mindspore2022/docs/api/api_python/mindspore.nn.rst

mindspore.nn
=============

神经网络Cell。

用于构建神经网络中的预定义构建块或计算单元。

MindSpore中 `mindspore.nn` 接口与上一版本相比，新增、删除和支持平台的变化信息请参考 `API Updates <https://gitee.com/mindspore/docs/blob/master/resource/api_updates/ops_api_updates.md>`_。

基本构成单元
------------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.Cell
    mindspore.nn.GraphCell

容器
-----------

.. mscnplatformautosummary::
    :toctree: nn

    mindspore.nn.CellList
    mindspore.nn.SequentialCell

卷积层
--------------------

.. mscnplatformautosummary::
    :toctree: nn

    mindspore.nn.Conv1d
    mindspore.nn.Conv1dTranspose
    mindspore.nn.Conv2d
    mindspore.nn.Conv2dTranspose
    mindspore.nn.Conv3d
    mindspore.nn.Conv3dTranspose

梯度
-----

.. mscnplatformautosummary::
    :toctree: nn

    mindspore.nn.Jvp
    mindspore.nn.Vjp

循环层
--------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.GRUCell
    mindspore.nn.GRU
    mindspore.nn.LSTMCell
    mindspore.nn.LSTM
    mindspore.nn.RNNCell
    mindspore.nn.RNN

稀疏层
-------------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.Embedding
    mindspore.nn.EmbeddingLookup
    mindspore.nn.MultiFieldEmbeddingLookup
    mindspore.nn.SparseToDense
    mindspore.nn.SparseTensorDenseMatmul

非线性激活函数
--------------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.CELU
    mindspore.nn.ELU
    mindspore.nn.FastGelu
    mindspore.nn.GELU
    mindspore.nn.get_activation
    mindspore.nn.HShrink
    mindspore.nn.HSigmoid
    mindspore.nn.HSwish
    mindspore.nn.LeakyReLU
    mindspore.nn.LogSigmoid
    mindspore.nn.LogSoftmax
    mindspore.nn.PReLU
    mindspore.nn.ReLU
    mindspore.nn.ReLU6
    mindspore.nn.Sigmoid
    mindspore.nn.Softmax
    mindspore.nn.SoftShrink
    mindspore.nn.Tanh

工具
-----

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.ClipByNorm
    mindspore.nn.Dense
    mindspore.nn.Dropout
    mindspore.nn.Flatten
    mindspore.nn.L1Regularizer
    mindspore.nn.Norm
    mindspore.nn.OneHot
    mindspore.nn.Pad
    mindspore.nn.Range
    mindspore.nn.ResizeBilinear
    mindspore.nn.Roll
    mindspore.nn.Tril
    mindspore.nn.Triu
    mindspore.nn.Unfold

图像
-----

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.CentralCrop
    mindspore.nn.ImageGradients
    mindspore.nn.MSSSIM
    mindspore.nn.PSNR
    mindspore.nn.SSIM

归一化层
---------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.BatchNorm1d
    mindspore.nn.BatchNorm2d
    mindspore.nn.BatchNorm3d
    mindspore.nn.GlobalBatchNorm
    mindspore.nn.GroupNorm
    mindspore.nn.InstanceNorm2d
    mindspore.nn.LayerNorm
    mindspore.nn.MatrixDiag
    mindspore.nn.MatrixDiagPart
    mindspore.nn.MatrixSetDiag
    mindspore.nn.SyncBatchNorm

池化层
--------------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.AvgPool1d
    mindspore.nn.AvgPool2d
    mindspore.nn.MaxPool1d
    mindspore.nn.MaxPool2d


量化
-----

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.ActQuant
    mindspore.nn.Conv2dBnAct
    mindspore.nn.Conv2dBnFoldQuant
    mindspore.nn.Conv2dBnFoldQuantOneConv
    mindspore.nn.Conv2dBnWithoutFoldQuant
    mindspore.nn.Conv2dQuant
    mindspore.nn.DenseBnAct
    mindspore.nn.DenseQuant
    mindspore.nn.FakeQuantWithMinMaxObserver
    mindspore.nn.MulQuant
    mindspore.nn.TensorAddQuant

损失函数
--------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.BCELoss
    mindspore.nn.BCEWithLogitsLoss
    mindspore.nn.CosineEmbeddingLoss
    mindspore.nn.DiceLoss
    mindspore.nn.FocalLoss
    mindspore.nn.L1Loss
    mindspore.nn.LossBase
    mindspore.nn.MSELoss
    mindspore.nn.MultiClassDiceLoss
    mindspore.nn.RMSELoss
    mindspore.nn.SampledSoftmaxLoss
    mindspore.nn.SmoothL1Loss
    mindspore.nn.SoftMarginLoss
    mindspore.nn.SoftmaxCrossEntropyWithLogits

优化器
-------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.Adagrad
    mindspore.nn.Adam
    mindspore.nn.AdamOffload
    mindspore.nn.AdamWeightDecay
    mindspore.nn.ASGD
    mindspore.nn.FTRL
    mindspore.nn.Lamb
    mindspore.nn.LARS
    mindspore.nn.LazyAdam
    mindspore.nn.Momentum
    mindspore.nn.Optimizer
    mindspore.nn.ProximalAdagrad
    mindspore.nn.RMSProp
    mindspore.nn.Rprop
    mindspore.nn.SGD
    mindspore.nn.thor

Wrapper
---------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.DistributedGradReducer
    mindspore.nn.DynamicLossScaleUpdateCell
    mindspore.nn.FixedLossScaleUpdateCell
    mindspore.nn.ForwardValueAndGrad
    mindspore.nn.GetNextSingleOp
    mindspore.nn.MicroBatchInterleaved
    mindspore.nn.ParameterUpdate
    mindspore.nn.PipelineCell
    mindspore.nn.TimeDistributed
    mindspore.nn.TrainOneStepCell
    mindspore.nn.TrainOneStepWithLossScaleCell
    mindspore.nn.WithEvalCell
    mindspore.nn.WithGradCell
    mindspore.nn.WithLossCell

数学运算
----------

.. mscnplatformautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.MatMul
    mindspore.nn.Moments
    mindspore.nn.ReduceLogSumExp

评估指标
--------

.. mscnautosummary::
    :toctree: nn
    :nosignatures:
    :template: classtemplate.rst

    mindspore.nn.Accuracy
    mindspore.nn.auc
    mindspore.nn.BleuScore
    mindspore.nn.ConfusionMatrix
    mindspore.nn.ConfusionMatrixMetric
    mindspore.nn.CosineSimilarity
    mindspore.nn.Dice
    mindspore.nn.F1
    mindspore.nn.Fbeta
    mindspore.nn.HausdorffDistance
    mindspore.nn.get_metric_fn
    mindspore.nn.Loss
    mindspore.nn.MAE
    mindspore.nn.MeanSurfaceDistance
    mindspore.nn.Metric
    mindspore.nn.MSE
    mindspore.nn.names
    mindspore.nn.OcclusionSensitivity
    mindspore.nn.Perplexity
    mindspore.nn.Precision
    mindspore.nn.Recall
    mindspore.nn.ROC
    mindspore.nn.RootMeanSquareDistance
    mindspore.nn.rearrange_inputs
    mindspore.nn.Top1CategoricalAccuracy
    mindspore.nn.Top5CategoricalAccuracy
    mindspore.nn.TopKCategoricalAccuracy

动态学习率
-----------

LearningRateSchedule类
^^^^^^^^^^^^^^^^^^^^^^^

本模块中的动态学习率都是LearningRateSchedule的子类，将LearningRateSchedule的实例传递给优化器。在训练过程中，优化器以当前step为输入调用该实例，得到当前的学习率。

.. code-block::

    import mindspore.nn as nn
    
    min_lr = 0.01
    max_lr = 0.1
    decay_steps = 4
    cosine_decay_lr = nn.CosineDecayLR(min_lr, max_lr, decay_steps)
    
    net = Net()
    optim = nn.Momentum(net.trainable_params(), learning_rate=cosine_decay_lr, momentum=0.9)

.. mscnplatformautosummary::
    :toctree: nn

    mindspore.nn.CosineDecayLR
    mindspore.nn.ExponentialDecayLR
    mindspore.nn.InverseDecayLR
    mindspore.nn.NaturalExpDecayLR
    mindspore.nn.PolynomialDecayLR
    mindspore.nn.WarmUpLR

Dynamic LR函数
^^^^^^^^^^^^^^

本模块中的动态学习率都是function，调用function并将结果传递给优化器。在训练过程中，优化器将result[current step]作为当前学习率。

.. code-block::

    import mindspore.nn as nn
    
    min_lr = 0.01
    max_lr = 0.1
    total_step = 6
    step_per_epoch = 1
    decay_epoch = 4
    
    lr= nn.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch, decay_epoch)
    
    net = Net()
    optim = nn.Momentum(net.trainable_params(), learning_rate=lr, momentum=0.9)
    
.. mscnautosummary::
    :toctree: nn

    mindspore.nn.cosine_decay_lr
    mindspore.nn.exponential_decay_lr
    mindspore.nn.inverse_decay_lr
    mindspore.nn.natural_exp_decay_lr
    mindspore.nn.piecewise_constant_lr
    mindspore.nn.polynomial_decay_lr
    mindspore.nn.warmup_lr