Python mxnet.Context方法代碼示例

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def benchmarking(net, opt, ctx):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]

    bs = opt.batch_size
    num_iterations = opt.num_iterations
    input_shape = (bs, 3,) + tuple(input_size)
    size = num_iterations * bs
    data = mx.random.uniform(-1.0, 1.0, shape=input_shape, ctx=ctx[0], dtype='float32')
    dry_run = 5

    from tqdm import tqdm
    with tqdm(total=size + dry_run * bs) as pbar:
        for n in range(dry_run + num_iterations):
            if n == dry_run:
                tic = time.time()
            output = net(data)
            output.wait_to_read()
            pbar.update(bs)
    speed = size / (time.time() - tic)
    print('With batch size %d , %d batches, throughput is %f imgs/sec' % (bs, num_iterations, speed)) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def _append_arrs(arrs, use_shared_mem=False, expand=False, batch_axis=0):
    """Internal impl for returning appened arrays as list."""
    if isinstance(arrs[0], mx.nd.NDArray):
        if use_shared_mem:
            out = [x.as_in_context(mx.Context('cpu_shared', 0)) for x in arrs]
        else:
            out = arrs
    else:
        if use_shared_mem:
            out = [mx.nd.array(x, ctx=mx.Context('cpu_shared', 0)) for x in arrs]
        else:
            out = [mx.nd.array(x) for x in arrs]

    # add batch axis
    if expand:
        out = [x.expand_dims(axis=batch_axis) for x in out]
    return out 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def _stack_arrs(arrs, use_shared_mem=False):
        """
        Internal imple for stacking arrays.
        """
        if isinstance(arrs[0], mx.nd.NDArray):
            if use_shared_mem:
                out = mx.nd.empty((len(arrs),) + arrs[0].shape, dtype=arrs[0].dtype,
                                  ctx=mx.Context("cpu_shared", 0))
                return mx.nd.stack(*arrs, out=out)
            else:
                return mx.nd.stack(*arrs)
        else:
            out = np.asarray(arrs)
            if use_shared_mem:
                return mx.nd.array(out, ctx=mx.Context("cpu_shared", 0))
            else:
                return mx.nd.array(out) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def seed_rngs(seed: int, ctx: Optional[Union[mx.Context, List[mx.Context]]] = None) -> None:
    """
    Seed the random number generators (Python, Numpy and MXNet).

    :param seed: The random seed.
    :param ctx: Random number generators in MXNet are device specific.
           If None, MXNet will set the state of each generator of each device using seed and device id. This will lead
           to different results on different devices. If ctx is provided, this function will seed
           device-specific generators with a fixed offset. E.g. for 2 devices and seed=13, seed for gpu(0) will be 13,
           14 for gpu(1). See https://beta.mxnet.io/api/gluon-related/_autogen/mxnet.random.seed.html.
    """
    logger.info("Random seed: %d", seed)
    np.random.seed(seed)
    random.seed(seed)
    if ctx is None:
        mx.random.seed(seed, ctx='all')
    else:
        if isinstance(ctx, mx.Context):
            ctx = [ctx]
        for i, c in enumerate(ctx):
            mx.random.seed(seed + i, ctx=c) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def get_gpu_memory_usage(ctx: Union[mx.context.Context, List[mx.context.Context]]) -> Dict[int, Tuple[int, int]]:
    """
    Returns used and total memory for GPUs identified by the given context list.

    :param ctx: List of MXNet context devices.
    :return: Dictionary of device id mapping to a tuple of (memory used, memory total).
    """
    if isinstance(ctx, mx.context.Context):
        ctx = [ctx]
    ctx = [c for c in ctx if c.device_type == 'gpu']
    if not ctx:
        return {}

    memory_data = {}  # type: Dict[int, Tuple[int, int]]
    for c in ctx:
        try:
            free, total = mx.context.gpu_memory_info(device_id=c.device_id)  # in bytes
            used = total - free
            memory_data[c.device_id] = (used * 1e-06, total * 1e-06)
        except mx.MXNetError:
            logger.exception("Failed retrieving memory data for gpu%d", c.device_id)
            continue
    log_gpu_memory_usage(memory_data)
    return memory_data 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def train(train_iter, test_iter, net, loss, trainer, ctx, num_epochs):
    """Train and evaluate a model."""
    print('training on', ctx)
    if isinstance(ctx, mx.Context):
        ctx = [ctx]
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n, m, start = 0.0, 0.0, 0, 0, time.time()
        for i, batch in enumerate(train_iter):
            Xs, ys, batch_size = _get_batch(batch, ctx)   
            with autograd.record():
                y_hats = [net(X) for X in Xs]
                ls = [loss(y_hat, y) for y_hat, y in zip(y_hats, ys)]
            for l in ls:
                l.backward()
            trainer.step(batch_size)
            train_l_sum += sum([l.sum().asscalar() for l in ls])
            n += sum([l.size for l in ls])
            train_acc_sum += sum([(y_hat.argmax(axis=1) == y).sum().asscalar()
                                 for y_hat, y in zip(y_hats, ys)])
            m += sum([y.size for y in ys])
        test_acc = evaluate_accuracy(test_iter, net, ctx)
        print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f, '
              'time %.1f sec'
              % (epoch + 1, train_l_sum / n, train_acc_sum / m, test_acc,
                 time.time() - start)) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def batchify(
    data: List[dict],
    dtype: DType,
    multi_processing: bool,
    single_process_ctx: Optional[mx.Context] = None,
    variable_length: bool = False,
) -> DataBatch:
    """reduce the list of dictionaries to a single dictionary, where values
        referenced by identical key are reduced using the stack function"""
    return {
        key: stack(
            data=[item[key] for item in data],
            multi_processing=multi_processing,
            dtype=dtype,
            single_process_ctx=single_process_ctx,
            variable_length=variable_length,
        )
        for key in data[0].keys()
    } 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def _as_in_context(batch: dict, ctx: mx.Context) -> DataBatch:
    """Move data into new context, should only be in main process."""
    assert (
        not MPWorkerInfo.worker_process
    ), "This function is not meant to be used in workers."
    batch = {
        k: v.as_in_context(ctx) if isinstance(v, nd.NDArray)
        # Workaround due to MXNet not being able to handle NDArrays with 0 in shape properly:
        else (
            stack(v, False, v.dtype, ctx)
            if isinstance(v[0], np.ndarray) and 0 in v[0].shape
            else v
        )
        for k, v in batch.items()
    }
    return batch 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def __init__(
        self,
        dataset: Dataset,
        *,
        transform: Transformation,
        batch_size: int,
        ctx: mx.Context,
        num_workers: Optional[int] = None,
        num_prefetch: Optional[int] = None,
        dtype: DType = np.float32,
        **kwargs,
    ) -> None:
        super().__init__(
            dataset=dataset,
            transform=transform,
            is_train=True,
            batch_size=batch_size,
            ctx=ctx,
            dtype=dtype,
            cyclic=False,
            num_workers=num_workers,
            num_prefetch=num_prefetch,
            shuffle_buffer_length=None,
            **kwargs,
        ) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def initialize_from_array(
        self, input_array: np.ndarray, ctx: mx.Context = get_mxnet_context()
    ):
        r"""
        Initialize the representation based on a numpy array.

        Parameters
        ----------
        input_array
            Numpy array.
        ctx
            MXNet context.
        """
        pass

    # noinspection PyMethodOverriding 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def __init__(self, num_classes, data_shape, max_iter, dtype):
        self.batch_size = data_shape[0]
        self.cur_iter = 0
        self.max_iter = max_iter
        self.dtype = dtype
        label = np.random.randint(0, num_classes, [self.batch_size,])
        data = np.random.uniform(-1, 1, data_shape)
        self.data = mx.nd.array(data, dtype=self.dtype, ctx=mx.Context('cpu_pinned', 0))
        self.label = mx.nd.array(label, dtype=self.dtype, ctx=mx.Context('cpu_pinned', 0)) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def test_ndarray_copy():
    c = mx.nd.array(np.random.uniform(-10, 10, (10, 10)))
    d = c.copyto(mx.Context('cpu', 0))
    assert np.sum(np.abs(c.asnumpy() != d.asnumpy())) == 0.0 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def train(epoch, ctx):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]
    net.initialize(mx.init.Orthogonal(), ctx=ctx)
    # re-initialize conv4's weight to be Orthogonal
    net.conv4.initialize(mx.init.Orthogonal(scale=1), force_reinit=True, ctx=ctx)
    trainer = gluon.Trainer(net.collect_params(), 'adam', {'learning_rate': opt.lr})
    loss = gluon.loss.L2Loss()

    for i in range(epoch):
        train_data.reset()
        for batch in train_data:
            data = gluon.utils.split_and_load(batch.data[0], ctx_list=ctx, batch_axis=0)
            label = gluon.utils.split_and_load(batch.label[0], ctx_list=ctx, batch_axis=0)
            outputs = []
            with ag.record():
                for x, y in zip(data, label):
                    z = net(x)
                    L = loss(z, y)
                    L.backward()
                    outputs.append(z)
            trainer.step(batch.data[0].shape[0])
            metric.update(label, outputs)

        name, acc = metric.get()
        metric.reset()
        print('training mse at epoch %d: %s=%f'%(i, name, acc))
        test(ctx)

    net.save_parameters('superres.params') 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def test_ndarray_cpu_shared_ctx():
    ctx = mx.Context('cpu_shared', 0)
    res = mx.nd.zeros((1, 2, 3), ctx=ctx)
    assert(res.context == ctx) 

# 需要導入模塊: import mxnet [as 別名]
# 或者: from mxnet import Context [as 別名]
def test_aggregator():
    """aggregate value on muliple devices"""

    def check_aggregator(kv, key, key_list, stype):
        # devices
        num_devs = 4
        devs = [mx.Context('cpu', i) for i in range(num_devs)]

        # single
        vals = [mx.nd.ones(shape, d).tostype(stype) for d in devs]
        outs = [mx.nd.empty(shape, d) for d in devs]

        kv.push(key, vals)
        kv.pull(key, out=outs)

        for out in outs:
            check_diff_to_scalar(out, num_devs)

        # list
        vals = [[mx.nd.ones(shape, d).tostype(stype)*2.0 for d in devs]] * len(key_list)
        outs = [[mx.nd.empty(shape, d) for d in devs]] * len(key_list)
        kv.push(key_list, vals)
        kv.pull(key_list, out=outs)

        for out in outs:
            for o in out:
                check_diff_to_scalar(o, num_devs * 2.0)

    stypes = ['default', 'row_sparse']
    for stype in stypes:
        check_aggregator(init_kv(), 3, keys, stype)
        check_aggregator(init_kv_with_str(), 'a', str_keys, stype)