediction method as function argument

Question

I'd like to specify which prediction method to use via function argument. Something like:

from sklearn.linear_model import LinearRegression

def Process(data_y_train, data_x_train, data_x_test, 
            model=LinearRegression, predict_method=predict):
  model_fit = model().fit(data_x_train, data_y_train)
  predicted_values = model_fit.predict_method(data_x_test)
  return predicted_values

Passing the model function via arugment model (e.g., LinearRegression, LogisticRegression) works well, but I'm having trouble passing the predict method (e.g., predict, predict_proba) via argument predict_method.

When I specify predict_method=predict, I get an error of 'name 'predict' is not defined'; if I specify predict_method=LinearRegression.predict, I get an error saying ''LinearRegression' object has no attribute 'predict_function''.

Per this discussion, I also tried

import sklearn.linear_model.LinearRegression

def Process(data_y_train, data_x_train, data_x_test, 
            model_module='sklearn.linear_model.LinearRegression',
            model=LinearRegression, predict_method='predict'):
  model_fit = model().fit(data_x_train, data_y_train)
  predict_call = getattr(__import__(model_module), predict_method)
  predicted_values = model_fit.predict_call(data_x_test)
  return predicted_values

But here I get an error: No module named LinearRegression.

Thank you for your help!

Answer 1

I notice that in your code, you're not using the predict_method parameter that you passed in anywhere in your code, so I don't think what you have written is what you were trying to do.

Currently, in your code, you are storing the output of the function model().fit(data_x_train, data_y_train) in the variable model_fit and then calling the predict_method attribute of that variable. If the above still doesn't work, that must be where the error is coming from, then.

I suspect what you want to do is the following:

def Process(data_y_train, data_x_train, data_x_test,
            model=LinearRegression, predict_method=LinearRegression.predict):
    model_instance = model() # create an instance of the class stored in the variable 'model'
    model_instance.fit(data_x_train, data_y_train) # run the function 'fit' belonging to that instance
    predicted_values = predict_method(model_instance,data_x_test) # run the method stored in the variable 'predict_method' - you have to pass the instance the method belongs to in the first parameter
    return predicted_values

Some more information:

• LinearRegression is a class. It defines a bunch of methods, etc.
• To create an instance of that class, you must do something like inst = LinearRegression(). The variable inst is now an instance of the class LinearRegression
• LinearRegression.predict is an example of an instance method. This means it needs an instance to run (or can be thought of as to 'operate on' in this case)
• I can therefore call inst.predict(x,y,z) but not LinearRegression.predict(x,y,z) directly.
• If you want to call LinearRegression.predict, you have to pass in the instance in the first argument: LinearRegression.predict(inst,x,y,z)

Regarding what you tried afterwards: calling a function from a string holding the function's name is not necessary in this situation and only increases the overhead, so it's probably not the correct way to go here :)

Hope this helps.