Isssue at implementation

[Marcel Caraciolo]

I have a simple doubt:

We have lots of interfaces to fetch one user such as:

preferences_from_user (user_id)

item_ids_from_user (item_id)

But it would be faster in case of using numpy and scipy arrays to fetch multiple users data. So instead of  one user_id as parameter I could pass more than one:

preferences_from_user(np.array([user_id,user_id2,...])

item_ids_from_user(np.array([user_id,user_id2,...])

But there is a problem there's lot ofs code already implemented using the first case.  So what's the best strategy for this now ?

I was thinking about putting new interfaces, only putting the name in the plurals and the developer can implement their own. And what's the place ? In the BaseRecommender or only in the MatrixDataModel ?

What do you think ?

Regards,

--
Marcel Pinheiro Caraciolo
M.S.C. Candidate at CIN/UFPE

   http://www.mobideia.com
   http://aimotion.blogspot.com/

[Marcel Caraciolo]

THe methods that would be added:

    def preferences_from_users(self, user_ids, order_by_id=True):

    def preferences_from_users(self, user_ids, order_by_id=True):

    def items_from_users(self, user_ids):

    def preferences_for_items(self, item_ids, order_by_id=True):

    def preference_value(self, user_id, item_ids):

    def set_preferences(self, user_ids, item_ids, values):

    def remove_preferences(self, user_ids, item_ids):

[Marcel Caraciolo]

I decided to add those methods and in a future release we will work on merging those methods, OK ?

It will be easier.

Cheers,

Marcel

[Bruno J. M. Melo]

While we have no stable release this time would be excellent to change the API without backward compatibility.

Any objections?

Regards,

On Wed, Aug 3, 2011 at 10:32 AM, Marcel Caraciolo <cara...@gmail.com> wrote:

--
Bruno J. M. Melo

[Marcel Caraciolo]

Hi Bruno,

I faced a problem in this task, because my goal was to create faster queries at the data model. Imagine a way to do it in batch instead of interating at each item of the array of user_ids.

The problem is the structure returned  at them. Using numpy you would have an array with rows, where each row correspond to a array of tuples. It would be a little complex to handle . But i think it would work.

Marcel