Python numpy.logical_not() Examples

def test_subsample_all_examples(self):
    numpy_labels = np.random.permutation(300)
    indicator = tf.constant(np.ones(300) == 1)
    numpy_labels = (numpy_labels - 200) > 0

    labels = tf.constant(numpy_labels)

    sampler = (balanced_positive_negative_sampler.
               BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 32)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 32) 

def test_subsample_selection(self):
    # Test random sampling when only some examples can be sampled:
    # 100 samples, 20 positives, 10 positives cannot be sampled
    numpy_labels = np.arange(100)
    numpy_indicator = numpy_labels < 90
    indicator = tf.constant(numpy_indicator)
    numpy_labels = (numpy_labels - 80) >= 0

    labels = tf.constant(numpy_labels)

    sampler = (balanced_positive_negative_sampler.
               BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 10)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 54)
      self.assertAllEqual(is_sampled, np.logical_and(is_sampled,
                                                     numpy_indicator)) 

def _get_result_map(self, mask):
        """Processing mask data"""

        # mask.shape[0]: row, mask.shape[1]: column
        result_map = np.zeros((mask.shape[1], mask.shape[0], self.nb_classes))
        # 0 (background pixel), 128 (face area pixel) or 255 (hair area pixel).
        skin = (mask == 128)
        hair = (mask == 255)

        if self.nb_classes == 2:
            # hair = (mask > 128)
            background = np.logical_not(hair)
            result_map[:, :, 0] = np.where(background, 1, 0)
            result_map[:, :, 1] = np.where(hair, 1, 0)
        elif self.nb_classes == 3:
            background = np.logical_not(hair + skin)
            result_map[:, :, 0] = np.where(background, 1, 0)
            result_map[:, :, 1] = np.where(skin, 1, 0)
            result_map[:, :, 2] = np.where(hair, 1, 0)
        else:
            raise Exception("error...")

        return result_map 

def test_subsample_all_examples(self):
    numpy_labels = np.random.permutation(300)
    indicator = tf.constant(np.ones(300) == 1)
    numpy_labels = (numpy_labels - 200) > 0

    labels = tf.constant(numpy_labels)

    sampler = (balanced_positive_negative_sampler.
               BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 32)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 32) 

def test_subsample_selection(self):
    # Test random sampling when only some examples can be sampled:
    # 100 samples, 20 positives, 10 positives cannot be sampled
    numpy_labels = np.arange(100)
    numpy_indicator = numpy_labels < 90
    indicator = tf.constant(numpy_indicator)
    numpy_labels = (numpy_labels - 80) >= 0

    labels = tf.constant(numpy_labels)

    sampler = (balanced_positive_negative_sampler.
               BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 10)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 54)
      self.assertAllEqual(is_sampled, np.logical_and(is_sampled,
                                                     numpy_indicator)) 

def find_outliers_upper_tail(mu):

    # remove values that are nan
    I = np.where(np.logical_and(np.logical_not(np.isnan(mu)), np.logical_not(np.isinf(mu))))[0]
    mu = mu[I]

    # calculate mean and sigma
    mean, sigma = mu.mean(), mu.std()

    # calculate the deviation in terms of sigma
    deviation = (mu - mean) / sigma

    # 2 * sigma is considered as an outlier
    S = I[np.where(deviation >= 2)[0]]

    if len(S) == 0 and deviation.max() > 1:
        S = I[[np.argmax(mu)]]

    return S if len(S) > 0 else None 

def test_class(self):
        """Tests container behavior."""
        model = kproxy_supercell.TDProxy(self.model_krhf, "hf", [self.k, 1, 1], density_fitting_hf)
        model.nroots = self.td_model_krhf.nroots
        assert not model.fast
        model.kernel()
        testing.assert_allclose(model.e, self.td_model_krhf.e, atol=1e-5)
        # Test real
        testing.assert_allclose(model.e.imag, 0, atol=1e-8)

        nocc = nvirt = 4
        testing.assert_equal(model.xy.shape, (len(model.e), 2, self.k, self.k, nocc, nvirt))

        # Test only non-degenerate roots
        d = abs(model.e[1:] - model.e[:-1]) < 1e-8
        d = numpy.logical_or(numpy.concatenate(([False], d)), numpy.concatenate((d, [False])))
        d = numpy.logical_not(d)
        assert_vectors_close(self.td_model_krhf.xy[d], model.xy[d], atol=1e-5) 

def _node_regr(self, configs, wf):
        """ 
        Return true if a given configuration is within nodal_cutoff 
        of the node 
        Also return the regularization polynomial if true, 
        f = a * r ** 2 + b * r ** 4 + c * r ** 3
        """
        ne = configs.configs.shape[1]
        d2 = 0.0
        for e in range(ne):
            d2 += np.sum(wf.gradient(e, configs.electron(e)) ** 2, axis=0)
        r = 1.0 / d2
        mask = r < self.nodal_cutoff ** 2

        c = 7.0 / (self.nodal_cutoff ** 6)
        b = -15.0 / (self.nodal_cutoff ** 4)
        a = 9.0 / (self.nodal_cutoff ** 2)

        f = a * r + b * r ** 2 + c * r ** 3
        f[np.logical_not(mask)] = 1.0

        return mask, f 

def _node_regr(self, configs, wf):
        """ 
        Return true if a given configuration is within nodal_cutoff 
        of the node 
        Also return the regularization polynomial if true, 
        f = a * r ** 2 + b * r ** 4 + c * r ** 3
        """
        ne = configs.configs.shape[1]
        d2 = 0.0
        for e in range(ne):
            d2 += np.sum(wf.gradient(e, configs.electron(e)) ** 2, axis=0)
        r = 1.0 / d2
        mask = r < self.nodal_cutoff ** 2

        c = 7.0 / (self.nodal_cutoff ** 6)
        b = -15.0 / (self.nodal_cutoff ** 4)
        a = 9.0 / (self.nodal_cutoff ** 2)

        f = a * r + b * r ** 2 + c * r ** 3
        f[np.logical_not(mask)] = 1.0

        return mask, f 

def color_pcl(pcl, K, im, color_axis=0, as_int=True, invalid_color=[0,0,0]):
  uvd = K @ pcl.T
  uvd /= uvd[2]
  uvd = np.round(uvd).astype(np.int32)
  mask = np.logical_and(uvd[0] >= 0, uvd[1] >= 0)
  color = np.empty((pcl.shape[0], 3), dtype=im.dtype)
  if color_axis == 0:
    mask = np.logical_and(mask, uvd[0] < im.shape[2])
    mask = np.logical_and(mask, uvd[1] < im.shape[1])
    uvd = uvd[:,mask]
    color[mask,:] = im[:,uvd[1],uvd[0]].T
  elif color_axis == 2:
    mask = np.logical_and(mask, uvd[0] < im.shape[1])
    mask = np.logical_and(mask, uvd[1] < im.shape[0])
    uvd = uvd[:,mask]
    color[mask,:] = im[uvd[1],uvd[0], :]
  else:
    raise Exception('invalid color_axis')
  color[np.logical_not(mask),:3] = invalid_color
  if as_int:
    color = (255.0 * color).astype(np.int32)
  return color 

def test_subsample_all_examples_dynamic(self):
    numpy_labels = np.random.permutation(300)
    indicator = tf.constant(np.ones(300) == 1)
    numpy_labels = (numpy_labels - 200) > 0

    labels = tf.constant(numpy_labels)

    sampler = (
        balanced_positive_negative_sampler.BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 32)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 32) 

def test_subsample_all_examples_static(self):
    numpy_labels = np.random.permutation(300)
    indicator = np.array(np.ones(300) == 1, np.bool)
    numpy_labels = (numpy_labels - 200) > 0

    labels = np.array(numpy_labels, np.bool)

    def graph_fn(indicator, labels):
      sampler = (
          balanced_positive_negative_sampler.BalancedPositiveNegativeSampler(
              is_static=True))
      return sampler.subsample(indicator, 64, labels)

    is_sampled = self.execute(graph_fn, [indicator, labels])
    self.assertTrue(sum(is_sampled) == 64)
    self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 32)
    self.assertTrue(sum(np.logical_and(
        np.logical_not(numpy_labels), is_sampled)) == 32) 

def test_subsample_selection_dynamic(self):
    # Test random sampling when only some examples can be sampled:
    # 100 samples, 20 positives, 10 positives cannot be sampled
    numpy_labels = np.arange(100)
    numpy_indicator = numpy_labels < 90
    indicator = tf.constant(numpy_indicator)
    numpy_labels = (numpy_labels - 80) >= 0

    labels = tf.constant(numpy_labels)

    sampler = (
        balanced_positive_negative_sampler.BalancedPositiveNegativeSampler())
    is_sampled = sampler.subsample(indicator, 64, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 64)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 10)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 54)
      self.assertAllEqual(is_sampled, np.logical_and(is_sampled,
                                                     numpy_indicator)) 

def test_subsample_selection_static(self):
    # Test random sampling when only some examples can be sampled:
    # 100 samples, 20 positives, 10 positives cannot be sampled.
    numpy_labels = np.arange(100)
    numpy_indicator = numpy_labels < 90
    indicator = np.array(numpy_indicator, np.bool)
    numpy_labels = (numpy_labels - 80) >= 0

    labels = np.array(numpy_labels, np.bool)

    def graph_fn(indicator, labels):
      sampler = (
          balanced_positive_negative_sampler.BalancedPositiveNegativeSampler(
              is_static=True))
      return sampler.subsample(indicator, 64, labels)

    is_sampled = self.execute(graph_fn, [indicator, labels])
    self.assertTrue(sum(is_sampled) == 64)
    self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 10)
    self.assertTrue(sum(np.logical_and(
        np.logical_not(numpy_labels), is_sampled)) == 54)
    self.assertAllEqual(is_sampled, np.logical_and(is_sampled, numpy_indicator)) 

def test_subsample_selection_larger_batch_size_static(self):
    # Test random sampling when total number of examples that can be sampled are
    # less than batch size:
    # 100 samples, 50 positives, 40 positives cannot be sampled, batch size 64.
    # It should still return 64 samples, with 4 of them that couldn't have been
    # sampled.
    numpy_labels = np.arange(100)
    numpy_indicator = numpy_labels < 60
    indicator = np.array(numpy_indicator, np.bool)
    numpy_labels = (numpy_labels - 50) >= 0

    labels = np.array(numpy_labels, np.bool)

    def graph_fn(indicator, labels):
      sampler = (
          balanced_positive_negative_sampler.BalancedPositiveNegativeSampler(
              is_static=True))
      return sampler.subsample(indicator, 64, labels)

    is_sampled = self.execute(graph_fn, [indicator, labels])
    self.assertTrue(sum(is_sampled) == 64)
    self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) >= 10)
    self.assertTrue(
        sum(np.logical_and(np.logical_not(numpy_labels), is_sampled)) >= 50)
    self.assertTrue(sum(np.logical_and(is_sampled, numpy_indicator)) == 60) 

def test_subsample_selection_no_batch_size(self):
    # Test random sampling when only some examples can be sampled:
    # 1000 samples, 6 positives (5 can be sampled).
    numpy_labels = np.arange(1000)
    numpy_indicator = numpy_labels < 999
    indicator = tf.constant(numpy_indicator)
    numpy_labels = (numpy_labels - 994) >= 0

    labels = tf.constant(numpy_labels)

    sampler = (balanced_positive_negative_sampler.
               BalancedPositiveNegativeSampler(0.01))
    is_sampled = sampler.subsample(indicator, None, labels)
    with self.test_session() as sess:
      is_sampled = sess.run(is_sampled)
      self.assertTrue(sum(is_sampled) == 500)
      self.assertTrue(sum(np.logical_and(numpy_labels, is_sampled)) == 5)
      self.assertTrue(sum(np.logical_and(
          np.logical_not(numpy_labels), is_sampled)) == 495)
      self.assertAllEqual(is_sampled, np.logical_and(is_sampled,
                                                     numpy_indicator)) 

def test_2d_with_missing(self):
        # Test cov on 2D variable w/ missing value
        x = self.data
        x[-1] = masked
        x = x.reshape(3, 4)
        valid = np.logical_not(getmaskarray(x)).astype(int)
        frac = np.dot(valid, valid.T)
        xf = (x - x.mean(1)[:, None]).filled(0)
        assert_almost_equal(cov(x),
                            np.cov(xf) * (x.shape[1] - 1) / (frac - 1.))
        assert_almost_equal(cov(x, bias=True),
                            np.cov(xf, bias=True) * x.shape[1] / frac)
        frac = np.dot(valid.T, valid)
        xf = (x - x.mean(0)).filled(0)
        assert_almost_equal(cov(x, rowvar=False),
                            (np.cov(xf, rowvar=False) *
                             (x.shape[0] - 1) / (frac - 1.)))
        assert_almost_equal(cov(x, rowvar=False, bias=True),
                            (np.cov(xf, rowvar=False, bias=True) *
                             x.shape[0] / frac)) 

def test_object_logical(self):
        a = np.array([3, None, True, False, "test", ""], dtype=object)
        assert_equal(np.logical_or(a, None),
                        np.array([x or None for x in a], dtype=object))
        assert_equal(np.logical_or(a, True),
                        np.array([x or True for x in a], dtype=object))
        assert_equal(np.logical_or(a, 12),
                        np.array([x or 12 for x in a], dtype=object))
        assert_equal(np.logical_or(a, "blah"),
                        np.array([x or "blah" for x in a], dtype=object))

        assert_equal(np.logical_and(a, None),
                        np.array([x and None for x in a], dtype=object))
        assert_equal(np.logical_and(a, True),
                        np.array([x and True for x in a], dtype=object))
        assert_equal(np.logical_and(a, 12),
                        np.array([x and 12 for x in a], dtype=object))
        assert_equal(np.logical_and(a, "blah"),
                        np.array([x and "blah" for x in a], dtype=object))

        assert_equal(np.logical_not(a),
                        np.array([not x for x in a], dtype=object))

        assert_equal(np.logical_or.reduce(a), 3)
        assert_equal(np.logical_and.reduce(a), None) 

def number_of_changes(self, percent):
        """
        The 'point of change' algorithm for each pixel,
        vectorized using numpy array operations
        Returns a 1D array
        """

        image_shape = self.shape
        j = np.ones(self.size, dtype=int)
        l = np.ones(self.size, dtype=int)
        result = np.zeros(self.size, dtype=np.float32)
        for repeat in range(1, self.k):
            # Numpy array indexing black magic to obtain an image mask
            # indicating if there is change at this (l, j) time point
            change_mask = (1 - self.K[l, j, np.arange(self.size)]) < percent

            # Where there is change
            l[change_mask] += j[change_mask]
            j[change_mask] = 1
            result[change_mask] += 1

            # Where there is no change
            j[np.logical_not(change_mask)] += 1

        return result 

def iterate_eos_scores(new_scores, eos_idx, existing_cases = None, beam_width=None)->Tuple[Sequence, Sequence, Sequence]:
    """
    Return the indices and scores corresponding to the eos word.
    Meaning of returned values is the same as for iterate_best_score
    """
    nb_cases, v_size = new_scores.shape
    num_cases = np.arange(nb_cases, dtype=np.int32)
    scores = -cuda.to_cpu(new_scores[:, eos_idx])
    if existing_cases is not None:
        need_to_return = np.logical_not(np.isin(num_cases, existing_cases))
        num_cases = num_cases[need_to_return]
        scores = scores[need_to_return]

    idx_in_cases = np.full(num_cases.shape[0], eos_idx, dtype=np.int32)

    if beam_width is not None:
        if beam_width < len(scores):
            idx_to_keep = np.argpartition(scores, beam_width)[:beam_width]
            scores = scores[idx_to_keep]
            num_cases = num_cases[idx_to_keep]
            idx_in_cases = idx_in_cases[idx_to_keep]

    return num_cases, idx_in_cases, scores 

def _shuffle_roidb_inds(self):
    """Randomly permute the training roidb."""
    # If the random flag is set, 
    # then the database is shuffled according to system time
    # Useful for the validation set
    if self._random:
      st0 = np.random.get_state()
      millis = int(round(time.time() * 1000)) % 4294967295
      np.random.seed(millis)
    
    if cfg.TRAIN.ASPECT_GROUPING:
      raise NotImplementedError
      '''
      widths = np.array([r['width'] for r in self._roidb])
      heights = np.array([r['height'] for r in self._roidb])
      horz = (widths >= heights)
      vert = np.logical_not(horz)
      horz_inds = np.where(horz)[0]
      vert_inds = np.where(vert)[0]
      inds = np.hstack((
          np.random.permutation(horz_inds),
          np.random.permutation(vert_inds)))
      inds = np.reshape(inds, (-1, 2))
      row_perm = np.random.permutation(np.arange(inds.shape[0]))
      inds = np.reshape(inds[row_perm, :], (-1,))
      self._perm = inds
      '''
    else:
      self._perm = np.random.permutation(np.arange(len(self._roidb)))
    # Restore the random state
    if self._random:
      np.random.set_state(st0)
      
    self._cur = 0 

def white_to_pial_vectors(white_surface, pial_surface):
        '''
        cortex.white_to_pial_vectors is a (3 x n) matrix of the unit direction vectors that point
          from the n vertices in the cortex's white surface to their equivalent positions in the
          pial surface.
        '''
        u = pial_surface.coordinates - white_surface.coordinates
        d = np.sqrt(np.sum(u**2, axis=0))
        z = np.isclose(d, 0)
        return pimms.imm_array(np.logical_not(z) / (d + z)) 

def normalize(u):
    '''
    normalize(u) yields a vetor with the same direction as u but unit length, or, if u has zero
    length, yields u.
    '''
    u = np.asarray(u)
    unorm = np.sqrt(np.sum(u**2, axis=0))
    z = np.isclose(unorm, 0)
    c = np.logical_not(z) / (unorm + z)
    return u * c 

def tetrahedral_barycentric_coordinates(tetra, pt):
    '''
    tetrahedral_barycentric_coordinates(tetrahedron, point) yields a list of weights for each vertex
      in the given tetrahedron in the same order as the vertices given. If all weights are 0, then
      the point is not inside the tetrahedron.
    '''
    # I found a description of this algorithm here (Nov. 2017):
    # http://steve.hollasch.net/cgindex/geometry/ptintet.html
    tetra = np.asarray(tetra)
    pt = np.asarray(pt)
    if tetra.shape[0] != 4:
        if tetra.shape[1] == 4:
            if tetra.shape[0] == 3:
                tetra = np.transpose(tetra, (1,0) if len(tetra.shape) == 2 else (1,0,2))
            else:
                tetra = np.transpose(tetra, (1,2,0))
        elif tetra.shape[1] == 3:
            tetra = np.transpose(tetra, (2,1,0))
        else:
            tetra = np.transpose(tetra, (2,0,1))
    elif tetra.shape[1] != 3:
        tetra = np.transpose(tetra, (0,2,1))
    if pt.shape[0] != 3: pt = pt.T
    # Okay, calculate the determinants...
    d_ = det_4x3(tetra[0], tetra[1], tetra[2], tetra[3])
    d0 = det_4x3(pt,       tetra[1], tetra[2], tetra[3])
    d1 = det_4x3(tetra[0], pt,       tetra[2], tetra[3])
    d2 = det_4x3(tetra[0], tetra[1], pt,       tetra[3])
    d3 = det_4x3(tetra[0], tetra[1], tetra[2], pt)
    s_ = np.sign(d_)
    z_ = np.logical_or(np.any([s_ * si == -1 for si in np.sign([d0,d1,d2,d3])], axis=0),
                       np.isclose(d_,0))
    x_ = np.logical_not(z_)
    d_inv = x_ / (x_ * d_ + z_)
    return np.asarray([d_inv * dq for dq in (d0,d1,d2,d3)]) 

def point_in_tetrahedron(tetra, pt):
    '''
    point_in_tetrahedron(tetrahedron, point) yields True if the given point is in the given 
      tetrahedron. If either tetrahedron or point (or both) are lists of shapes/points, then this
      calculation is automatically threaded over all the given arguments.
    '''
    bcs = tetrahedral_barycentric_coordinates(tetra, pt)
    return np.logical_not(np.all(np.isclose(bcs, 0), axis=0)) 

def _retinotopy_vectors_to_float(ang, ecc, wgt, weight_min=0):
    ok = np.isfinite(wgt) & np.isfinite(ecc) & np.isfinite(ang)
    ok[ok] &= wgt[ok] > weight_min
    bad = np.logical_not(ok)
    if np.sum(bad) > 0:
        wgt = np.array(wgt)
        wgt[bad] = 0
    return (ang, ecc, wgt) 

def zinv(x, null=0):
    '''
    zinv(x) yields 1/x if x is not close to 0 and 0 otherwise. Automatically threads over arrays and
      supports sparse-arrays.

    The optional argument null (default: 0) may be given to specify that zeros in the arary x should
    instead be replaced with the given value. Note that if this value is not equal to 0, then any
    sparse array passed to zinv must be reified.

    The zinv function never raises an error due to divide-by-zero; if you desire this behavior, use
    the inv function instead.
    '''
    if sps.issparse(x):
        if null != 0: return zinv(x.toarray(), null=null)
        x = x.copy()
        x.data = zinv(x.data)
        try: x.eliminate_zeros()
        except Exception: pass
        return x
    else:
        x = np.asarray(x)
        z = np.isclose(x, 0)
        r = np.logical_not(z) / (x + z)
        if null == 0: return r
        r[z] = null
        return r 

def czdivide(a, b, null=0):
    '''
    czdivide(a, b) returns the quotient a / b as a numpy array object. Like numpy's divide function
      or a/b syntax, czdivide will thread over the latest dimension possible. Unlike numpy's divide,
      czdivide works with sparse matrices. Additionally, czdivide multiplies a by the zinv of b, so
      divide-by-zero entries are replaced with 0 in the result.

    The optional argument null (default: 0) may be given to specify that zeros in the arary b should
    instead be replaced with the given value in the result. Note that if this value is not equal to
    0, then any sparse array passed as argument b must be reified.

    The czdivide function never raises an error due to divide-by-zero; if you desire this behavior,
    use the cdivide function instead.
    '''
    if null == 0:         return a.multiply(zinv(b)) if sps.issparse(a) else a * zinv(b)
    elif sps.issparse(b): b = b.toarray()
    else:                 b = np.asarray(b)
    z = np.isclose(b, 0)
    q = np.logical_not(z)
    zi = q / (b + z)
    if sps.issparse(a):
        r = a.multiply(zi).tocsr()
    else:
        r = np.asarray(a) * zi
    r[np.ones(a.shape, dtype=np.bool)*z] = null
    return r 

def test_predicted_scores_are_within_range(self):
    ocr_model = self.create_model()

    _, _, scores = ocr_model.char_predictions(self.fake_logits)
    with self.test_session() as sess:
      scores_np = sess.run(scores)

    values_in_range = (scores_np >= 0.0) & (scores_np <= 1.0)
    self.assertTrue(
        np.all(values_in_range),
        msg=('Scores contains out of the range values %s' %
             scores_np[np.logical_not(values_in_range)])) 

def bin_points(XYZ_cms, map_size, z_bins, xy_resolution):
  """Bins points into xy-z bins
  XYZ_cms is ... x H x W x3
  Outputs is ... x map_size x map_size x (len(z_bins)+1)
  """
  sh = XYZ_cms.shape
  XYZ_cms = XYZ_cms.reshape([-1, sh[-3], sh[-2], sh[-1]])
  n_z_bins = len(z_bins)+1
  map_center = (map_size-1.)/2.
  counts = []
  isvalids = []
  for XYZ_cm in XYZ_cms:
    isnotnan = np.logical_not(np.isnan(XYZ_cm[:,:,0]))
    X_bin = np.round(XYZ_cm[:,:,0] / xy_resolution + map_center).astype(np.int32)
    Y_bin = np.round(XYZ_cm[:,:,1] / xy_resolution + map_center).astype(np.int32)
    Z_bin = np.digitize(XYZ_cm[:,:,2], bins=z_bins).astype(np.int32)

    isvalid = np.array([X_bin >= 0, X_bin < map_size, Y_bin >= 0, Y_bin < map_size,
                        Z_bin >= 0, Z_bin < n_z_bins, isnotnan])
    isvalid = np.all(isvalid, axis=0)

    ind = (Y_bin * map_size + X_bin) * n_z_bins + Z_bin
    ind[np.logical_not(isvalid)] = 0
    count = np.bincount(ind.ravel(), isvalid.ravel().astype(np.int32),
                         minlength=map_size*map_size*n_z_bins)
    count = np.reshape(count, [map_size, map_size, n_z_bins])
    counts.append(count)
    isvalids.append(isvalid)
  counts = np.array(counts).reshape(list(sh[:-3]) + [map_size, map_size, n_z_bins])
  isvalids = np.array(isvalids).reshape(list(sh[:-3]) + [sh[-3], sh[-2], 1])
  return counts, isvalids