Python numpy.int8() Examples

def parse_data(path, dataset, flatten):
    if dataset != 'train' and dataset != 't10k':
        raise NameError('dataset must be train or t10k')

    label_file = os.path.join(path, dataset + '-labels-idx1-ubyte')
    with open(label_file, 'rb') as file:
        _, num = struct.unpack(">II", file.read(8))
        labels = np.fromfile(file, dtype=np.int8)  # int8
        new_labels = np.zeros((num, 10))
        new_labels[np.arange(num), labels] = 1

    img_file = os.path.join(path, dataset + '-images-idx3-ubyte')
    with open(img_file, 'rb') as file:
        _, num, rows, cols = struct.unpack(">IIII", file.read(16))
        imgs = np.fromfile(file, dtype=np.uint8).reshape(num, rows, cols)  # uint8
        imgs = imgs.astype(np.float32) / 255.0
        if flatten:
            imgs = imgs.reshape([num, -1])

    return imgs, new_labels 

def find_match(self, pred, gt):
    '''
    Match component to balls.
    '''
    batch_size, n_frames_input, n_components, _ = pred.shape
    diff = pred.reshape(batch_size, n_frames_input, n_components, 1, 2) - \
               gt.reshape(batch_size, n_frames_input, 1, n_components, 2)
    diff = np.sum(np.sum(diff ** 2, axis=-1), axis=1)
    # Direct indices
    indices = np.argmin(diff, axis=2)
    ambiguous = np.zeros(batch_size, dtype=np.int8)
    for i in range(batch_size):
      _, counts = np.unique(indices[i], return_counts=True)
      if not np.all(counts == 1):
        ambiguous[i] = 1
    return indices, ambiguous 

def test_quantize_float32_to_int8():
    shape = rand_shape_nd(4)
    data = rand_ndarray(shape, 'default', dtype='float32')
    min_range = mx.nd.min(data)
    max_range = mx.nd.max(data)
    qdata, min_val, max_val = mx.nd.contrib.quantize(data, min_range, max_range, out_type='int8')
    data_np = data.asnumpy()
    min_range = min_range.asscalar()
    max_range = max_range.asscalar()
    real_range = np.maximum(np.abs(min_range), np.abs(max_range))
    quantized_range = 127.0
    scale = quantized_range / real_range
    assert qdata.dtype == np.int8
    assert min_val.dtype == np.float32
    assert max_val.dtype == np.float32
    assert same(min_val.asscalar(), -real_range)
    assert same(max_val.asscalar(), real_range)
    qdata_np = (np.sign(data_np) * np.minimum(np.abs(data_np) * scale + 0.5, quantized_range)).astype(np.int8)
    assert_almost_equal(qdata.asnumpy(), qdata_np, atol = 1) 

def test_quantized_flatten():
    def check_quantized_flatten(shape, qdtype):
        if qdtype == 'uint8':
            data_low = 0.0
            data_high = 127.0
        else:
            data_low = -127.0
            data_high = 127.0
        qdata = mx.nd.random.uniform(low=data_low, high=data_high, shape=shape).astype(qdtype)
        min_data = mx.nd.array([-1023.343], dtype='float32')
        max_data = mx.nd.array([2343.324275], dtype='float32')
        qoutput, min_output, max_output = mx.nd.contrib.quantized_flatten(qdata, min_data, max_data)
        assert qoutput.ndim == 2
        assert qoutput.shape[0] == qdata.shape[0]
        assert qoutput.shape[1] == np.prod(qdata.shape[1:])
        assert same(qdata.asnumpy().flatten(), qoutput.asnumpy().flatten())
        assert same(min_data.asnumpy(), min_output.asnumpy())
        assert same(max_data.asnumpy(), max_output.asnumpy())

    for qdtype in ['int8', 'uint8']:
        check_quantized_flatten((10,), qdtype)
        check_quantized_flatten((10, 15), qdtype)
        check_quantized_flatten((10, 15, 18), qdtype)
        check_quantized_flatten((3, 4, 23, 23), qdtype) 

def main():
    print('create y...')
    y = np.random.randint(2, size=N_OBS)
    print('create X...')
    row = np.random.randint(N_OBS, size=N_VALUE)
    col = np.random.randint(N_FEATURE, size=N_VALUE)
    data = np.ones(N_VALUE)
    X = sparse.csr_matrix((data, (row, col)), dtype=np.int8)

    print('train...')
    profiler = cProfile.Profile(subcalls=True, builtins=True, timeunit=0.001,)
    clf = FTRL(interaction=False)
    profiler.enable()
    clf.fit(X, y)
    profiler.disable()
    profiler.print_stats()

    p = clf.predict(X)
    print('AUC: {:.4f}'.format(auc(y, p)))

    assert auc(y, p) > .5 

def frompointer(pointer, count, dtype=float):
    '''Interpret a buffer that the pointer refers to as a 1-dimensional array.

    Args:
        pointer : int or ctypes pointer
            address of a buffer
        count : int
            Number of items to read.
        dtype : data-type, optional
            Data-type of the returned array; default: float.

    Examples:

    >>> s = numpy.ones(3, dtype=numpy.int32)
    >>> ptr = s.ctypes.data
    >>> frompointer(ptr, count=6, dtype=numpy.int16)
    [1, 0, 1, 0, 1, 0]
    '''
    dtype = numpy.dtype(dtype)
    count *= dtype.itemsize
    buf = (ctypes.c_char * count).from_address(pointer)
    a = numpy.ndarray(count, dtype=numpy.int8, buffer=buf)
    return a.view(dtype) 

def test_no_offset_scale():
    # Specific tests of no-offset scaling
    SAW = SlopeArrayWriter
    # Floating point
    for data in ((-128, 127),
                  (-128, 126),
                  (-128, -127),
                  (-128, 0),
                  (-128, -1),
                  (126, 127),
                  (-127, 127)):
        aw = SAW(np.array(data, dtype=np.float32), np.int8)
        assert_equal(aw.slope, 1.0)
    aw = SAW(np.array([-126, 127 * 2.0], dtype=np.float32), np.int8)
    assert_equal(aw.slope, 2)
    aw = SAW(np.array([-128 * 2.0, 127], dtype=np.float32), np.int8)
    assert_equal(aw.slope, 2)
    # Test that nasty abs behavior does not upset us
    n = -2**15
    aw = SAW(np.array([n, n], dtype=np.int16), np.uint8)
    assert_array_almost_equal(aw.slope, n / 255.0, 5) 

def test_calculate_scale():
    # Test for special cases in scale calculation
    npa = np.array
    # Here the offset handles it
    res = calculate_scale(npa([-2, -1], dtype=np.int8), np.uint8, True)
    assert_equal(res, (1.0, -2.0, None, None))
    # Not having offset not a problem obviously
    res = calculate_scale(npa([-2, -1], dtype=np.int8), np.uint8, 0)
    assert_equal(res, (-1.0, 0.0, None, None))
    # Case where offset handles scaling
    res = calculate_scale(npa([-1, 1], dtype=np.int8), np.uint8, 1)
    assert_equal(res, (1.0, -1.0, None, None))
    # Can't work for no offset case
    assert_raises(ValueError,
                  calculate_scale, npa([-1, 1], dtype=np.int8), np.uint8, 0)
    # Offset trick can't work when max is out of range
    res = calculate_scale(npa([-1, 255], dtype=np.int16), np.uint8, 1)
    assert_not_equal(res, (1.0, -1.0, None, None)) 

def test_a2f_min_max():
    str_io = BytesIO()
    for in_dt in (np.float32, np.int8):
        for out_dt in (np.float32, np.int8):
            arr = np.arange(4, dtype=in_dt)
            # min thresholding
            data_back = write_return(arr, str_io, out_dt, 0, 0, 1, 1)
            assert_array_equal(data_back, [1, 1, 2, 3])
            # max thresholding
            data_back = write_return(arr, str_io, out_dt, 0, 0, 1, None, 2)
            assert_array_equal(data_back, [0, 1, 2, 2])
            # min max thresholding
            data_back = write_return(arr, str_io, out_dt, 0, 0, 1, 1, 2)
            assert_array_equal(data_back, [1, 1, 2, 2])
    # Check that works OK with scaling and intercept
    arr = np.arange(4, dtype=np.float32)
    data_back = write_return(arr, str_io, np.int, 0, -1, 0.5, 1, 2)
    assert_array_equal(data_back * 0.5 - 1, [1, 1, 2, 2])
    # Even when scaling is negative
    data_back = write_return(arr, str_io, np.int, 0, 1, -0.5, 1, 2)
    assert_array_equal(data_back * -0.5 + 1, [1, 1, 2, 2]) 

def test_can_cast():
    tests = ((np.float32, np.float32, True, True, True),
             (np.float64, np.float32, True, True, True),
             (np.complex128, np.float32, False, False, False),
             (np.float32, np.complex128, True, True, True),
             (np.float32, np.uint8, False, True, True),
             (np.uint32, np.complex128, True, True, True),
             (np.int64, np.float32, True, True, True),
             (np.complex128, np.int16, False, False, False),
             (np.float32, np.int16, False, True, True),
             (np.uint8, np.int16, True, True, True),
             (np.uint16, np.int16, False, True, True),
             (np.int16, np.uint16, False, False, True),
             (np.int8, np.uint16, False, False, True),
             (np.uint16, np.uint8, False, True, True),
             )
    for intype, outtype, def_res, scale_res, all_res in tests:
        assert_equal(def_res, can_cast(intype, outtype))
        assert_equal(scale_res, can_cast(intype, outtype, False, True))
        assert_equal(all_res, can_cast(intype, outtype, True, True)) 

def testEnumArrayTypeItem(self):
        mapping = {'RED': 0, 'GREEN': 1, 'BLUE': 2}
        dt_enum = special_dtype(enum=(np.int8, mapping))
        typeItem = hdf5dtype.getTypeItem(dt_enum)
        dt_array = np.dtype('(2,3)'+dt_enum.str, metadata=dict(dt_enum.metadata))

        typeItem = hdf5dtype.getTypeItem(dt_array)

        self.assertEqual(typeItem['class'], 'H5T_ARRAY')
        self.assertTrue("dims" in typeItem)
        self.assertEqual(typeItem["dims"], (2,3))
        baseItem = typeItem['base']
        self.assertEqual(baseItem['class'], 'H5T_ENUM')
        self.assertTrue('mapping' in baseItem)
        self.assertEqual(baseItem['mapping']['GREEN'], 1)
        self.assertTrue("base" in baseItem)
        basePrim = baseItem["base"]
        self.assertEqual(basePrim["class"], 'H5T_INTEGER')
        self.assertEqual(basePrim['base'], 'H5T_STD_I8LE')
        typeSize = hdf5dtype.getItemSize(typeItem)
        self.assertEqual(typeSize, 6)  # one-byte for base enum type * shape of (2,3) 

def test_payload_getitem_setitem(self, item):
        data = self.payload.data
        sel_data = data[item]
        assert np.all(self.payload[item] == sel_data)
        payload = self.Payload(self.payload.words.copy(), sample_shape=(2,),
                               bps=8, complex_data=False)
        assert payload == self.payload
        payload[item] = 1 - sel_data
        check = self.payload.data
        check[item] = 1 - sel_data
        assert np.all(payload[item] == 1 - sel_data)
        assert np.all(payload.data == check)
        assert np.all(payload[:]
                      == payload.words.view(np.int8).reshape(-1, 2))
        assert payload != self.payload
        payload[item] = sel_data
        assert np.all(payload[item] == sel_data)
        assert payload == self.payload
        payload = self.Payload.fromdata(data + 1j * data, bps=8)
        sel_data = payload.data[item]
        assert np.all(payload[item] == sel_data)
        payload[item] = 1 - sel_data
        check = payload.data
        check[item] = 1 - sel_data
        assert np.all(payload.data == check) 

def test_payload_getitem_setitem(self, item):
        data = self.payload.data
        sel_data = data[item]
        assert np.all(self.payload[item] == sel_data)
        payload = self.Payload(self.payload.words.copy(), sample_shape=(2,),
                               bps=8, complex_data=False)
        assert payload == self.payload
        payload[item] = 1 - sel_data
        check = self.payload.data
        check[item] = 1 - sel_data
        assert np.all(payload[item] == 1 - sel_data)
        assert np.all(payload.data == check)
        assert np.all(payload[:]
                      == payload.words.view(np.int8).reshape(-1, 2))
        assert payload != self.payload
        payload[item] = sel_data
        assert np.all(payload[item] == sel_data)
        assert payload == self.payload
        payload = self.Payload.fromdata(data + 1j * data, bps=8)
        sel_data = payload.data[item]
        assert np.all(payload[item] == sel_data)
        payload[item] = 1 - sel_data
        check = payload.data
        check[item] = 1 - sel_data
        assert np.all(payload.data == check) 

def _convert(self, vals):
        res = {}
        for k, v in vals.items():
            if isinstance(v, (np.int, np.int8, np.int16, np.int32, np.int64)):
                v = int(v)
            elif isinstance(v, (np.float, np.float16, np.float32, np.float64)):
                v = float(v)
            elif isinstance(v, Labels):
                v = list(v)
            elif isinstance(v, np.ndarray):
                v = v.tolist()
            elif isinstance(v, dict):
                v = self._convert(v)
            res[k] = v
        return res 

def _toscalar(v):
    if isinstance(v, (np.float16, np.float32, np.float64,
                      np.uint8, np.uint16, np.uint32, np.uint64,
                      np.int8, np.int16, np.int32, np.int64)):
        return np.asscalar(v)
    else:
        return v 

def convert(self, complex_iq):
        intlv = self._interleave(complex_iq)
        clipped = self._clip(intlv)
        converted = 127. * clipped
        hackrf_out = converted.astype(np.int8)
        return hackrf_out 

def annotate(anno_file, chromo, pos):
    #anno_file = dat.GzipFile(anno_file, 'r')
    anno_file = get_fh(anno_file, 'r')
    anno = pd.read_table(anno_file, header=None, usecols=[0, 1, 2],
                         dtype={0: 'str', 1: 'int32', 2: 'int32'})
    anno_file.close()
    anno.columns = ['chromo', 'start', 'end']
    anno.chromo = anno.chromo.str.upper().str.replace('CHR', '')
    anno = anno.loc[anno.chromo == chromo]
    anno.sort_values('start', inplace=True)
    start, end = an.join_overlapping(anno.start.values, anno.end.values)
    anno = np.array(an.is_in(pos, start, end), dtype='int8')
    return anno 

def parse_type(self, hdat, dataobj=None):
        dtype = super(MGHImageType, self).parse_type(hdat, dataobj=dataobj)
        if   np.issubdtype(dtype, np.floating): dtype = np.float32
        elif np.issubdtype(dtype, np.int8):     dtype = np.int8
        elif np.issubdtype(dtype, np.int16):    dtype = np.int16
        elif np.issubdtype(dtype, np.integer):  dtype = np.int32
        else: raise ValueError('Could not deduce appropriate MGH type for dtype %s' % dtype)
        return dtype 

def read_data(label_url, image_url):
    with gzip.open(download_data(label_url)) as flbl:
        magic, num = struct.unpack(">II", flbl.read(8))
        label = np.fromstring(flbl.read(), dtype=np.int8)
    with gzip.open(download_data(image_url), 'rb') as fimg:
        magic, num, rows, cols = struct.unpack(">IIII", fimg.read(16))
        image = np.fromstring(fimg.read(), dtype=np.uint8).reshape(len(label), rows, cols)
    return label, image 

def read_data(label, image):
    """
    download and read data into numpy
    """
    base_url = 'http://yann.lecun.com/exdb/mnist/'
    with gzip.open(download_file(base_url+label, os.path.join('data',label))) as flbl:
        magic, num = struct.unpack(">II", flbl.read(8))
        label = np.fromstring(flbl.read(), dtype=np.int8)
    with gzip.open(download_file(base_url+image, os.path.join('data',image)), 'rb') as fimg:
        magic, num, rows, cols = struct.unpack(">IIII", fimg.read(16))
        image = np.fromstring(fimg.read(), dtype=np.uint8).reshape(len(label), rows, cols)
    return (label, image) 

def get_mnist():
    """Download and load the MNIST dataset

    Returns
    -------
    dict
        A dict containing the data
    """
    def read_data(label_url, image_url):
        with gzip.open(mx.test_utils.download(label_url)) as flbl:
            struct.unpack(">II", flbl.read(8))
            label = np.frombuffer(flbl.read(), dtype=np.int8)
        with gzip.open(mx.test_utils.download(image_url), 'rb') as fimg:
            _, _, rows, cols = struct.unpack(">IIII", fimg.read(16))
            image = np.frombuffer(fimg.read(), dtype=np.uint8).reshape(len(label), rows, cols)
            image = image.reshape(image.shape[0], 1, 28, 28).astype(np.float32)/255
        return (label, image)

    # changed to mxnet.io for more stable hosting
    # path = 'http://yann.lecun.com/exdb/mnist/'
    path = 'http://data.mxnet.io/data/mnist/'
    (train_lbl, train_img) = read_data(
        path+'train-labels-idx1-ubyte.gz', path+'train-images-idx3-ubyte.gz')
    (test_lbl, test_img) = read_data(
        path+'t10k-labels-idx1-ubyte.gz', path+'t10k-images-idx3-ubyte.gz')
    return {'train_data':train_img, 'train_label':train_lbl,
            'test_data':test_img, 'test_label':test_lbl} 

def test_ndarray_elementwise():
    nrepeat = 10
    maxdim = 4
    all_type = [np.float32, np.float64, np.float16, np.uint8, np.int8, np.int32, np.int64]
    real_type = [np.float32, np.float64, np.float16]
    for repeat in range(nrepeat):
        for dim in range(1, maxdim):
            check_with_uniform(lambda x, y: x + y, 2, dim, type_list=all_type)
            check_with_uniform(lambda x, y: x - y, 2, dim, type_list=all_type)
            check_with_uniform(lambda x, y: x * y, 2, dim, type_list=all_type)
            check_with_uniform(lambda x, y: x / y, 2, dim, type_list=real_type)
            check_with_uniform(lambda x, y: x / y, 2, dim, rmin=1, type_list=all_type)
            check_with_uniform(mx.nd.sqrt, 1, dim, np.sqrt, rmin=0)
            check_with_uniform(mx.nd.square, 1, dim, np.square, rmin=0)
            check_with_uniform(lambda x: mx.nd.norm(x).asscalar(), 1, dim, np.linalg.norm) 

def _pad(seq, target_len):
  """Pads a sequence of word embeddings up to the target length.

  Args:
    seq: Sequence of word embeddings.
    target_len: Desired padded sequence length.

  Returns:
    embeddings: Input sequence padded with zero embeddings up to the target
      length.
    mask: A 0/1 vector with zeros corresponding to padded embeddings.

  Raises:
    ValueError: If len(seq) is not in the interval (0, target_len].
  """
  seq_len = len(seq)
  if seq_len <= 0 or seq_len > target_len:
    raise ValueError("Expected 0 < len(seq) <= %d, got %d" % (target_len,
                                                              seq_len))

  emb_dim = seq[0].shape[0]
  padded_seq = np.zeros(shape=(target_len, emb_dim), dtype=seq[0].dtype)
  mask = np.zeros(shape=(target_len,), dtype=np.int8)
  for i in range(seq_len):
    padded_seq[i] = seq[i]
    mask[i] = 1
  return padded_seq, mask 

def GenerateSample(filename, code_shape, layer_depth):
  # {0, +1} binary codes.
  # No conversion since the output file is expected to store
  # codes using {0, +1} codes (and not {-1, +1}).
  code = synthetic_model.GenerateSingleCode(code_shape)
  code = np.round(code)

  # Reformat the code so as to be compatible with what is generated
  # by the image encoder.
  # The image encoder generates a tensor of size:
  # iteration_count x batch_size x height x width x iteration_depth.
  # Here: batch_size = 1
  if code_shape[-1] % layer_depth != 0:
    raise ValueError('Number of layers is not an integer')
  height = code_shape[0]
  width = code_shape[1]
  code = code.reshape([1, height, width, -1, layer_depth])
  code = np.transpose(code, [3, 0, 1, 2, 4])

  int_codes = code.astype(np.int8)
  exported_codes = np.packbits(int_codes.reshape(-1))

  output = io.BytesIO()
  np.savez_compressed(output, shape=int_codes.shape, codes=exported_codes)
  with tf.gfile.FastGFile(filename, 'wb') as code_file:
    code_file.write(output.getvalue()) 

def decode(self, ids, strip_extraneous=False):
    del strip_extraneous
    label_id = ids
    if isinstance(label_id, np.ndarray):
      label_id = np.squeeze(label_id).astype(np.int8).tolist()
    assert isinstance(label_id, list)
    assert len(label_id) == self.vocab_size
    return self._class_labels[label_id.index(1)] 

def load_dataset(path):
    download_dataset(path)

    # training data
    data = [np.load(os.path.join(path, 'cifar-10-batches-py',
                                 'data_batch_%d' % (i + 1))) for i in range(5)]
    X_train = np.vstack([d['data'] for d in data])
    y_train = np.hstack([np.asarray(d['labels'], np.int8) for d in data])

    # test data
    data = np.load(os.path.join(path, 'cifar-10-batches-py', 'test_batch'))
    X_test = data['data']
    y_test = np.asarray(data['labels'], np.int8)

    # reshape
    X_train = X_train.reshape(-1, 3, 32, 32)
    X_test = X_test.reshape(-1, 3, 32, 32)

    # normalize
    try:
        mean_std = np.load(os.path.join(path, 'cifar-10-mean_std.npz'))
        mean = mean_std['mean']
        std = mean_std['std']
    except IOError:
        mean = X_train.mean(axis=(0, 2, 3), keepdims=True).astype(np.float32)
        std = X_train.std(axis=(0, 2, 3), keepdims=True).astype(np.float32)
        np.savez(os.path.join(path, 'cifar-10-mean_std.npz'),
                 mean=mean, std=std)
    X_train = (X_train - mean) / std
    X_test = (X_test - mean) / std

    return X_train, y_train, X_test, y_test 

def _create_shared_arr(self):
        TYPE = {
            np.float32: ctypes.c_float,
            np.float64: ctypes.c_double,
            np.uint8: ctypes.c_uint8,
            np.int8: ctypes.c_int8,
            np.int32: ctypes.c_int32,
        }
        ctype = TYPE[self.output_dtype]
        arr = mp.RawArray(ctype, int(np.prod(self.output_shape)))
        return arr 

def int_abs(arr):
    """ Absolute values of array taking care of max negative int values

    Parameters
    ----------
    arr : array-like

    Returns
    -------
    abs_arr : array
        array the same shape as `arr` in which all negative numbers have been
        changed to positive numbers with the magnitude.

    Examples
    --------
    This kind of thing is confusing in base numpy:

    >>> import numpy as np
    >>> np.abs(np.int8(-128))
    -128

    ``int_abs`` fixes that:

    >>> int_abs(np.int8(-128))
    128
    >>> int_abs(np.array([-128, 127], dtype=np.int8))
    array([128, 127], dtype=uint8)
    >>> int_abs(np.array([-128, 127], dtype=np.float32))
    array([ 128.,  127.], dtype=float32)
    """
    arr = np.array(arr, copy=False)
    dt = arr.dtype
    if dt.kind == 'u':
        return arr
    if dt.kind != 'i':
        return np.absolute(arr)
    out = arr.astype(np.dtype(dt.str.replace('i', 'u')))
    return np.choose(arr < 0, (arr, arr * -1), out=out) 

def able_int_type(values):
    """ Find the smallest integer numpy type to contain sequence `values`

    Prefers uint to int if minimum is >= 0

    Parameters
    ----------
    values : sequence
        sequence of integer values

    Returns
    -------
    itype : None or numpy type
        numpy integer type or None if no integer type holds all `values`

    Examples
    --------
    >>> able_int_type([0, 1]) == np.uint8
    True
    >>> able_int_type([-1, 1]) == np.int8
    True
    """
    if any([v % 1 for v in values]):
        return None
    mn = min(values)
    mx = max(values)
    if mn >= 0:
        for ityp in np.sctypes['uint']:
            if mx <= np.iinfo(ityp).max:
                return ityp
    for ityp in np.sctypes['int']:
        info = np.iinfo(ityp)
        if mn >= info.min and mx <= info.max:
            return ityp
    return None 

def __init__(self, array, out_dtype=None, calc_scale=True):
        """ Initialize array writer

        Parameters
        ----------
        array : array-like
            array-like object
        out_dtype : None or dtype
            dtype with which `array` will be written.  For this class,
            `out_dtype`` needs to be the same as the dtype of the input `array`
            or a swapped version of the same.
        \*\*kwargs : keyword arguments

        Examples
        --------
        >>> arr = np.array([0, 255], np.uint8)
        >>> aw = ArrayWriter(arr)
        >>> aw = ArrayWriter(arr, np.int8) #doctest: +IGNORE_EXCEPTION_DETAIL
        Traceback (most recent call last):
            ...
        WriterError: Scaling needed but cannot scale
        """
        self._array = np.asanyarray(array)
        arr_dtype = self._array.dtype
        if out_dtype is None:
            out_dtype = arr_dtype
        else:
            out_dtype = np.dtype(out_dtype)
        self._out_dtype = out_dtype
        self._finite_range = None
        if self.scaling_needed():
            raise WriterError("Scaling needed but cannot scale")