Saving language models

[larsgrobe]

Dear all,

I am wondering if noone ever wanted to persistently save a model for later reuse. Currently, the training functions (lda, nmf, elda) do not return the gensim-model itself, so one cannot directly use the gensim-functions to save and load models.

I could imagine two ways to allow for persistent models:

1. By making the gensim-model a member of TopicModel, and implementing .save() and .load() member function.
2. By adding a file-name parameter to the training functions. The functions could then test if the given file exists - if yes, it would be loaded and training by-passed, otherwise the file would be written after training.

Best, Lars.

[larsgrobe]

I have implemented the second approach, since it maintains compability and does not interfere much with existing code:

• If no filename is given, the training functions behave as before.
• If a filename is given and the file does not exist, the trained model is saved.
• If filename is given and points to an existing file, the model is loaded and training bypassed.

Persistent models should not only provide an interface to other modules that can handle gensim-models, but allow replicability (given that each training may produce different models) and a more explorative approach to nlp.

PR: #84

Best, Lars.