Python torch.sort() Examples

def forward(self, x, x_len, atten_mask):
        CoAtt = torch.bmm(x, x.transpose(1, 2))
        CoAtt = atten_mask.unsqueeze(1) * CoAtt - (1 - atten_mask).unsqueeze(1) * INF
        CoAtt = torch.softmax(CoAtt, dim=-1)
        new_x = torch.cat([torch.bmm(CoAtt, x), x], -1)

        sorted_x_len, indx = torch.sort(x_len, 0, descending=True)
        new_x = pack_padded_sequence(new_x[indx], sorted_x_len.data.tolist(), batch_first=True)

        h0 = to_cuda(torch.zeros(2, x_len.size(0), self.hidden_size // 2), self.use_cuda)
        c0 = to_cuda(torch.zeros(2, x_len.size(0), self.hidden_size // 2), self.use_cuda)
        packed_h, (packed_h_t, _) = self.model(new_x, (h0, c0))

        # restore the sorting
        _, inverse_indx = torch.sort(indx, 0)
        packed_h_t = torch.cat([packed_h_t[i] for i in range(packed_h_t.size(0))], -1)
        restore_packed_h_t = packed_h_t[inverse_indx]
        output = restore_packed_h_t
        return output 

def lovasz_softmax_flat(probas, labels, only_present=False):
    """
    Multi-class Lovasz-Softmax loss
      probas: [P, C] Variable, class probabilities at each prediction (between 0 and 1)
      labels: [P] Tensor, ground truth labels (between 0 and C - 1)
      only_present: average only on classes present in ground truth
    """
    C = probas.size(1)
    losses = []
    for c in range(C):
        fg = (labels == c).float() # foreground for class c
        if only_present and fg.sum() == 0:
            continue
        errors = (Variable(fg) - probas[:, c]).abs()
        errors_sorted, perm = torch.sort(errors, 0, descending=True)
        perm = perm.data
        fg_sorted = fg[perm]
        losses.append(torch.dot(errors_sorted, Variable(lovasz_grad(fg_sorted))))
    return mean(losses) 

def _threshold_and_support(input, dim=0):
    """
    Sparsemax building block: compute the threshold
    Parameters:
        input: any dimension
        dim: dimension along which to apply the sparsemax
    Returns:
        the threshold value
    """
    input_srt, _ = torch.sort(input, descending=True, dim=dim)
    input_cumsum = input_srt.cumsum(dim) - 1
    rhos = _make_ix_like(input, dim)
    support = rhos * input_srt > input_cumsum

    support_size = support.sum(dim=dim).unsqueeze(dim)
    tau = input_cumsum.gather(dim, support_size - 1)
    tau /= support_size.to(input.dtype)
    return tau, support_size 

def average_precision(output, target, difficult_examples=True):

        # sort examples
        sorted, indices = torch.sort(output, dim=0, descending=True)

        # Computes prec@i
        pos_count = 0.
        total_count = 0.
        precision_at_i = 0.
        for i in indices:
            label = target[i]
            if difficult_examples and label == 0:
                continue
            if label == 1:
                pos_count += 1
            total_count += 1
            if label == 1:
                precision_at_i += pos_count / total_count
        precision_at_i /= pos_count
        return precision_at_i 

def forward(self, inputs, targets):

        N, C, H, W = inputs.size()
        masks = torch.zeros(N, C, H, W).to(targets.device).scatter_(1, targets.view(N, 1, H, W), 1)

        loss = 0.

        for mask, input in zip(masks.view(N, -1), inputs.view(N, -1)):

            max_margin_errors = 1. - ((mask * 2 - 1) * input)
            errors_sorted, indices = torch.sort(max_margin_errors, descending=True)
            labels_sorted = mask[indices.data]

            inter = labels_sorted.sum() - labels_sorted.cumsum(0)
            union = labels_sorted.sum() + (1. - labels_sorted).cumsum(0)
            iou = 1. - inter / union

            p = len(labels_sorted)
            if p > 1:
                iou[1:p] = iou[1:p] - iou[0:-1]

            loss += torch.dot(nn.functional.relu(errors_sorted), iou)

        return loss / N 

def test_websocket_worker_multiple_output_response(hook, start_remote_worker):
    """Evaluates that you can do basic tensor operations using
    WebsocketServerWorker."""
    server, remote_proxy = start_remote_worker(id="socket_multiple_output", hook=hook, port=8771)

    x = torch.tensor([1.0, 3, 2])
    x = x.send(remote_proxy)

    p1, p2 = torch.sort(x)
    x1, x2 = p1.get(), p2.get()

    assert (x1 == torch.tensor([1.0, 2, 3])).all()
    assert (x2 == torch.tensor([0, 2, 1])).all()

    x.get()  # retrieve remote object before closing the websocket connection

    remote_proxy.close()
    server.terminate() 

def cxcy_to_gcxgcy(cxcy, priors_cxcy):
    """
    Encode bounding boxes (that are in center-size form) w.r.t. the corresponding prior boxes (that are in center-size form).

    For the center coordinates, find the offset with respect to the prior box, and scale by the size of the prior box.
    For the size coordinates, scale by the size of the prior box, and convert to the log-space.

    In the model, we are predicting bounding box coordinates in this encoded form.

    :param cxcy: bounding boxes in center-size coordinates, a tensor of size (n_priors, 4)
    :param priors_cxcy: prior boxes with respect to which the encoding must be performed, a tensor of size (n_priors, 4)
    :return: encoded bounding boxes, a tensor of size (n_priors, 4)
    """

    # The 10 and 5 below are referred to as 'variances' in the original Caffe repo, completely empirical
    # They are for some sort of numerical conditioning, for 'scaling the localization gradient'
    # See https://github.com/weiliu89/caffe/issues/155
    return torch.cat([(cxcy[:, :2] - priors_cxcy[:, :2]) / (priors_cxcy[:, 2:] / 10),  # g_c_x, g_c_y
                      torch.log(cxcy[:, 2:] / priors_cxcy[:, 2:]) * 5], 1)  # g_w, g_h 

def value(self):
        """Returns the model's average precision for each class
        Return:
            ap (FloatTensor): 1xK tensor, with avg precision for each class k
        """

        if self.scores.numel() == 0:
            return 0
        ap = torch.zeros(self.scores.size(1))
        rg = torch.arange(1, self.scores.size(0)).float()

        # compute average precision for each class
        for k in range(self.scores.size(1)):
            # sort scores
            scores = self.scores[:, k]
            targets = self.targets[:, k]

            # compute average precision
            ap[k] = AveragePrecisionMeter.average_precision(scores, targets, self.difficult_examples)
        return ap 

def nms(boxes, nms_thresh):
    if len(boxes) == 0:
        return boxes

    det_confs = torch.zeros(len(boxes))
    for i in range(len(boxes)):
        det_confs[i] = 1-boxes[i][4]

    _,sortIds = torch.sort(det_confs)
    out_boxes = []
    for i in range(len(boxes)):
        box_i = boxes[sortIds[i]]
        if box_i[4] > 0:
            out_boxes.append(box_i)
            for j in range(i+1, len(boxes)):
                box_j = boxes[sortIds[j]]
                if bbox_iou(box_i, box_j, x1y1x2y2=False) > nms_thresh:
                    #print(box_i, box_j, bbox_iou(box_i, box_j, x1y1x2y2=False))
                    box_j[4] = 0
    return out_boxes 

def apply_packed_sequence(rnn, embedding, lengths):
    """ Runs a forward pass of embeddings through an rnn using packed sequence.
    Args:
       rnn: The RNN that that we want to compute a forward pass with.
       embedding (FloatTensor b x seq x dim): A batch of sequence embeddings.
       lengths (LongTensor batch): The length of each sequence in the batch.

    Returns:
       output: The output of the RNN `rnn` with input `embedding`
    """
    # Sort Batch by sequence length
    lengths_sorted, permutation = torch.sort(lengths, descending=True)
    embedding_sorted = embedding[permutation]

    # Use Packed Sequence
    embedding_packed = pack(embedding_sorted, lengths_sorted, batch_first=True)
    outputs_packed, (hidden, cell) = rnn(embedding_packed)
    outputs_sorted, _ = unpack(outputs_packed, batch_first=True)
    # Restore original order
    _, permutation_rev = torch.sort(permutation, descending=False)
    outputs = outputs_sorted[permutation_rev]
    hidden, cell = hidden[:, permutation_rev], cell[:, permutation_rev]
    return outputs, (hidden, cell) 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.relu(errors_sorted), Variable(grad))
    return loss 

def lovasz_hinge_flat(self, logits, labels):
        """
        Binary Lovasz hinge loss
          logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
          labels: [P] Tensor, binary ground truth labels (0 or 1)
          ignore: label to ignore
        """
        if len(labels) == 0:
            # only void pixels, the gradients should be 0
            return logits.sum() * 0.
        signs = 2. * labels.float() - 1.
        errors = (1. - logits * Variable(signs))
        errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
        perm = perm.data
        gt_sorted = labels[perm]
        grad = lovasz_grad(gt_sorted)
        loss = torch.dot(F.relu(errors_sorted), Variable(grad))
        return loss 

def scores_to_ranks(scores: torch.Tensor):
    """Convert model output scores into ranks."""
    batch_size, num_rounds, num_options = scores.size()
    scores = scores.view(-1, num_options)

    # sort in descending order - largest score gets highest rank
    sorted_ranks, ranked_idx = scores.sort(1, descending=True)

    # i-th position in ranked_idx specifies which score shall take this
    # position but we want i-th position to have rank of score at that
    # position, do this conversion
    ranks = ranked_idx.clone().fill_(0)
    for i in range(ranked_idx.size(0)):
        for j in range(num_options):
            ranks[i][ranked_idx[i][j]] = j
    # convert from 0-99 ranks to 1-100 ranks
    ranks += 1
    ranks = ranks.view(batch_size, num_rounds, num_options)
    return ranks 

def __init__(self, dim=-1, k=None):
        """sparsemax: normalizing sparse transform (a la softmax).

        Solves the projection:

            min_p ||x - p||_2   s.t.    p >= 0, sum(p) == 1.

        Parameters
        ----------
        dim : int
            The dimension along which to apply sparsemax.

        k : int or None
            number of largest elements to partial-sort over. For optimal
            performance, should be slightly bigger than the expected number of
            nonzeros in the solution. If the solution is more than k-sparse,
            this function is recursively called with a 2*k schedule.
            If `None`, full sorting is performed from the beginning.
        """
        self.dim = dim
        self.k = k
        super(Sparsemax, self).__init__() 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.relu(errors_sorted), Variable(grad))
    return loss 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.elu(errors_sorted) + 1, Variable(grad))
    return loss 

def lovasz_hinge_flat(self, logits, labels):
        """
        Binary Lovasz hinge loss
          logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
          labels: [P] Tensor, binary ground truth labels (0 or 1)
          ignore: label to ignore
        """
        if len(labels) == 0:
            # only void pixels, the gradients should be 0
            return logits.sum() * 0.
        signs = 2. * labels.float() - 1.
        errors = (1. - logits * Variable(signs))
        errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
        perm = perm.data
        gt_sorted = labels[perm]
        grad = lovasz_grad(gt_sorted)
        loss = torch.dot(F.relu(errors_sorted), Variable(grad))
        return loss 

def perturb_s_and_get_filtered_rank(embedding, w, s, r, o, test_size, triplets_to_filter):
    """ Perturb subject in the triplets
    """
    num_entities = embedding.shape[0]
    ranks = []
    for idx in range(test_size):
        if idx % 100 == 0:
            print("test triplet {} / {}".format(idx, test_size))
        target_s = s[idx]
        target_r = r[idx]
        target_o = o[idx]
        filtered_s = filter_s(triplets_to_filter, target_s, target_r, target_o, num_entities)
        target_s_idx = int((filtered_s == target_s).nonzero())
        emb_s = embedding[filtered_s]
        emb_r = w[target_r]
        emb_o = embedding[target_o]
        emb_triplet = emb_s * emb_r * emb_o
        scores = torch.sigmoid(torch.sum(emb_triplet, dim=1))
        _, indices = torch.sort(scores, descending=True)
        rank = int((indices == target_s_idx).nonzero())
        ranks.append(rank)
    return torch.LongTensor(ranks) 

def perturb_o_and_get_filtered_rank(embedding, w, s, r, o, test_size, triplets_to_filter):
    """ Perturb object in the triplets
    """
    num_entities = embedding.shape[0]
    ranks = []
    for idx in range(test_size):
        if idx % 100 == 0:
            print("test triplet {} / {}".format(idx, test_size))
        target_s = s[idx]
        target_r = r[idx]
        target_o = o[idx]
        filtered_o = filter_o(triplets_to_filter, target_s, target_r, target_o, num_entities)
        target_o_idx = int((filtered_o == target_o).nonzero())
        emb_s = embedding[target_s]
        emb_r = w[target_r]
        emb_o = embedding[filtered_o]
        emb_triplet = emb_s * emb_r * emb_o
        scores = torch.sigmoid(torch.sum(emb_triplet, dim=1))
        _, indices = torch.sort(scores, descending=True)
        rank = int((indices == target_o_idx).nonzero())
        ranks.append(rank)
    return torch.LongTensor(ranks) 

def __calculate_eps__(self, matrix, num_nodes, avg_degree):
        r"""Calculates threshold necessary to achieve a given average degree.

        Args:
            matrix (Tensor): Adjacency matrix or edge weights.
            num_nodes (int): Number of nodes.
            avg_degree (int): Target average degree.

        :rtype: (:class:`float`)
        """
        sorted_edges = torch.sort(matrix.flatten(), descending=True).values
        if avg_degree * num_nodes > len(sorted_edges):
            return -np.inf

        left = sorted_edges[avg_degree * num_nodes - 1]
        right = sorted_edges[avg_degree * num_nodes]
        return (left + right) / 2.0 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * signs)

    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.elu(errors_sorted), grad)
    return loss 

def lovasz_softmax_flat(probas, labels, only_present=False):
    """
    Multi-class Lovasz-Softmax loss
      probas: [P, C] Variable, class probabilities at each prediction (between 0 and 1)
      labels: [P] Tensor, ground truth labels (between 0 and C - 1)
      only_present: average only on classes present in ground truth
    """
    C = probas.size(1)
    losses = []
    for c in range(C):
        fg = (labels == c).float()  # foreground for class c
        if only_present and fg.sum() == 0:
            continue

        errors = (fg - probas[:, c]).abs()
        errors_sorted, perm = torch.sort(errors, 0, descending=True)
        perm = perm.data
        fg_sorted = fg[perm]
        losses.append(torch.dot(errors_sorted, lovasz_grad(fg_sorted)))
    return mean(losses) 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.relu(errors_sorted), Variable(grad))
    return loss 

def lovasz_softmax_flat(probas, labels, only_present=False):
    """
    Multi-class Lovasz-Softmax loss
      probas: [P, C] Variable, class probabilities at each prediction (between 0 and 1)
      labels: [P] Tensor, ground truth labels (between 0 and C - 1)
      only_present: average only on classes present in ground truth
    """
    C = probas.size(1)
    losses = []
    for c in range(C):
        fg = (labels == c).float() # foreground for class c
        if only_present and fg.sum() == 0:
            continue

        errors = (fg - probas[:, c]).abs()
        errors_sorted, perm = torch.sort(errors, 0, descending=True)
        perm = perm.data
        fg_sorted = fg[perm]
        losses.append(torch.dot(errors_sorted, lovasz_grad(fg_sorted)))
    return mean(losses) 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.relu(errors_sorted), Variable(grad))
    return loss 

def lovasz_hinge_flat(logits, labels):
    """
    Binary Lovasz hinge loss
      logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
      labels: [P] Tensor, binary ground truth labels (0 or 1)
      ignore: label to ignore
    """
    if len(labels) == 0:
        # only void pixels, the gradients should be 0
        return logits.sum() * 0.
    signs = 2. * labels.float() - 1.
    errors = (1. - logits * Variable(signs))
    errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
    perm = perm.data
    gt_sorted = labels[perm]
    grad = lovasz_grad(gt_sorted)
    loss = torch.dot(F.elu(errors_sorted) + 1, Variable(grad))
    return loss 

def _keep_top_k(self, boxes, box_inds, end_at, top_k, thresh):
        '''Set dynamic class threshold based on average detections per class per frame
        (before nms) and keep boxes above this threshold.
        '''
        box_list = boxes[:end_at].tolist()
        box_list = [box for box in box_list if box is not None]
        if len(box_list)==0:
            return
        X = torch.cat(box_list, dim=0)
        if X.size(0) == 0:
            return

        scores,_ = torch.sort(X[:,4], descending=True)
        # set threshold for this class
        thresh = scores[min(scores.numel(), top_k)]
        for image_index in range(end_at):
            if boxes[image_index] is not None and boxes[image_index].size(0)>0:
                bbox = boxes[image_index]
                keep = torch.nonzero(bbox[:,4]>=thresh).view(-1)
                if keep.numel()==0:
                    continue
                boxes[image_index] = bbox[keep]
                box_inds[image_index] = box_inds[image_index][keep]
        return boxes, box_inds, thresh 

def select(self, t, candidate_logprob, **kwargs):
        selected_logprob, selected_idx = torch.sort(candidate_logprob.view(self.b_s, -1), -1, descending=True)
        selected_logprob, selected_idx = selected_logprob[:, :self.beam_size], selected_idx[:, :self.beam_size]
        return selected_idx, selected_logprob 

def sort_1d(input):
    return th.sort(input) 

def sort_and_rank(score, target):
    _, indices = torch.sort(score, dim=1, descending=True)
    indices = torch.nonzero(indices == target.view(-1, 1))
    indices = indices[:, 1].view(-1)
    return indices