Source code for surprise.dataset

"""
the :mod:`dataset` module defines some tools for managing datasets.

Users may use both *built-in* and user-defined datasets (see the
:ref:`getting_started` page for examples). Right now, four built-in datasets
are available:

* The `movielens-100k <http://grouplens.org/datasets/movielens/>`_ dataset.
* The `movielens-1m <http://grouplens.org/datasets/movielens/>`_ dataset.
* The `Jester <http://eigentaste.berkeley.edu/dataset/>`_ dataset 2.

Built-in datasets can all be loaded (or downloaded if you haven't already)
using the :meth:`Dataset.load_builtin` method. For each built-in dataset,
Surprise also provide predefined :class:`readers <Reader>` which are useful if
you want to use a custom dataset that has the same format as a built-in one.

Summary:

.. autosummary::
    :nosignatures:

    Dataset.load_builtin
    Dataset.load_from_file
    Dataset.load_from_folds
    Dataset.folds
    DatasetAutoFolds.split
    Reader
    Trainset
"""


from __future__ import (absolute_import, division, print_function,
                        unicode_literals)
from collections import defaultdict
from collections import namedtuple
import sys
import os
import zipfile
import itertools
import random

import numpy as np

from .six.moves import input
from .six.moves.urllib.request import urlretrieve
from .six.moves import range
from .six import iteritems


# directory where builtin datasets are stored. For now it's in the home
# directory under the .surprise_data. May be ask user to define it?
DATASETS_DIR = os.path.expanduser('~') + '/.surprise_data/'

# a builtin dataset has
# - an url (where to download it)
# - a path (where it is located on the filesystem)
# - the parameters of the corresponding reader
BuiltinDataset = namedtuple('BuiltinDataset', ['url', 'path', 'reader_params'])

BUILTIN_DATASETS = {
    'ml-100k':
        BuiltinDataset(
            url='http://files.grouplens.org/datasets/movielens/ml-100k.zip',
            path=DATASETS_DIR + 'ml-100k/ml-100k/u.data',
            reader_params=dict(line_format='user item rating timestamp',
                               rating_scale=(1, 5),
                               sep='\t')
        ),
    'ml-1m':
        BuiltinDataset(
            url='http://files.grouplens.org/datasets/movielens/ml-1m.zip',
            path=DATASETS_DIR + 'ml-1m/ml-1m/ratings.dat',
            reader_params=dict(line_format='user item rating timestamp',
                               rating_scale=(1, 5),
                               sep='::')
        ),
    'jester':
        BuiltinDataset(
            url='http://eigentaste.berkeley.edu/dataset/jester_dataset_2.zip',
            path=DATASETS_DIR + 'jester/jester_ratings.dat',
            reader_params=dict(line_format='user item rating',
                               rating_scale=(-10, 10))
        )
}


[docs]class Dataset: """Base class for loading datasets. Note that you should never instantiate the :class:`Dataset` class directly (same goes for its derived classes), but instead use one of the three available methods for loading datasets.""" def __init__(self, reader): self.reader = reader self.r_min = reader.inf + reader.offset self.r_max = reader.sup + reader.offset @classmethod
[docs] def load_builtin(cls, name='ml-100k'): """Load a built-in dataset. If the dataset has not already been loaded, it will be downloaded and saved. You will have to split your dataset using the :meth:`split <DatasetAutoFolds.split>` method. See an example in the :ref:`User Guide <load_builtin_example>`. Args: name(:obj:`string`): The name of the built-in dataset to load. Accepted values are 'ml-100k', 'ml-1m', and 'jester'. Default is 'ml-100k'. Returns: A :obj:`Dataset` object. Raises: ValueError: If the ``name`` parameter is incorrect. """ try: dataset = BUILTIN_DATASETS[name] except KeyError: raise ValueError('unknown dataset ' + name + '. Accepted values are ' + ', '.join(BUILTIN_DATASETS.keys()) + '.') # if dataset does not exist, offer to download it if not os.path.isfile(dataset.path): answered = False while not answered: print('Dataset ' + name + ' could not be found. Do you want ' 'to download it? [Y/n] ', end='') choice = input().lower() if choice in ['yes', 'y', '', 'omg this is so nice of you!!']: answered = True elif choice in ['no', 'n', 'hell no why would i want that?!']: answered = True print("Ok then, I'm out!") sys.exit() if not os.path.exists(DATASETS_DIR): os.makedirs(DATASETS_DIR) print('Trying to download dataset from ' + dataset.url + '...') urlretrieve(dataset.url, DATASETS_DIR + 'tmp.zip') with zipfile.ZipFile(DATASETS_DIR + 'tmp.zip', 'r') as tmp_zip: tmp_zip.extractall(DATASETS_DIR + name) os.remove(DATASETS_DIR + 'tmp.zip') print('Done! Dataset', name, 'has been saved to', DATASETS_DIR + name) reader = Reader(**dataset.reader_params) return cls.load_from_file(file_path=dataset.path, reader=reader)
@classmethod
[docs] def load_from_file(cls, file_path, reader): """Load a dataset from a (custom) file. Use this if you want to use a custom dataset and all of the ratings are stored in one file. You will have to split your dataset using the :meth:`split <DatasetAutoFolds.split>` method. See an example in the :ref:`User Guide <load_from_file_example>`. Args: file_path(:obj:`string`): The path to the file containing ratings. reader(:obj:`Reader`): A reader to read the file. """ return DatasetAutoFolds(ratings_file=file_path, reader=reader)
@classmethod
[docs] def load_from_folds(cls, folds_files, reader): """Load a dataset where folds (for cross-validation) are predifined by some files. The purpose of this method is to cover a common use case where a dataset is already split into predefined folds, such as the movielens-100k dataset which defines files u1.base, u1.test, u2.base, u2.test, etc... It can also be used when you don't want to perform cross-validation but still want to specify your training and testing data (which comes down to 1-fold cross-validation anyway). See an example in the :ref:`User Guide <load_from_folds_example>`. Args: folds_files(:obj:`iterable` of :obj:`tuples`): The list of the folds. A fold is a tuple of the form ``(path_to_train_file, path_to_test_file)``. reader(:obj:`Reader`): A reader to read the files. """ return DatasetUserFolds(folds_files=folds_files, reader=reader)
def read_ratings(self, file_name): """Return a list of ratings (user, item, rating, timestamp) read from file_name""" with open(os.path.expanduser(file_name)) as f: raw_ratings = [self.reader.parse_line(line) for line in itertools.islice(f, self.reader.skip_lines, None)] return raw_ratings
[docs] def folds(self): """Generator function to iterate over the folds of the Dataset. See :ref:`User Guide <iterate_over_folds>` for usage. Yields: tuple: :class:`Trainset` and testset of current fold. """ for raw_trainset, raw_testset in self.raw_folds(): trainset = self.construct_trainset(raw_trainset) testset = self.construct_testset(raw_testset) yield trainset, testset
def construct_trainset(self, raw_trainset): raw2inner_id_users = {} raw2inner_id_items = {} current_u_index = 0 current_i_index = 0 rm = defaultdict(int) ur = defaultdict(list) ir = defaultdict(list) # user raw id, item raw id, rating, time stamp for urid, irid, r, timestamp in raw_trainset: try: uid = raw2inner_id_users[urid] except KeyError: uid = current_u_index raw2inner_id_users[urid] = current_u_index current_u_index += 1 try: iid = raw2inner_id_items[irid] except KeyError: iid = current_i_index raw2inner_id_items[irid] = current_i_index current_i_index += 1 rm[uid, iid] = r ur[uid].append((iid, r)) ir[iid].append((uid, r)) n_users = len(ur) # number of users n_items = len(ir) # number of items trainset = Trainset(rm, ur, ir, n_users, n_items, self.r_min, self.r_max, raw2inner_id_users, raw2inner_id_items) return trainset def construct_testset(self, raw_testset): return [(ruid, riid, r) for (ruid, riid, r, _) in raw_testset]
class DatasetUserFolds(Dataset): """A derived class from :class:`Dataset` for which folds (for cross-validation) are predefined.""" def __init__(self, folds_files=None, reader=None): Dataset.__init__(self, reader) self.folds_files = folds_files # check that all files actually exist. for train_test_files in self.folds_files: for f in train_test_files: if not os.path.isfile(os.path.expanduser(f)): raise ValueError('File ' + str(f) + ' does not exist.') def raw_folds(self): for train_file, test_file in self.folds_files: raw_train_ratings = self.read_ratings(train_file) raw_test_ratings = self.read_ratings(test_file) yield raw_train_ratings, raw_test_ratings
[docs]class DatasetAutoFolds(Dataset): """A derived class from :class:`Dataset` for which folds (for cross-validation) are not predefined. (Or for when there are no folds at all).""" def __init__(self, ratings_file=None, reader=None): Dataset.__init__(self, reader) self.ratings_file = ratings_file self.n_folds = 5 self.shuffle = True self.raw_ratings = self.read_ratings(self.ratings_file)
[docs] def build_full_trainset(self): """Do not split the dataset into folds and just return a trainset as is, built from the whole dataset. User can then query for predictions, as shown in the :ref:`User Guide <train_on_whole_trainset>`. Returns: The :class:`Trainset`. """ return self.construct_trainset(self.raw_ratings)
def raw_folds(self): if self.shuffle: random.shuffle(self.raw_ratings) self.shuffle = False # set to false for future calls to raw_folds def k_folds(seq, n_folds): """Inspired from scikit learn KFold method.""" if n_folds > len(seq) or n_folds < 2: raise ValueError('Incorrect value for n_folds.') start, stop = 0, 0 for fold_i in range(n_folds): start = stop stop += len(seq) // n_folds if fold_i < len(seq) % n_folds: stop += 1 yield seq[:start] + seq[stop:], seq[start:stop] return k_folds(self.raw_ratings, self.n_folds)
[docs] def split(self, n_folds=5, shuffle=True): """Split the dataset into folds for futur cross-validation. If you forget to call :meth:`split`, the dataset will be automatically shuffled and split for 5-folds cross-validation. You can obtain repeatable splits over your all your experiments by seeding the RNG: :: import random random.seed(my_seed) # call this before you call split! Args: n_folds(:obj:`int`): The number of folds. shuffle(:obj:`bool`): Whether to shuffle ratings before splitting. If ``False``, folds will always be the same each time the experiment is run. Default is ``True``. """ self.n_folds = n_folds self.shuffle = shuffle
[docs]class Reader(): """The Reader class is used to parse a file containing ratings. Such a file is assumed to specify only one rating per line, and each line needs to respect the following structure: :: user ; item ; rating ; [timestamp] where the order of the fields and the seperator (here ';') may be arbitrarily defined (see below). brackets indicate that the timestamp field is optional. Args: name(:obj:`string`, optional): If specified, a Reader for one of the built-in datasets is returned and any other parameter is ignored. Accepted values are 'ml-100k', 'ml-1m', and 'jester'. Default is ``None``. line_format(:obj:`string`): The fields names, in the order at which they are encountered on a line. Example: ``'item user rating'``. sep(char): the separator between fields. Example : ``';'``. rating_scale(:obj:`tuple`, optional): The rating scale used for every rating. Default is ``(1, 5)``. skip_lines(:obj:`int`, optional): Number of lines to skip at the beginning of the file. Default is ``0``. """ def __init__(self, name=None, line_format=None, sep=None, rating_scale=(1, 5), skip_lines=0): if name: try: self.__init__(**BUILTIN_DATASETS[name].reader_params) except KeyError: raise ValueError('unknown reader ' + name + '. Accepted values are ' + ', '.join(BUILTIN_DATASETS.keys()) + '.') else: self.sep = sep self.skip_lines = skip_lines self.inf, self.sup = rating_scale self.offset = -self.inf + 1 if self.inf <= 0 else 0 splitted_format = line_format.split() entities = ['user', 'item', 'rating'] if 'timestamp' in splitted_format: self.with_timestamp = True entities.append('timestamp') else: self.with_timestamp = False # check that all fields are correct if any(field not in entities for field in splitted_format): raise ValueError('line_format parameter is incorrect.') self.indexes = [splitted_format.index(entity) for entity in entities] def parse_line(self, line): '''Parse a line. Args: line(str): The line to parse Returns: tuple: User id, item id, rating and timestamp. The timestamp is set to ``None`` if it does no exist. ''' line = line.split(self.sep) try: if self.with_timestamp: uid, iid, r, timestamp = (line[i].strip().strip('"') for i in self.indexes) else: uid, iid, r = (line[i].strip().strip('"') for i in self.indexes) timestamp = None except IndexError: raise ValueError(('Impossible to parse line.' + ' Check the line_format and sep parameters.')) return uid, iid, float(r) + self.offset, timestamp
[docs]class Trainset: """A trainset contains all useful data that constitutes a training set. It is used by the :meth:`train() <surprise.prediction_algorithms.algo_base.AlgoBase.train>` method of every prediction algorithm. You should not try to built such an object on your own but rather use the :meth:`Dataset.folds` method or the :meth:`DatasetAutoFolds.build_full_trainset` method. Attributes: rm(:obj:`defaultdict` of :obj:`int`): A dictionary containing all known ratings. Keys are tuples (user_inner__id, item_inner_id), values are ratings. ``rm`` stands for *ratings matrix*, even though it's not a proper matrix object. ur(:obj:`defaultdict` of :obj:`list`): A dictionary containing lists of tuples of the form ``(item_inner_id, rating)``. Keys are user inner ids. ``ur`` stands for *user ratings*. ir(:obj:`defaultdict` of :obj:`list`): A dictionary containing lists of tuples of the form ``(user_inner_id, rating)``. Keys are item inner ids. ``ir`` stands for *item ratings*. n_users: Total number of users :math:`|U|`. n_items: Total number of items :math:`|I|`. n_ratings: Total number of ratings :math:`|R_{train}|`. r_min: Minimum value of the rating scale. r_max: Maximum value of the rating scale. global_mean: The mean of all ratings :math:`\\mu`. """ def __init__(self, rm, ur, ir, n_users, n_items, r_min, r_max, raw2inner_id_users, raw2inner_id_items): self.rm = rm self.ur = ur self.ir = ir self.n_users = n_users self.n_items = n_items self.n_ratings = len(self.rm) self.r_min = r_min self.r_max = r_max self._raw2inner_id_users = raw2inner_id_users self._raw2inner_id_items = raw2inner_id_items self._global_mean = None
[docs] def knows_user(self, uid): """Indicate if the user is part of the trainset. A user is part of the trainset if the user has at least one rating. Args: uid: The (inner) user id. See :ref:`this note<raw_inner_note>`. Returns: ``True`` if user is part of the trainset, else ``False``. """ return uid in self.ur
[docs] def knows_item(self, iid): """Indicate if the item is part of the trainset. An item is part of the trainset if the item was rated at least once. Args: iid: The (inner) item id. See :ref:`this note<raw_inner_note>`. Returns: ``True`` if item is part of the trainset, else ``False``. """ return iid in self.ir
[docs] def to_inner_uid(self, ruid): """Convert a raw **user** id to an inner id. See :ref:`this note<raw_inner_note>`. Args: ruid: The user raw id. Returns: The user inner id. Raises: ValueError: When user is not part of the trainset. """ try: return self._raw2inner_id_users[ruid] except KeyError: raise ValueError(('User ' + str(ruid) + ' is not part of the trainset.'))
[docs] def to_inner_iid(self, riid): """Convert a raw **item** id to an inner id. See :ref:`this note<raw_inner_note>`. Args: riid: The item raw id. Returns: The item inner id. Raises: ValueError: When item is not part of the trainset. """ try: return self._raw2inner_id_items[riid] except KeyError: raise ValueError(('Item ' + str(riid) + ' is not part of the trainset.'))
[docs] def all_ratings(self): """Generator function to iterate over all ratings. Yields: A tuple ``(uid, iid, rating)`` where ids are inner ids. """ for u, u_ratings in iteritems(self.ur): for i, r in u_ratings: yield u, i, r
[docs] def all_users(self): """Generator function to iterate over all users. Yields: Inner id of users. """ return range(self.n_users)
[docs] def all_items(self): """Generator function to iterate over all items. Yields: Inner id of items. """ return range(self.n_items)
@property def global_mean(self): """Return the mean of all ratings. It's only computed once.""" if self._global_mean is None: self._global_mean = np.mean( [r for (_, _, r) in self.all_ratings()]) return self._global_mean