Skip to content

neurite.py.data

Data resources for neurite.

DataSplit 

DataSplit(train=None, val=None, test=None)

initialize DataSplit object, maintains splits of train, val, test

data can be accessed using member variables, like data.train

or as a victionary, using: data['train']

Args: train ([type], optional): [description]. Defaults to None. val ([type], optional): [description]. Defaults to None. test ([type], optional): [description]. Defaults to None.

Source code in neurite/py/data.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
def __init__(self, train=None, val=None, test=None):
    """ initialize DataSplit object, maintains splits of train, val, test

    data can be accessed using member variables, like
    data.train

    or as a victionary, using:
    data['train']

    Args:
        train ([type], optional): [description]. Defaults to None.
        val ([type], optional): [description]. Defaults to None.
        test ([type], optional): [description]. Defaults to None.
    """
    self._splits = []
    self.train = None
    self.val = None
    self.test = None

map_fn 

map_fn(lambda_fn, splits=None)

apply function to each of the data splits

Args: lambda_fn (function): function that takes in one input splits (list, optional): which splits to do processing on. Defaults to ['train', 'val', 'test'].

Source code in neurite/py/data.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
def map_fn(self, lambda_fn, splits=None):
    """ apply function to each of the data splits

    Args:
        lambda_fn (function): function that takes in one input
        splits (list, optional): which splits to do processing on.
            Defaults to ['train', 'val', 'test'].
    """

    if splits is None:
        splits = ['train', 'val', 'test']

    for attr in splits:
        self[attr] = lambda_fn[self[attr]]

load_dataset 

load_dataset(dataset)

Downloads a dataset and caches it in the user's home directory.

Source code in neurite/py/data.py 
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
def load_dataset(dataset):
    """
    Downloads a dataset and caches it in the user's home directory.
    """
    import urllib.request
    datadir = Path.home().joinpath('.neurite')
    datadir.mkdir(exist_ok=True)

    if dataset == '2D-OASIS-TUTORIAL':
        filename = datadir.joinpath('2D-OASIS-TUTORIAL.npz')
        if not filename.exists():
            url = 'https://surfer.nmr.mgh.harvard.edu/pub/data/voxelmorph/2D-OASIS-TUTORIAL.npz'
            urllib.request.urlretrieve(url, filename)
            print(f'Cached dataset in {datadir}.')
        return np.load(filename)['images']
    else:
        raise ValueError(f'Unknown dataset {dataset}.')

split_dataset 

split_dataset(data, ratios, axis=0, randomize=True, rand_seed=None)

split a dataset used to split train in train/val, for example

can input single numpy array or list

Source code in neurite/py/data.py 
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
def split_dataset(data, ratios, axis=0, randomize=True, rand_seed=None):
    """
    split a dataset
    used to split train in train/val, for example

    can input single numpy array or list
    """

    nb_groups = len(ratios)
    is_list = isinstance(data, (list, tuple))

    if is_list:
        nb_items = len(data)
        assert axis == 0, \
            'if data is a list or tuple, axis needs to be 0. got: %d' % axis
    else:
        assert type(data) is np.ndarray, \
            'data should be list, tuple or numpy array, got: %s' % type(data)
        nb_items = data.shape[axis]

    # get slice indices
    cratios = np.cumsum(ratios) / np.sum(ratios)
    sl_idx = [0] + [np.round(c * nb_items).astype(int) for c in cratios]

    # prepare a list of all indices, and shuffle if necessary
    rng = list(range(nb_items))
    if randomize:
        if rand_seed:
            random.seed(rand_seed)
        random.shuffle(rng)

    # extract data
    if is_list:
        split = [data[rng[sl_idx[f]:sl_idx[f + 1]]] for f in range(nb_groups)]
    else:
        split = [np.take(data, rng[sl_idx[f]:sl_idx[f + 1]], axis=axis) for f in range(nb_groups)]

    return split