Package utils

Load documents from a text file. If batch_size is given, behaves like a generator and returns one batch at a time. Each document is represented as a list of tuples, where each tuple contains a word id and a word count.

Each line of the text file is assumed to contain one document and should start with the number of unique words in that document, followed by the words. Each word should be represented by its id and a number of occurences separated by a colon. For example:

       6 5600:2 293:1 5548:1 2577:1 3733:3 2677:2

Parameters:

• batch_size (int) - the number of documents to return at once
• stochastic (bool) - if True, the batch size is drawn from a Poisson distribution

Returns: list/generator

returns either a list of documents or a generator of lists of documents

Load users from a text file. If `batch_size` is given, behaves like a generator and returns one batch at a time. Each user is represented as a list of tuples, where each tuple contains an item id and a rating.

Each line is assumed to contain a 3-tuple of a user id, an item id, and a rating. The ratings of users should be grouped. For example:

       1488844   1  3
       1488844   8  4
       1488844  17  2
       1488844  30  3
       8850131  33  4
       8850131  35  1
       8850131  86  5

Parameters:

• filepath (str) - path to file containing data
• batch_size (int) - the number of users to return at once
• stochastic (bool) - if True, the batch size is drawn from a Poisson distribution
• threshold (int) - only load users whose rating is greater or equal this threshold

Returns: list/generator

returns either a list of users or a generator of lists of users

Like load_users, but users are stored in a dictionary instead of a list. The keys correspond to user IDs and the values are lists of item/rating pairs.

Returns: dict/generator

returns either a dictionary of users or a generator of dictionaries of users

Randomly selects $k$ out of $n$ elements.

Parameters:

• k (int) - the number of elements to pick
• n (int) - the number of elements to pick from

Returns: list

a list of $k$ indices