Python nd.ones方法代碼示例

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def test_compute_quantile_loss() -> None:
    y_true = nd.ones(shape=(10, 10, 10))
    y_pred = nd.zeros(shape=(10, 10, 10, 2))

    quantiles = [0.5, 0.9]

    loss = QuantileLoss(quantiles)

    correct_qt_loss = [1.0, 1.8]

    for idx, q in enumerate(quantiles):
        assert (
            nd.mean(
                loss.compute_quantile_loss(
                    nd.ndarray, y_true, y_pred[:, :, :, idx], q
                )
            )
            - correct_qt_loss[idx]
            < 1e-5
        ), f"computing quantile loss at quantile {q} fails!" 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def forward(self, graph, feat):
        r"""Compute APPNP layer.

        Parameters
        ----------
        graph : DGLGraph
            The graph.
        feat : mx.NDArray
            The input feature of shape :math:`(N, *)` :math:`N` is the
            number of nodes, and :math:`*` could be of any shape.

        Returns
        -------
        mx.NDArray
            The output feature of shape :math:`(N, *)` where :math:`*`
            should be the same as input shape.
        """
        with graph.local_scope():
            norm = mx.nd.power(mx.nd.clip(
                graph.in_degrees().astype(feat.dtype), a_min=1, a_max=float("inf")), -0.5)
            shp = norm.shape + (1,) * (feat.ndim - 1)
            norm = norm.reshape(shp).as_in_context(feat.context)
            feat_0 = feat
            for _ in range(self._k):
                # normalization by src node
                feat = feat * norm
                graph.ndata['h'] = feat
                graph.edata['w'] = self.edge_drop(
                    nd.ones((graph.number_of_edges(), 1), ctx=feat.context))
                graph.update_all(fn.u_mul_e('h', 'w', 'm'),
                                 fn.sum('m', 'h'))
                feat = graph.ndata.pop('h')
                # normalization by dst node
                feat = feat * norm
                feat = (1 - self._alpha) * feat + self._alpha * feat_0
            return feat 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def validate(val_data, val_dataset, net, ctx):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]

    val_metric.reset()

    from tqdm import tqdm
    for batch in tqdm(val_data):
        data, scale, center, score, imgid = val_batch_fn(batch, ctx)

        outputs = [net(X) for X in data]
        if opt.flip_test:
            data_flip = [nd.flip(X, axis=3) for X in data]
            outputs_flip = [net(X) for X in data_flip]
            outputs_flipback = [flip_heatmap(o, val_dataset.joint_pairs, shift=True) for o in outputs_flip]
            outputs = [(o + o_flip)/2 for o, o_flip in zip(outputs, outputs_flipback)]

        if opt.dsnt:
            outputs = [net_dsnt(X)[0] for X in outputs]

        if len(outputs) > 1:
            outputs_stack = nd.concat(*[o.as_in_context(mx.cpu()) for o in outputs], dim=0)
        else:
            outputs_stack = outputs[0].as_in_context(mx.cpu())

        if opt.dsnt:
            preds = (outputs_stack - 0.5) * scale.expand_dims(axis=1) + center.expand_dims(axis=1)
            maxvals = nd.ones(preds.shape[0:2]+(1, ))
        else:
            preds, maxvals = get_final_preds(outputs_stack, center.asnumpy(), scale.asnumpy())
        val_metric.update(preds, maxvals, score, imgid)

    metric_name, metric_score = val_metric.get()
    print("Inference Completed! %s = %.4f" % (metric_name, metric_score))
    return 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def test_data_parallel():
    # test gluon.contrib.parallel.DataParallelModel
    net = nn.HybridSequential()
    with net.name_scope():
        net.add(nn.Conv2D(in_channels=1, channels=20, kernel_size=5))
        net.add(nn.Activation('relu'))
        net.add(nn.MaxPool2D(pool_size=2, strides=2))
        net.add(nn.Conv2D(in_channels=20, channels=50, kernel_size=5))
        net.add(nn.Activation('relu'))
        net.add(nn.MaxPool2D(pool_size=2, strides=2))
        # The Flatten layer collapses all axis, except the first one, into one axis.
        net.add(nn.Flatten())
        net.add(nn.Dense(512,in_units=800))
        net.add(nn.Activation('relu'))
        net.add(nn.Dense(10, in_units=512))

    net.collect_params().initialize()
    criterion = gluon.loss.SoftmaxCELoss(axis=1)

    def test_net_sync(net, criterion, sync, nDevices):
        ctx_list = [mx.cpu(0) for i in range(nDevices)]
        net = DataParallelModel(net, ctx_list, sync=sync)
        criterion = DataParallelCriterion(criterion, ctx_list, sync=sync)
        iters = 100
        # train mode
        for i in range(iters):
            x = mx.random.uniform(shape=(8, 1, 28, 28))
            t = nd.ones(shape=(8))
            with autograd.record():
                y = net(x)
                loss = criterion(y, t)
                autograd.backward(loss)
        # evaluation mode
        for i in range(iters):
            x = mx.random.uniform(shape=(8, 1, 28, 28))
            y = net(x)

    test_net_sync(net, criterion, True, 1)
    test_net_sync(net, criterion, True, 2)
    test_net_sync(net, criterion, False, 1)
    test_net_sync(net, criterion, False, 2) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def __init__(
        self,
        quantiles: List[float],
        quantile_weights: List[float] = None,
        weight=None,
        batch_axis=0,
        **kwargs,
    ) -> None:
        """
        Represents the quantile loss used to fit decoders that learn quantiles.

        Parameters
        ----------
        quantiles
            list of quantiles to compute loss over.

        quantile_weights
            weights of the quantiles.

        weight:
            weighting of the loss.

        batch_axis:
            indicates axis that represents the batch.
        """
        super().__init__(weight, batch_axis, **kwargs)

        self.quantiles = quantiles
        self.num_quantiles = len(quantiles)
        self.quantile_weights = (
            nd.ones(self.num_quantiles) / self.num_quantiles
            if not quantile_weights
            else quantile_weights
        )

    # noinspection PyMethodOverriding 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def test_global_norm():
    data = list()
    for i in range(1, 6):
        data.append(np.ones((i * 10, i * 10)) * i)
    gnorm = np.asscalar(np.sqrt(sum([np.sum(np.square(d)) for d in data])))
    assert np.isclose(gnorm, global_norm([nd.array(d) for d in data]).asscalar()) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def test_broadcast_like():
    x = nd.ones((1, 2)) * 10
    y = nd.ones((100, 100, 2)) * 20
    assert mx.test_utils.almost_equal(x.broadcast_like(y).asnumpy(), broadcast_like(nd, x, y).asnumpy()) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import ones [as 別名]
def test_data_parallel():
    # test gluon.contrib.parallel.DataParallelModel
    net = nn.HybridSequential()
    with net.name_scope():
        net.add(nn.Conv2D(in_channels=1, channels=5, kernel_size=5))
        net.add(nn.Activation('relu'))
        net.add(nn.MaxPool2D(pool_size=2, strides=2))
        net.add(nn.Conv2D(in_channels=5, channels=5, kernel_size=5))
        net.add(nn.Activation('relu'))
        net.add(nn.MaxPool2D(pool_size=2, strides=2))
        # The Flatten layer collapses all axis, except the first one, into one axis.
        net.add(nn.Flatten())
        net.add(nn.Dense(8,in_units=80))
        net.add(nn.Activation('relu'))
        net.add(nn.Dense(10, in_units=8))

    net.collect_params().initialize()
    criterion = gluon.loss.SoftmaxCELoss(axis=1)

    def test_net_sync(net, criterion, sync, nDevices):
        ctx_list = [mx.cpu(0) for i in range(nDevices)]
        net = DataParallelModel(net, ctx_list, sync=sync)
        criterion = DataParallelCriterion(criterion, ctx_list, sync=sync)
        iters = 10
        bs = 2
        # train mode
        for i in range(iters):
            x = mx.random.uniform(shape=(bs, 1, 28, 28))
            t = nd.ones(shape=(bs))
            with autograd.record():
                y = net(x)
                loss = criterion(y, t)
                autograd.backward(loss)
        # evaluation mode
        for i in range(iters):
            x = mx.random.uniform(shape=(bs, 1, 28, 28))
            y = net(x)
        nd.waitall()

    # test_net_sync(net, criterion, True, 1)
    test_net_sync(net, criterion, True, 2)
    # test_net_sync(net, criterion, False, 1)
    test_net_sync(net, criterion, False, 2)