Python torch.bincount方法代码示例

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def _torch_hist(label_true, label_pred, n_class):
    """Calculates the confusion matrix for the labels
    
    Args:
        label_true ([type]): [description]
        label_pred ([type]): [description]
        n_class ([type]): [description]
    
    Returns:
        [type]: [description]
    """
    
    assert len(label_true.shape) == 1, "Labels need to be 1D"
    assert len(label_pred.shape) == 1, "Predictions need to be 1D"
    mask = (label_true >= 0) & (label_true < n_class)
    hist = torch.bincount(n_class * label_true[mask] + label_pred[mask], minlength=n_class ** 2).reshape(
        n_class, n_class
    )
    return hist 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def split(data, batch):
    node_slice = torch.cumsum(torch.from_numpy(np.bincount(batch)), 0)
    node_slice = torch.cat([torch.tensor([0]), node_slice])

    row, _ = data.edge_index
    edge_slice = torch.cumsum(torch.from_numpy(np.bincount(batch[row])), 0)
    edge_slice = torch.cat([torch.tensor([0]), edge_slice])

    # Edge indices should start at zero for every graph.
    data.edge_index -= node_slice[batch[row]].unsqueeze(0)
    data.__num_nodes__ = torch.bincount(batch).tolist()

    slices = {'edge_index': edge_slice}
    if data.x is not None:
        slices['x'] = node_slice
    if data.edge_attr is not None:
        slices['edge_attr'] = edge_slice
    if data.y is not None:
        if data.y.size(0) == batch.size(0):
            slices['y'] = node_slice
        else:
            slices['y'] = torch.arange(0, batch[-1] + 2, dtype=torch.long)

    return data, slices 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def __init__(self, centroids, assignments, bias, in_features, out_features):
        super(PQLinear, self).__init__()
        self.block_size = centroids.size(1)
        self.n_centroids = centroids.size(0)
        self.in_features = in_features
        self.out_features = out_features
        # check compatibility
        if self.in_features % self.block_size != 0:
            raise ValueError("Wrong PQ sizes")
        if len(assignments) % self.out_features != 0:
            raise ValueError("Wrong PQ sizes")
        # define parameters
        self.centroids = nn.Parameter(centroids, requires_grad=True)
        self.register_buffer("assignments", assignments)
        self.register_buffer("counts", torch.bincount(assignments).type_as(centroids))
        if bias is not None:
            self.bias = nn.Parameter(bias)
        else:
            self.register_parameter("bias", None) 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def test_main():
    '''
    Test the unified segmenter.
    '''
    from PIL import Image
    testim = Image.open('script/testdata/test_church_242.jpg')
    tensor_im = (torch.from_numpy(numpy.asarray(testim)).permute(2, 0, 1)
            .float() / 255 * 2 - 1)[None, :, :, :].cuda()
    segmenter = UnifiedParsingSegmenter()
    seg = segmenter.segment_batch(tensor_im)
    bc = torch.bincount(seg.view(-1))
    labels, cats = segmenter.get_label_and_category_names()
    for label in bc.nonzero()[:,0]:
        if label.item():
            # What is the prediction for this class?
            pred, mask = segmenter.predict_single_class(tensor_im, label.item())
            assert mask.sum().item() == bc[label].item()
            assert len(((seg == label).max(1)[0] - mask).nonzero()) == 0
            inside_pred = pred[mask].mean().item()
            outside_pred = pred[~mask].mean().item()
            print('%s (%s, #%d): %d pixels, pred %.2g inside %.2g outside' %
                (labels[label.item()] + (label.item(), bc[label].item(),
                    inside_pred, outside_pred))) 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def KMeans(x_i, c_j, Nits = 10, ranges = None):

    D = x_i.shape[1]
    for i in range(10):
        # Points -> Nearest cluster
        labs_i = nn_search(x_i, c_j, ranges = ranges)
        # Class cardinals:
        Ncl = torch.bincount(labs_i.view(-1)).type(dtype)
        # Compute the cluster centroids with torch.bincount:
        for d in range(D):  # Unfortunately, vector weights are not supported...
            c_j[:, d] = torch.bincount(labs_i.view(-1), weights=x_i[:, d]) / Ncl
    
    return c_j, labs_i


##############################################
# On the subject
# ~~~~~~~~~~~~~~~~~~~~~~~~
#
# For new subject (unlabelled), we perform a simple Kmean
# on R^60 to obtain a cluster of the data.
# 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def forward(self, simmat, dtoks, qtoks):
        # THIS IS SLOW ... Any way to make this faster? Maybe it's not worth doing on GPU?
        BATCH, CHANNELS, QLEN, DLEN = simmat.shape
        # +1e-5 to nudge scores of 1 to above threshold
        bins = ((simmat + 1.000001) / 2. * (self.bins - 1)).int()
        # set weights of 0 for padding (in both query and doc dims)
        weights = ((dtoks != -1).reshape(BATCH, 1, DLEN).expand(BATCH, QLEN, DLEN) * \
                  (qtoks != -1).reshape(BATCH, QLEN, 1).expand(BATCH, QLEN, DLEN)).float()

        # no way to batch this... loses gradients here. https://discuss.pytorch.org/t/histogram-function-in-pytorch/5350
        bins, weights = bins.cpu(), weights.cpu()
        histogram = []
        for superbins, w in zip(bins, weights):
            result = []
            for b in superbins:
                result.append(torch.stack([torch.bincount(q, x, self.bins) for q, x in zip(b, w)], dim=0))
            result = torch.stack(result, dim=0)
            histogram.append(result)
        histogram = torch.stack(histogram, dim=0)

        # back to GPU
        histogram = histogram.to(simmat.device)
        return (histogram.float() + 1e-5).log() 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def update(self, output: Sequence[torch.Tensor]) -> None:
        self._check_shape(output)
        y_pred, y = output

        self._num_examples += y_pred.shape[0]

        # target is (batch_size, ...)
        y_pred = torch.argmax(y_pred, dim=1).flatten()
        y = y.flatten()

        target_mask = (y >= 0) & (y < self.num_classes)
        y = y[target_mask]
        y_pred = y_pred[target_mask]

        indices = self.num_classes * y + y_pred
        m = torch.bincount(indices, minlength=self.num_classes ** 2).reshape(self.num_classes, self.num_classes)
        self.confusion_matrix += m.to(self.confusion_matrix) 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def forward(self, simmat, dlens, dtoks, qtoks):
        BATCH, CHANNELS, QLEN, DLEN = simmat.shape

        # +1e-5 to nudge scores of 1 to above threshold
        bins = ((simmat + 1.00001) / 2. * (self.bins - 1)).int()
        weights = ((dtoks != -1).reshape(BATCH, 1, DLEN).expand(BATCH, QLEN, DLEN) * \
                      (qtoks != -1).reshape(BATCH, QLEN, 1).expand(BATCH, QLEN, DLEN)).float()
        # apparently no way to batch this... https://discuss.pytorch.org/t/histogram-function-in-pytorch/5350
        bins, weights = bins.cpu(), weights.cpu() # WARNING: this line (and the similar line below) improve performance tenfold when on GPU
        histogram = []
        for superbins, w in zip(bins, weights):
            result = []
            for b in superbins:
                result.append(torch.stack([torch.bincount(q, x, self.bins) for q, x in zip(b, w)], dim=0))
            result = torch.stack(result, dim=0)
            histogram.append(result)
        histogram = torch.stack(histogram, dim=0)
        histogram = histogram.to(simmat.device) # WARNING: this line (and the similar line above) improve performance tenfold when on GPU
        return histogram 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def fast_hist(label_true, label_pred):
    n_class = settings.N_CLASSES
    mask = (label_true >= 0) & (label_true < n_class)
    hist = torch.bincount(
        n_class * label_true[mask].int() + label_pred[mask].int(),
        minlength=n_class ** 2,
    ).reshape(n_class, n_class)
    return hist 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def update(self, a, b):
        n = self.num_classes
        if self.mat is None:
            self.mat = torch.zeros((n, n), dtype=torch.int64, device=a.device)
        with torch.no_grad():
            k = (a >= 0) & (a < n)
            inds = n * a[k].to(torch.int64) + b[k]
            self.mat += torch.bincount(inds, minlength=n ** 2).reshape(n, n) 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def __init__(self, centroids, assignments, num_embeddings, embedding_dim,
                     padding_idx=None, max_norm=None, norm_type=2.,
                     scale_grad_by_freq=False, sparse=False, _weight=None):
        super(PQEmbedding, self).__init__()
        self.block_size = centroids.size(1)
        self.n_centroids = centroids.size(0)
        self.num_embeddings = num_embeddings
        self.embedding_dim = embedding_dim
        if padding_idx is not None:
            if padding_idx > 0:
                assert padding_idx < self.num_embeddings, 'Padding_idx must be within num_embeddings'
            elif padding_idx < 0:
                assert padding_idx >= -self.num_embeddings, 'Padding_idx must be within num_embeddings'
                padding_idx = self.num_embeddings + padding_idx
        self.padding_idx = padding_idx
        self.max_norm = max_norm
        self.norm_type = norm_type
        self.scale_grad_by_freq = scale_grad_by_freq
        self.sparse = sparse
        # check compatibility
        if self.embedding_dim % self.block_size != 0:
            raise ValueError("Wrong PQ sizes")
        if len(assignments) % self.num_embeddings != 0:
            raise ValueError("Wrong PQ sizes")
        # define parameters
        self.centroids = nn.Parameter(centroids, requires_grad=True)
        self.register_buffer("assignments", assignments)
        self.register_buffer("counts", torch.bincount(assignments).type_as(centroids)) 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def energy_spectrum(vel):
    """
    Compute energy spectrum given a velocity field
    :param vel: tensor of shape (N, 3, res, res, res)
    :return spec: tensor of shape(N, res/2)
    :return k: tensor of shape (res/2,), frequencies corresponding to spec
    """
    device = vel.device
    res = vel.shape[-2:]

    assert(res[0] == res[1])
    r = res[0]
    k_end = int(r/2)
    vel_ = pad_rfft3(vel, onesided=False) # (N, 3, res, res, res, 2)
    uu_ = (torch.norm(vel_, dim=-1) / r**3)**2
    e_ = torch.sum(uu_, dim=1)  # (N, res, res, res)
    k = fftfreqs(res).to(device) # (3, res, res, res)
    rad = torch.norm(k, dim=0) # (res, res, res)
    k_bin = torch.arange(k_end, device=device).float()+1
    bins = torch.zeros(k_end+1).to(device)
    bins[1:-1] = (k_bin[1:]+k_bin[:-1])/2
    bins[-1] = k_bin[-1]
    bins = bins.unsqueeze(0)
    bins[1:] += 1e-3
    inds = searchsorted(bins, rad.flatten().unsqueeze(0)).squeeze().int()
    # bincount = torch.histc(inds.cpu(), bins=bins.shape[1]+1).to(device)
    bincount = torch.bincount(inds)
    asort = torch.argsort(inds.squeeze())
    sorted_e_ = e_.view(e_.shape[0], -1)[:, asort]
    csum_e_ = torch.cumsum(sorted_e_, dim=1)
    binloc = torch.cumsum(bincount, dim=0).long()-1
    spec_ = csum_e_[:,binloc[1:]] - csum_e_[:,binloc[:-1]]
    spec_ = spec_[:, :-1]
    spec_ = spec_ * 2 * np.pi * (k_bin.float()**2) / bincount[1:-1].float()
    return spec_, k_bin


##################### COMPUTE STATS ########################### 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def forward(self, inputs: torch.Tensor) -> torch.Tensor:
        """
        # Parameters

        inputs : `torch.Tensor`
            Shape `(batch_size, timesteps, sequence_length)` of word ids
            representing the current batch.

        # Returns

        `torch.Tensor`
            The bag-of-words representations for the input sequence, shape
            `(batch_size, vocab_size)`
        """
        bag_of_words_vectors = []

        mask = get_text_field_mask({"tokens": {"tokens": inputs}})
        if self._ignore_oov:
            # also mask out positions corresponding to oov
            mask &= inputs != self._oov_idx
        for document, doc_mask in zip(inputs, mask):
            document = torch.masked_select(document, doc_mask)
            vec = torch.bincount(document, minlength=self.vocab_size).float()
            vec = vec.view(1, -1)
            bag_of_words_vectors.append(vec)
        bag_of_words_output = torch.cat(bag_of_words_vectors, 0)

        if self._projection:
            projection = self._projection
            bag_of_words_output = projection(bag_of_words_output)
        return bag_of_words_output 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def _fast_hist(true, pred, num_classes):
    mask = (true >= 0) & (true < num_classes)
    hist = torch.bincount(
        num_classes * true[mask] + pred[mask],
        minlength=num_classes ** 2,
    ).reshape(num_classes, num_classes).float()
    return hist 

# 需要导入模块: import torch [as 别名]
# 或者: from torch import bincount [as 别名]
def region_based_classification_single(self, sample, radius):
        """

        :param sample: one sample (1*channel*H*W)
        :param radius:
        :return:
        """
        self.model.eval()

        assert sample.shape[0] == 1, "the sample parameter should be one example in numpy format"
        copy_sample = np.copy(sample)

        with torch.no_grad():
            copy_sample = torch.from_numpy(copy_sample).to(self.device)

            # prepare the hypercube samples (size=num_points) for the sample (size=1)
            hypercube_samples = copy_sample.repeat(self.num_points, 1, 1, 1).to(self.device).float()
            random_space = torch.Tensor(*hypercube_samples.size()).to(self.device).float()
            random_space.uniform_(-radius, radius)
            hypercube_samples = torch.clamp(hypercube_samples + random_space, min=0.0, max=1.0)

            # predicting for hypercube samples
            hypercube_preds = self.model(hypercube_samples)
            hypercube_labels = torch.max(hypercube_preds, dim=1)[1]

            # voting for predicted labels
            bin_count = torch.bincount(hypercube_labels)
            rc_label = torch.max(bin_count, dim=0)[1]

            return rc_label.cpu().numpy()