当前位置: 首页>>代码示例>>Python>>正文


Python gluon.ParameterDict方法代码示例

本文整理汇总了Python中mxnet.gluon.ParameterDict方法的典型用法代码示例。如果您正苦于以下问题:Python gluon.ParameterDict方法的具体用法?Python gluon.ParameterDict怎么用?Python gluon.ParameterDict使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在mxnet.gluon的用法示例。


在下文中一共展示了gluon.ParameterDict方法的14个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。

示例1: model_fn

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def model_fn(model_dir):
    """Load the gluon model. Called once when hosting service starts.

    Args:
        model_dir: The directory where model files are stored.

    Returns:
        a model (in this case a Gluon network)
    """
    symbol = mx.sym.load("%s/model.json" % model_dir)
    outputs = mx.symbol.softmax(data=symbol, name="softmax_label")
    inputs = mx.sym.var("data")
    param_dict = gluon.ParameterDict("model_")
    net = gluon.SymbolBlock(outputs, inputs, param_dict)
    net.load_params("%s/model.params" % model_dir, ctx=mx.cpu())
    return net 
开发者ID:aws,项目名称:sagemaker-python-sdk,代码行数:18,代码来源:script.py

示例2: test_paramdict

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def test_paramdict():
    ctx = mx.cpu(1)
    params0 = gluon.ParameterDict('net_')
    params0.get('w0', shape=(10, 10))
    params0.get('w1', shape=(10, 10), stype='row_sparse')
    all_row_ids = mx.nd.arange(0, 10, ctx=ctx)
    # check param names
    assert list(params0.keys()) == ['net_w0', 'net_w1']
    params0.initialize(ctx=ctx)
    trainer0 = mx.gluon.Trainer(params0, 'sgd')
    prev_w0 = params0.get('w0').data(ctx)
    prev_w1 = params0.get('w1').row_sparse_data(all_row_ids)
    # save params
    params0.save('test_paramdict.params')

    # load params
    params1 = gluon.ParameterDict('net_')
    params1.get('w0', shape=(10, 10))
    params1.get('w1', shape=(10, 10), stype='row_sparse')
    params1.load('test_paramdict.params', ctx)
    trainer1 = mx.gluon.Trainer(params1, 'sgd')

    # compare the values before and after save/load
    cur_w0 = params1.get('w0').data(ctx)
    cur_w1 = params1.get('w1').row_sparse_data(all_row_ids)
    mx.test_utils.assert_almost_equal(prev_w0.asnumpy(), cur_w0.asnumpy())
    mx.test_utils.assert_almost_equal(prev_w1.asnumpy(), cur_w1.asnumpy())

    # create a new param dict with dense params, and load from the checkpoint
    # of sparse & dense params
    params2 = gluon.ParameterDict('net_')
    params2.get('w0', shape=(10, 10))
    params2.get('w1', shape=(10, 10))
    params2.load('test_paramdict.params', ctx)

    # compare the values before and after save/load
    cur_w0 = params2.get('w0').data(ctx)
    cur_w1 = params2.get('w1').data(ctx)
    mx.test_utils.assert_almost_equal(prev_w0.asnumpy(), cur_w0.asnumpy())
    mx.test_utils.assert_almost_equal(prev_w1.asnumpy(), cur_w1.asnumpy()) 
开发者ID:awslabs,项目名称:dynamic-training-with-apache-mxnet-on-aws,代码行数:42,代码来源:test_gluon.py

示例3: test_sparse_hybrid_block

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def test_sparse_hybrid_block():
    params = gluon.ParameterDict('net_')
    params.get('weight', shape=(5,5), stype='row_sparse', dtype='float32')
    params.get('bias', shape=(5), dtype='float32')
    net = gluon.nn.Dense(5, params=params)
    net.initialize()
    x = mx.nd.ones((2,5))
    # an exception is expected when forwarding a HybridBlock w/ sparse param
    y = net(x) 
开发者ID:awslabs,项目名称:dynamic-training-with-apache-mxnet-on-aws,代码行数:11,代码来源:test_gluon.py

示例4: test_trainer_reset_kv

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def test_trainer_reset_kv():
    def check_trainer_reset_kv(kv):
        params = gluon.ParameterDict()
        x = params.get('x', shape=(10,), lr_mult=1.0)
        params.initialize(ctx=[mx.cpu(0), mx.cpu(1)], init='zeros')
        trainer = gluon.Trainer(params, 'sgd', {'learning_rate': 0.1}, kvstore=kv)
        params.save('test_trainer_reset_kv.params')
        with mx.autograd.record():
            for w in x.list_data():
                y = w + 1
                y.backward()
        trainer.step(1)
        assert trainer._kvstore.type == kv
        # load would reset kvstore
        mx.nd.waitall()
        params.load('test_trainer_reset_kv.params')
        assert trainer._kvstore is None
        assert trainer._kv_initialized is False
        with mx.autograd.record():
            for w in x.list_data():
                y = w + 1
                y.backward()
        trainer.step(1)
        # the updated parameter should be based on the loaded checkpoint
        assert (x.data(mx.cpu()) == -0.2).asnumpy().all()

    kvs = ['local', 'device']
    for kv in kvs:
        check_trainer_reset_kv(kv) 
开发者ID:awslabs,项目名称:dynamic-training-with-apache-mxnet-on-aws,代码行数:31,代码来源:test_gluon_trainer.py

示例5: test_trainer_sparse_kv

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def test_trainer_sparse_kv():
    def check_trainer_sparse_kv(kv, stype, grad_stype, update_on_kv):
        params = gluon.ParameterDict()
        x = params.get('x', shape=(10,1), lr_mult=1.0, stype=stype, grad_stype=grad_stype)
        params.initialize(ctx=[mx.cpu(0), mx.cpu(1)], init='zeros')
        trainer = gluon.Trainer(params, 'sgd', {'learning_rate': 0.1}, kvstore=kv)
        all_rows = mx.nd.arange(0, 10, ctx=mx.cpu(0))
        ws = x.list_data() if stype == 'default' else x.list_row_sparse_data(all_rows)
        with mx.autograd.record():
            for w in ws:
                y = w + 1
                y.backward()
        trainer.step(1)
        assert trainer._kvstore.type == kv
        assert trainer._kv_initialized
        assert trainer._update_on_kvstore is update_on_kv
        # the updated parameter should be based on the loaded checkpoint
        mx.nd.waitall()
        updated_w = x.data(mx.cpu(0)) if stype == 'default' else x.row_sparse_data(all_rows)
        assert (updated_w == -0.2).asnumpy().all()

    kvs = ['local', 'device']
    for kv in kvs:
        check_trainer_sparse_kv(kv, 'default', 'default', True)
        check_trainer_sparse_kv(kv, 'default', 'row_sparse', False)
        check_trainer_sparse_kv(kv, 'row_sparse', 'row_sparse', True) 
开发者ID:awslabs,项目名称:dynamic-training-with-apache-mxnet-on-aws,代码行数:28,代码来源:test_gluon_trainer.py

示例6: __init__

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def __init__(self, constants=None, dtype=None, context=None):
        self.dtype = dtype if dtype is not None else get_default_dtype()
        self.mxnet_context = context if context is not None else get_default_device()
        self._constants = {}
        self._var_ties = {}
        if constants is not None:
            constant_uuids = {
                (k.uuid if isinstance(k, ModelComponent) else k): v
                for k, v in constants.items()}
            self._constants.update(constant_uuids)
        self._params = ParameterDict() 
开发者ID:amzn,项目名称:MXFusion,代码行数:13,代码来源:inference_parameters.py

示例7: initialize_params

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def initialize_params(self, graphs, observed_uuid):
        """
        :param graphs: a list of graphs in which the parameters will be optimized.
        :type graphs: a list of FactorGraph
        :param observed_uuid: Parameter Variables that are passed in directly as data, not to be inferred.
        :type observed_uuid: list, set
        """
        if self._params is not None:
            warnings.warn("InferenceParameters has already been initialized.  The existing one will be overwritten.")

        self._params = ParameterDict()
        for g in graphs:
            for var in g.get_constants():
                self._constants[var.uuid] = var.constant

            excluded = set(self._constants.keys()).union(observed_uuid)
            for var in g.get_parameters(excluded=excluded):
                var_shape = realize_shape(var.shape, self._constants)
                init = initializer.Constant(var.initial_value_before_transformation) \
                    if var.initial_value is not None else None

                self._params.get(name=var.uuid, shape=var_shape,
                                 dtype=self.dtype,
                                 allow_deferred_init=True, init=init)
            for m in g.modules.values():
                m.initialize_hidden_parameters(self._params, excluded, self._constants)

        self._params.initialize(ctx=self.mxnet_context) 
开发者ID:amzn,项目名称:MXFusion,代码行数:30,代码来源:inference_parameters.py

示例8: initialize_hidden_parameters

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def initialize_hidden_parameters(self, param_dict=None, excluded=None,
                                     constants=None):
        """
        Initialize all the hidden parameters.

        :param param_dict: the MXNet ParameterDict for parameter initialization
        :type param_dict: MXNet ParameterDict
        :param excluded: the set of variables that are excluded from initialization
        :type excluded: set(str(UUID))
        :param constants: the constants discovered during initialization, to be used for shape inference
        :type constants: {str(UUID): float or int}
        """
        if param_dict is None:
            param_dict = ParameterDict()
        if excluded is None:
            excluded = set()
        if constants is None:
            constants = {}
        for g in [self._module_graph]+self._extra_graphs:
            for var in g.get_parameters(
                    excluded=set([v.uuid for _, v in self.inputs] +
                                 [v.uuid for _, v in self.outputs]
                                 ).union(constants.keys()).union(excluded),
                    include_inherited=True):

                var_shape = realize_shape(var.shape, constants)
                init = initializer.Constant(var.initial_value_before_transformation) \
                    if var.initial_value is not None else None
                param_dict.get(name=var.uuid, shape=var_shape, dtype=self.dtype,
                               allow_deferred_init=True, init=init)
        return param_dict 
开发者ID:amzn,项目名称:MXFusion,代码行数:33,代码来源:module.py

示例9: _override_block_parameters

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def _override_block_parameters(self, input_kws):
        """
        When a probabilistic distribution is defined for the parameters of a Gluon block (in ParameterDict), a special
        treatment is necessary because otherwise these parameters will be directly exposed to a gradient optimizer as
        free parameters.

        For each parameters of the Gluon bock with probabilistic distribution, this method dynamically sets its values
        as the outcome of upstream computation and ensure the correct gradient can be estimated via automatic
        differentiation.

        :param **input_kws: the dict of inputs to the functions. The key in the dict should match with the name of
        inputs specified in the
            inputs of FunctionEvaluation.
        :type **input_kws: {variable name: MXNet NDArray or MXNet Symbol}
        """
        for bn in self.parameter_names:
            if bn in input_kws:
                val = input_kws[bn]
                param = self._gluon_block.collect_params()[bn]

                if isinstance(val, mx.ndarray.ndarray.NDArray):
                    ctx = val.context
                    ctx_list = param._ctx_map[ctx.device_typeid&1]
                    if ctx.device_id >= len(ctx_list) or ctx_list[ctx.device_id] is None:
                        raise Exception
                    dev_id = ctx_list[ctx.device_id]
                    param._data[dev_id] = val
                else:
                    param._var = val 
开发者ID:amzn,项目名称:MXFusion,代码行数:31,代码来源:mxfusion_gluon_function.py

示例10: get_weights

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def get_weights(self) -> gluon.ParameterDict:
        """
        :return: a ParameterDict containing all network weights
        """
        return self.model.collect_params() 
开发者ID:NervanaSystems,项目名称:coach,代码行数:7,代码来源:architecture.py

示例11: set_weights

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def set_weights(self, weights: gluon.ParameterDict, new_rate: float=1.0) -> None:
        """
        Sets the network weights from the given ParameterDict
        :param new_rate: ratio for adding new and old weight values: val=rate*weights + (1-rate)*old_weights
        """
        old_weights = self.model.collect_params()
        for name, p in weights.items():
            name = name[len(weights.prefix):]  # Strip prefix
            old_p = old_weights[old_weights.prefix + name]  # Add prefix
            old_p.set_data(new_rate * p._reduce() + (1 - new_rate) * old_p._reduce()) 
开发者ID:NervanaSystems,项目名称:coach,代码行数:12,代码来源:architecture.py

示例12: __init__

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def __init__(self, name: str, param_dict: gluon.ParameterDict):
        self._name = name
        self._param_dict = param_dict 
开发者ID:NervanaSystems,项目名称:coach,代码行数:5,代码来源:savers.py

示例13: test_paramdict

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def test_paramdict():
    params = gluon.ParameterDict('net_')
    params.get('weight', shape=(10, 10))
    assert list(params.keys()) == ['net_weight']
    params.initialize(ctx=mx.cpu())
    params.save('test.params')
    params.load('test.params', mx.cpu()) 
开发者ID:mahyarnajibi,项目名称:SNIPER-mxnet,代码行数:9,代码来源:test_gluon.py

示例14: load_parameters

# 需要导入模块: from mxnet import gluon [as 别名]
# 或者: from mxnet.gluon import ParameterDict [as 别名]
def load_parameters(uuid_map=None,
                        mxnet_parameters=None,
                        variable_constants=None,
                        mxnet_constants=None,
                        context=None, dtype=None,
                        current_params=None):
        """
        Loads back a set of InferenceParameters from files.
        :param mxnet_parameters: These are the parameters of
                                     the previous inference algorithm.
        These are in a {uuid: mx.nd.array} mapping.
        :type mxnet_parameters: Dict of {uuid: mx.nd.array}
        :param mxnet_constants: These are the constants in mxnet format
                                    from the previous inference algorithm.
        These are in a {uuid: mx.nd.array} mapping.
        :type mxnet_constants:  Dict of {uuid: mx.nd.array}
        :param variable_constants: These are the constants in
                                       primitive format from the previous
        inference algorithm.
        :type variable_constants: dict of {uuid: constant primitive}
        """
        def with_uuid_map(item, uuid_map):
            if uuid_map is not None:
                return uuid_map[item]
            else:
                return item
        ip = InferenceParameters(context=context, dtype=dtype)

        mapped_params = {with_uuid_map(k, uuid_map): v
                         for k, v in mxnet_parameters.items()}

        new_paramdict = ParameterDict()
        if current_params is not None:
            new_paramdict.update(current_params)

        # Do this because we need to map the uuids to the new Model
        # before loading them into the ParamDict
        for name, mapped_param in mapped_params.items():
            new_paramdict[name]._load_init(mapped_param, ip.mxnet_context)
        ip._params = new_paramdict

        new_mxnet_constants = {}
        new_variable_constants = {}
        new_variable_constants = {with_uuid_map(k, uuid_map): v
                                  for k, v in variable_constants.items()}
        new_mxnet_constants = {with_uuid_map(k, uuid_map): v
                               for k, v in mxnet_constants.items()}

        ip._constants = {}
        ip._constants.update(new_variable_constants)
        ip._constants.update(new_mxnet_constants)
        return ip 
开发者ID:amzn,项目名称:MXFusion,代码行数:54,代码来源:inference_parameters.py


注:本文中的mxnet.gluon.ParameterDict方法示例由纯净天空整理自Github/MSDocs等开源代码及文档管理平台,相关代码片段筛选自各路编程大神贡献的开源项目,源码版权归原作者所有,传播和使用请参考对应项目的License;未经允许,请勿转载。