Python nd.mean方法代码示例

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def evaluate_accuracy_multi(data_iterator, net, ctx):
    data_iterator.reset()
    acc = 0
    dummy_label = np.zeros((0,6))
    dummy_pred = np.zeros((0,6))
    t1 = time.time()
    for i, batch in enumerate(data_iterator):
        data, label = _get_batch_multi(batch, ctx, False)
        # acc += np.mean([accuracy(net(X), Y) for X, Y in zip(data, label)])
        # acc += np.mean([roc_auc_score(Y.asnumpy(), net(X).asnumpy()) for X, Y in zip(data, label)])
        output = np.vstack((net(X).asnumpy() for X in data))
        labels = np.vstack((Y.asnumpy() for Y in label))
        dummy_label = np.vstack((dummy_label, labels)) 
        dummy_pred = np.vstack((dummy_pred, output))
    # return acc / (i+1)
    # print dummy_label.shape, dummy_pred.shape
    dummy_pred_label = dummy_pred > 0.5
    for i in range(dummy_label.shape[1]):
        print i, confusion_matrix(dummy_label[:,i], dummy_pred_label[:,i])

    return roc_auc_score(dummy_label, dummy_pred), accuracy(dummy_pred, dummy_label), time.time() - t1 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def train(net,epochs, ctx, train_data,test_data,
            margin_loss, reconstructions_loss, 
            batch_size,scale_factor):
    num_classes = 10
    trainer = gluon.Trainer(
        net.collect_params(),'sgd', {'learning_rate': 0.05, 'wd': 5e-4})

    for epoch in range(epochs):
        train_loss = 0.0
        for batch_idx, (data, label) in tqdm(enumerate(train_data), total=len(train_data), ncols=70, leave=False, unit='b'):
            label = label.as_in_context(ctx)
            data = data.as_in_context(ctx)
            with autograd.record():
                prob, X_l2norm, reconstructions = net(data, label)
                loss1 = margin_loss(data, num_classes,  label, X_l2norm)
                loss2 = reconstructions_loss(reconstructions, data)
                loss = loss1 + scale_factor * loss2
                loss.backward()
            trainer.step(batch_size)
            train_loss += nd.mean(loss).asscalar()
        test_acc = test(test_data, net, ctx)
        print('Epoch:{}, TrainLoss:{:.5f}, TestAcc:{}'.format(epoch,train_loss / len(train_data),test_acc)) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
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!" 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def _evaluate_accuracy(self, X, Y, batch_size=64):
        data_loader = self.generate_batch(X, Y, batch_size, shuffled=False)

        softmax_loss = gluon.loss.SoftmaxCrossEntropyLoss()

        num_batches = len(X) // batch_size

        metric = mx.metric.Accuracy()
        loss_avg = 0.
        for i, (data, label) in enumerate(data_loader):
            data = data.as_in_context(self.model_ctx)
            label = label.as_in_context(self.model_ctx)
            output = self.model(data)
            predictions = nd.argmax(output, axis=1)
            loss = softmax_loss(output, label)
            metric.update(preds=predictions, labels=label)
            loss_avg = loss_avg * i / (i + 1) + nd.mean(loss).asscalar() / (i + 1)

            if i + 1 == num_batches:
                break
        return metric.get()[1], loss_avg 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def facc(label, pred):
    """ evaluate accuracy """
    pred = pred.ravel()
    label = label.ravel()
    return ((pred > 0.5) == label).mean()


# setting 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def evaluate(data):
    acc_test = mx.metric.Accuracy()
    test_loss = 0.0
    cnt = 0
    for epoch_percent, batch_slots in batch_iter(data,batch_size,shuffle=False):
        batch_sequence, batch_label = zip(*batch_slots)
        batch_sequence = nd.array(batch_sequence,ctx)
        batch_label = nd.array(batch_label,ctx)
        output = net(batch_sequence)
        loss = SCE(output,batch_label)
        acc_test.update(preds=[output],labels=[batch_label])
        test_loss += nd.mean(loss).asscalar()
        cnt = cnt+1
    return acc_test.get()[1],test_loss/cnt 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def accuracy(output, label):
    return nd.mean(output.argmax(axis=1) == label).asscalar() 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def accuracy(output, label):
    L = -label*np.log2(output) - (1-label) * np.log2(1-output)
    return np.mean(L) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def train(train_data, test_data, net, loss, trainer,
          ctx, num_epochs, print_batches=None):
    """Train a network"""
    min_loss = 100000
    for epoch in range(num_epochs):
        train_loss = 0.
        train_acc = 0.
        n = 0
        for i, batch in enumerate(train_data):
            data, label = _get_batch(batch, ctx)
            with autograd.record():
                output = net(data)
                L = loss(output, label)
                L.backward()
            trainer.step(data.shape[0], ignore_stale_grad=True)
            train_loss += nd.mean(L).asscalar()
            train_acc += accuracy(output, label)
            n = i + 1
            if print_batches and n % print_batches == 0:
                test_acc = evaluate_accuracy(test_data, net, ctx)
                test_data.reset()
                print("Batch %d. Loss: %f, Train acc %f, Test Loss %f" % (
                n, train_loss/n, train_acc/n, test_acc))
                if test_acc < min_loss:
                    min_loss = test_acc
                    net.save_params('net.params')
        test_acc = evaluate_accuracy(test_data, net, ctx)
        train_data.reset()
        test_data.reset()
        print("Epoch %d. Loss: %f, Train acc %f, Test Loss %f" % (
              epoch, train_loss/n, train_acc/n, test_acc))
        if test_acc < min_loss:
            min_loss = test_acc
            net.save_params('net.params') 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def train_multi(train_data, test_data, iteration, net, loss, trainer,
          ctx, num_epochs, print_batches=None, pos_tr_ratio=None):
    """Train a network"""
    min_loss = 0
    for epoch in range(num_epochs):
        train_loss = 0.
        train_acc = 0.
        n = 0
        for i, batch in enumerate(train_data):
            data, label = _get_batch_multi(batch, ctx)
            with autograd.record():
                losses = [loss(net(X), Y, pos_tr_ratio) for X, Y in zip(data, label)]
                for l in losses:
                    l.backward()
            trainer.step(batch.data[0].shape[0], ignore_stale_grad=True)
            train_loss += np.mean([nd.mean(l).asscalar() for l in losses])
            # train_acc += accuracy(output, label)
            n = i + 1
            if print_batches and n % print_batches == 0:
                test_acc, test_loss, eval_time = evaluate_accuracy_multi(test_data, net, ctx)
                print("Batch %d. Loss: %f, Test roc_auc: %f, test_loss: %f , eval time: %f" % (
                n, train_loss/n, test_acc, test_loss, eval_time))
                if test_acc > min_loss:
                    min_loss = test_acc
                    net.save_params('net'+str(iteration)+'.params')
          
        train_data.reset()
        test_acc, test_loss, eval_time = evaluate_accuracy_multi(test_data, net, ctx)
        print("Epoch %d. Loss: %f, roc_auc: %f, test_loss: %f , eval time: %f" % (
              epoch, train_loss/n, test_acc, test_loss, eval_time))
        if test_acc > min_loss:
            min_loss = test_acc
            net.save_params('net'+str(iteration)+'.params') 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def CapLoss(y_pred, y_true):
    L = y_true * nd.square(nd.maximum(0., 0.9 - y_pred)) + \
        0.5 * (1 - y_true) * nd.square(nd.maximum(0., y_pred - 0.1))
    return nd.mean(nd.sum(L, 1)) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def EntropyLoss1(y_pred, y_true, train_pos_ratio):
    scale = 10
    train_pos_ratio = array(train_pos_ratio, ctx=y_pred.context, dtype=np.float32) * scale
    train_neg_ratio = (scale - train_pos_ratio)
    L = - y_true*nd.log2(y_pred) * train_neg_ratio - (1-y_true) * nd.log2(1-y_pred)*train_pos_ratio 
    return nd.mean(L) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def EntropyLoss(y_pred, y_true):
    L = - y_true*nd.log2(y_pred) - (1-y_true) * nd.log2(1-y_pred)
    return nd.mean(L) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def EntropyLoss1(y_pred, y_true):
    train_pos_ratio = array([ 0.09584448, 0.00999555, 0.05294822, 0.00299553, 0.04936361, 0.00880486], ctx=y_pred.context, dtype=np.float32)*10
    train_neg_ratio = (1.0-train_pos_ratio)*10
    L = - y_true*nd.log2(y_pred) * train_neg_ratio - (1-y_true) * nd.log2(1-y_pred) * train_pos_ratio
    return nd.mean(L) 

# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import mean [as 别名]
def meta_knowledge(self, feature):
        return self.geo_encoder(nd.mean(feature, axis=0))