Adding inference to Linear Regression model

[kavyasrinet]

Adding inference to the fluid fit_a_line model, this will help us when we re-write the models for the book.
It will also help @kexinzhao and me understand how inference works, before we start working on inference wrapper for Mobile.

We have a few questions:

1. Is the way we perform inference in this example correct ?
2. How do we determine the program that gets run inside of exe.run during inference ? (Should we clone the fluid.default_main_program() before we start training or is the method I used here, correct ? )

[dzhwinter]

You must clone the main_program first.

[QiJune]

Please refer to https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/fluid/tests/book/test_recognize_digits_mlp.py#L35.
We need to create a inference program first.

[dzhwinter]

How do we determine the program that gets run inside of exe.run during inference? (Should we clone the fluid.default_main_program() before we start training or is the method I used here, correct ? )

inference_program = fluid.default_main_program().clone()

After discussed with @reyoung @QiJune , we believe that we should provide a general interface of train/test. Which means for most users, the framework will do the clone inside.

[kavyasrinet]

Thanks for pointing that out @dzhwinter and @QiJune .
I have added an inference_program for the inference part as of now, and that is just a clone of the main_program since in this example, I am not fetching the accuracy or other attributes. Is my understanding correct here ?